{"category":"presentations","count":4,"items":[{"id":"we-robot-robotaxi-reveal-2024-10-10","type":"video","url":"https://www.youtube.com/watch?v=6v6dbxPlsXs","title":"'We, Robot' Robotaxi Reveal","titles":{"en":"'We, Robot' Robotaxi Reveal","de":"'We, Robot' Robotaxi Reveal","fr":"'We, Robot' Robotaxi Reveal"},"date":"2024-10-10","summary":"Tesla's 'We, Robot' event where Musk unveiled the Cybercab robotaxi, Robovan and Optimus robots.","text":"statements made in this presentation are forward-looking statements that are subject to risks and uncertainties actual results May differ materially from those projected more details can be found in the written materials ladies and gentlemen here everywhere online around the world outer space I'm France and on behalf of everybody at Tesla welcome to we robot just want to thank Warner Brothers for hosting us here as you know this is the birthplace of many epic films many of them depicting a vision of the future we're here tonight to experience that future that is closer than you think and who better than Elon right to show us that future so it looks like elon's on his way so let's welcome him all here what's up oh e welcome welcome to the Wii robot party so we we have uh we have quite a show for you tonight I think think you're going to like it uh as you can see I just arrived in the robot taxi the Cyber cab and uh there's 20 more where that came from so they've been traveling they're all there's no people in them as you can see the cars just going by with no people and we have uh we have 50 fully autonomous cars here tonight uh so you'll see model wise and the Cyber cab uh all driverless uh you'll you'll be able to take a ride in the Cyber cab there's no steering wheel or pedals so I hope this goes well we'll find out so so you see a lot of sci-fi movies where uh the future is is dark and dismal where uh it's not a future you want to be in so you know like I I love Blade Runner but uh I don't know if we want that future I think we want that uh that duster he's wearing but uh but but not but not the uh the ble apocalypse uh we want to have a fun fun exciting future that if you could look in a crystal bowl and see the future you'd be like yes I wish I could be there now that's what we want so so when we think about transport today there's a lot of kind of pain that we take for granted that we think is normal um like having to drive around La uh in like three hours of traffic um yeah and people that live in LA I mean you know try to get from Pasadena to you know Elsa gundo during rush hour is like you can fly to you know another City faster and you can get to cross town La so and you have to drive the whole way unless you're in a Tesla of course our Tesla already uh does quite well at this uh you know supervised self-driving so supervised full self full self-driving is actually working quite well I'm sure there's people in the crowd you you're using that uh yeah so it we'll move from supervised full self- driving to un unsupervised full self- driving where the car you could you could fall asleep and wake up at your destination so uh but there's also a challenge for a lot of people that cars cost too much I mean when you factor in everything that goes into a car and the car insurance and the car payments and the storage of the car it's it's very expensive so with the and you say like how often are what how many hours a week are cars used your average passenger car is only used about 10 hours a week out of 168 hours so the vast majority of the time cars are just doing nothing but if they're autonomous they could be used I don't know five times more maybe maybe 10 times more so you could actually for the the same car would have five times as much value maybe maybe 10 times as much value it's it's there's 168 hours in the week and like I said only 10 of them I use for driving so and then then a bunch of those hours are looking for a parking spot which is you know can be pretty annoying at times so so we want with autonomy you get you get your time back this is a very big deal so it's it's not just a sa like it it'll save lives like a lot of lives um and prevent injuries I I think we'll see autonomous cars become 10 times safer than a human um I me you think of times past were there were there used to be an elevator operator in every elevator uh but uh once in a while they get you know they get tired and accidentally sh somebody in a half uh you know so so now we have automated elevators you just get in an elevator and you press a button and you don't even think about it and it just takes takes you to the floor and if you did see an elevator operator with a big relay switch you'd be like that's weird um now that's that's how cars will be um and and it's not just the lives saved in injuries but if you look at the think about the cumulative time that people spend in a car and the time they will get back that they can now spend well I guess on their phones or or or watching a movie or doing work or whatever you want to do um you can think of the the car in an autonomous world as being like just a little Lounge you're just sitting in in a comfortable little lounge and you can do whatever you want while you're in this comfortable little lounge and when you get out you will be at your destination so yeah it's going to be awesome so so in fact we we I think the the cost of autonomous transport will be so low that you can think of it like individualized mass transit um the like the average cost of of a bus per mile for a city um not not the ticket price because that is subsidized but the average price is about a dollar a mile whereas the cost of uh cyber cab uh we we think probably over time from the operating cost is probably going to be around 20 cents a mile um and price including taxes and and everything else probably ends up being 30 or 4 40 cents a Miley one so yes and you will be able to buy one yes exactly U and uh we expect the cost to be below $330,000 and I think there'll be an interesting um you know business model where like let's say somebody is an uh you know UB LIF driver today uh they where they can actually sort of manage a fleet of cars and like a sort of manage I don't know 10 20 cars and just sort of you know take care of them like a like a Shepherd uh tends their flock you have a little your flock of cars and you're the shepherd and you take care of your a flock of cars I think that would be pretty cool um and um it's I think it's going to be it's going to be a glorious future it's going to be really something special so AV yes good all excellent questions um so we do expect actually to to start uh fully autonomous unsupervised FSD uh in Texas and California next year and that that's obviously that's with the model 3 in model Y and then we we we expect to be in production with the the Cyber cap which is which is really um highly optimized for autonomous transport uh in probably well I tend to be a little optimistic with time frames um but but in in in 2026 so yeah before 2027 let me put it that way um and uh we'll make this this vehicle in very high volume and um but well before that you will you will experience the uh a robotic taxi bya the model 3 and model y program and model snx to uh but to the the model the the your 3 and Y will be will achieve U unsupervised full soft driving um with with permission in where wherever Regulators essentially approve it in the US and then and then to follow in outside the US so and it's cyber TR 2 yes of course sorry I don't want to beet yes yes all our cars are basically um old cars that we make so let's let's not get new on here um all right next slide so one of the reasons why or the the the computer can be so much better than a person is that we have millions of cars that are training uh in on driving so it's like like it's like living millions of lives simultaneously and seeing very unusual situations that a person in their entire lifetime would not see but hopefully um yeah exactly so it's so with with that amount of training data it's obviously going to be much better than what a human could be um because you can't live A Million Lives um and it's also it can see an all dire simultaneously and it doesn't get tired or or text or any of those things so uh it will naturally be like I said uh 10 20 30 times safer than a human just um for all those reasons um and and I want to emphasize that the solution that we have is is AI and vision so there's no um expense of equipment needed so the the model 3 and model Y and snx that we made today will be capable of full autonomy unsupervised um and and that means that a cost of producing the vehicle is is low um now we we are going to actually over speec the computer for the Cyber cab uh so our ai5 computer um will be somewhat overspeed and uh because I think there's actually also an opportunity sort of like in Amazon web services where if the car is driving for 50 hours a week There's still over 100 hours left and it it there's a potential there to have a massive amount of distributed inference compute where if you've got like say a fleet of 100 million vehicles and a kilowatt of efficient inference compute you have 100 gaw of of compute which is really quite substantial um and uh if it's there you might as well use it so um yeah so that's that I think will make sense so all right so our autonomous future is is here um as I said we've got 50 tesas driving autonomously um we're trying to give you a sense of what what cities will be like in the future and uh when you when you get in you'll see like it's really quite a wild experience to just be in a car with no steering wheel no pedals no controls and it feels great um so and we you know we have enough Vehicles here so everyone should be able to to try it out and uh experience this the set that we built here um it's a very big set so it's like really we we've used Ser uh I don't know 20 30 Acres or something like that it's really big so it's it goes on the ride's long um and we we set it up we set it up to feel like a like a ride like a park ride so it'll be it'll be cool uh and you'll get to experience it tonight something we're also doing is uh and it's really high time we did this is uh inductive charging so the robot taxi has no plug it it just uh goes over the inductive charger and charges so yeah it's kind of how it should be thanks guys I love you too um so one of the things that like is really interesting is how will this affect this the cities that we live in and when when you drive around a city or when the car drives you around a city you'll see there's like there's a lot of parking lots there's there's parking lots everywhere parking garages uh there and and so what would happen if you have an autonomous world is that you can now turn parking lots into parks and uh so from we're taking we're taking the in lot out of parking lot um welcome um so there's a lot of opportunity to create uh green space in the cities that we live in so I think that would be quite fantastic oh and uh also so what what what what happens if you need a vehicle that uh is bigger than a model y the the Roven the Roven is uh this is a we we're going to make this and it's going to look like that now can you imagine going down the streets and you see this coming towards you that' be sick so this this can carry up to 20 people and it can also uh transport Goods so you can config good for goods transport within a city uh or transport of up to 20 people at a time so this is going to the Roven is what's going to solve for high density so if you if you want to take a sports team somewhere or um you're looking to to really get uh the cost of travel down to I don't know 51 cents a mile then you can use the Roven some people call it the Roba van but uh so yeah um you know one of the things that we want to do and we've seen this with the Cyber truck is we want to change the the look of the roads the future should look like the future so um speaking of robots so everything we've developed for our cars the batteries Power Electronics uh the advanced Motors gearboxes the the software the uh the AI inference computer it all actually applies to a humanoid robot so the same techniques it's just a robot with arms and legs instead of a robot with with wheels and uh We've made a lot of progress with the Optimus and uh as you can see we we started up with someone um in a robot suit uh sort of and then we've progressed dramatically year after year so if you extrapolate this you're really going to have something expect spacular something that anyone could own um so you can have your own personal R2-D2 C3PO and I think at scale the you know this would cost something like I don't know 20 $330,000 probably less less than a car is my prediction long term you know take us a minute to get to the long term but um but fundamentally at scale the Optimus robot you should be able to buy an Optimus robot for I think probably 20 to $330,000 long to him so and and and what can it do it can it'll basically do anything you want so it can um be a teacher babysit your kids it can walk your dog mow your lawn get the groceries just be your friend serve drinks um whatever you can think of it will do and yeah it's going to be awesome and I I I think this will be the biggest product ever of any kind yeah because I think everyone of the 8 billion people of Earth I think everyone's going to want their Optimist buddy and there's going to be some maybe two uh and then there will be they'll be producing products and services I I predict actually provided we to address risks of digital super intelligence uh 80% will 80% prob probability of good a good outcome look on the right side um the cup is 80% full um the uh the cost of products and services will decline dramatically and basically anyone will be able to have any products and services they they want it will be an age of abundance the likes of which people have not almost no one has envisioned it will be something special so now one of the things we wanted to show tonight was uh that Optimus is not a canned video it's not walled off the Optimus robots will walk among you please please be nice to the Optimus robots so you'll be able to walk right up to them and um they'll serve drinks at the bar and uh you'll directly I mean that's it's it's a wild experience just to have humanoid robots and it's they're there they just in front of you uh so yeah with that um let's party I love you guys too if you look at that gazebo over there let's get the party started what is baby don't hurtt baby don't hurt me don't hurt me no more I don't know why you're not there I give you my love but you don't care so what is right and what is wrong give me a What is love baby don't hurt me don't hurt me no what is baby don't hurt me don't hurt me no more w oh w okay e n I I I he right oh we hey he I one he know no she's not oh hello everyone Midwest is currently prioritizing VIP rides GA guests please make your way to our other amazing zones to board your Robo taxi experience thank you down he he he he he hey the rain the rain the rank the rank the rank he he he he hey he a a e my you w you w for e he he he he hey he","textByLang":{"en":"statements made in this presentation are forward-looking statements that are subject to risks and uncertainties actual results May differ materially from those projected more details can be found in the written materials ladies and gentlemen here everywhere online around the world outer space I'm France and on behalf of everybody at Tesla welcome to we robot just want to thank Warner Brothers for hosting us here as you know this is the birthplace of many epic films many of them depicting a vision of the future we're here tonight to experience that future that is closer than you think and who better than Elon right to show us that future so it looks like elon's on his way so let's welcome him all here what's up oh e welcome welcome to the Wii robot party so we we have uh we have quite a show for you tonight I think think you're going to like it uh as you can see I just arrived in the robot taxi the Cyber cab and uh there's 20 more where that came from so they've been traveling they're all there's no people in them as you can see the cars just going by with no people and we have uh we have 50 fully autonomous cars here tonight uh so you'll see model wise and the Cyber cab uh all driverless uh you'll you'll be able to take a ride in the Cyber cab there's no steering wheel or pedals so I hope this goes well we'll find out so so you see a lot of sci-fi movies where uh the future is is dark and dismal where uh it's not a future you want to be in so you know like I I love Blade Runner but uh I don't know if we want that future I think we want that uh that duster he's wearing but uh but but not but not the uh the ble apocalypse uh we want to have a fun fun exciting future that if you could look in a crystal bowl and see the future you'd be like yes I wish I could be there now that's what we want so so when we think about transport today there's a lot of kind of pain that we take for granted that we think is normal um like having to drive around La uh in like three hours of traffic um yeah and people that live in LA I mean you know try to get from Pasadena to you know Elsa gundo during rush hour is like you can fly to you know another City faster and you can get to cross town La so and you have to drive the whole way unless you're in a Tesla of course our Tesla already uh does quite well at this uh you know supervised self-driving so supervised full self full self-driving is actually working quite well I'm sure there's people in the crowd you you're using that uh yeah so it we'll move from supervised full self- driving to un unsupervised full self- driving where the car you could you could fall asleep and wake up at your destination so uh but there's also a challenge for a lot of people that cars cost too much I mean when you factor in everything that goes into a car and the car insurance and the car payments and the storage of the car it's it's very expensive so with the and you say like how often are what how many hours a week are cars used your average passenger car is only used about 10 hours a week out of 168 hours so the vast majority of the time cars are just doing nothing but if they're autonomous they could be used I don't know five times more maybe maybe 10 times more so you could actually for the the same car would have five times as much value maybe maybe 10 times as much value it's it's there's 168 hours in the week and like I said only 10 of them I use for driving so and then then a bunch of those hours are looking for a parking spot which is you know can be pretty annoying at times so so we want with autonomy you get you get your time back this is a very big deal so it's it's not just a sa like it it'll save lives like a lot of lives um and prevent injuries I I think we'll see autonomous cars become 10 times safer than a human um I me you think of times past were there were there used to be an elevator operator in every elevator uh but uh once in a while they get you know they get tired and accidentally sh somebody in a half uh you know so so now we have automated elevators you just get in an elevator and you press a button and you don't even think about it and it just takes takes you to the floor and if you did see an elevator operator with a big relay switch you'd be like that's weird um now that's that's how cars will be um and and it's not just the lives saved in injuries but if you look at the think about the cumulative time that people spend in a car and the time they will get back that they can now spend well I guess on their phones or or or watching a movie or doing work or whatever you want to do um you can think of the the car in an autonomous world as being like just a little Lounge you're just sitting in in a comfortable little lounge and you can do whatever you want while you're in this comfortable little lounge and when you get out you will be at your destination so yeah it's going to be awesome so so in fact we we I think the the cost of autonomous transport will be so low that you can think of it like individualized mass transit um the like the average cost of of a bus per mile for a city um not not the ticket price because that is subsidized but the average price is about a dollar a mile whereas the cost of uh cyber cab uh we we think probably over time from the operating cost is probably going to be around 20 cents a mile um and price including taxes and and everything else probably ends up being 30 or 4 40 cents a Miley one so yes and you will be able to buy one yes exactly U and uh we expect the cost to be below $330,000 and I think there'll be an interesting um you know business model where like let's say somebody is an uh you know UB LIF driver today uh they where they can actually sort of manage a fleet of cars and like a sort of manage I don't know 10 20 cars and just sort of you know take care of them like a like a Shepherd uh tends their flock you have a little your flock of cars and you're the shepherd and you take care of your a flock of cars I think that would be pretty cool um and um it's I think it's going to be it's going to be a glorious future it's going to be really something special so AV yes good all excellent questions um so we do expect actually to to start uh fully autonomous unsupervised FSD uh in Texas and California next year and that that's obviously that's with the model 3 in model Y and then we we we expect to be in production with the the Cyber cap which is which is really um highly optimized for autonomous transport uh in probably well I tend to be a little optimistic with time frames um but but in in in 2026 so yeah before 2027 let me put it that way um and uh we'll make this this vehicle in very high volume and um but well before that you will you will experience the uh a robotic taxi bya the model 3 and model y program and model snx to uh but to the the model the the your 3 and Y will be will achieve U unsupervised full soft driving um with with permission in where wherever Regulators essentially approve it in the US and then and then to follow in outside the US so and it's cyber TR 2 yes of course sorry I don't want to beet yes yes all our cars are basically um old cars that we make so let's let's not get new on here um all right next slide so one of the reasons why or the the the computer can be so much better than a person is that we have millions of cars that are training uh in on driving so it's like like it's like living millions of lives simultaneously and seeing very unusual situations that a person in their entire lifetime would not see but hopefully um yeah exactly so it's so with with that amount of training data it's obviously going to be much better than what a human could be um because you can't live A Million Lives um and it's also it can see an all dire simultaneously and it doesn't get tired or or text or any of those things so uh it will naturally be like I said uh 10 20 30 times safer than a human just um for all those reasons um and and I want to emphasize that the solution that we have is is AI and vision so there's no um expense of equipment needed so the the model 3 and model Y and snx that we made today will be capable of full autonomy unsupervised um and and that means that a cost of producing the vehicle is is low um now we we are going to actually over speec the computer for the Cyber cab uh so our ai5 computer um will be somewhat overspeed and uh because I think there's actually also an opportunity sort of like in Amazon web services where if the car is driving for 50 hours a week There's still over 100 hours left and it it there's a potential there to have a massive amount of distributed inference compute where if you've got like say a fleet of 100 million vehicles and a kilowatt of efficient inference compute you have 100 gaw of of compute which is really quite substantial um and uh if it's there you might as well use it so um yeah so that's that I think will make sense so all right so our autonomous future is is here um as I said we've got 50 tesas driving autonomously um we're trying to give you a sense of what what cities will be like in the future and uh when you when you get in you'll see like it's really quite a wild experience to just be in a car with no steering wheel no pedals no controls and it feels great um so and we you know we have enough Vehicles here so everyone should be able to to try it out and uh experience this the set that we built here um it's a very big set so it's like really we we've used Ser uh I don't know 20 30 Acres or something like that it's really big so it's it goes on the ride's long um and we we set it up we set it up to feel like a like a ride like a park ride so it'll be it'll be cool uh and you'll get to experience it tonight something we're also doing is uh and it's really high time we did this is uh inductive charging so the robot taxi has no plug it it just uh goes over the inductive charger and charges so yeah it's kind of how it should be thanks guys I love you too um so one of the things that like is really interesting is how will this affect this the cities that we live in and when when you drive around a city or when the car drives you around a city you'll see there's like there's a lot of parking lots there's there's parking lots everywhere parking garages uh there and and so what would happen if you have an autonomous world is that you can now turn parking lots into parks and uh so from we're taking we're taking the in lot out of parking lot um welcome um so there's a lot of opportunity to create uh green space in the cities that we live in so I think that would be quite fantastic oh and uh also so what what what what happens if you need a vehicle that uh is bigger than a model y the the Roven the Roven is uh this is a we we're going to make this and it's going to look like that now can you imagine going down the streets and you see this coming towards you that' be sick so this this can carry up to 20 people and it can also uh transport Goods so you can config good for goods transport within a city uh or transport of up to 20 people at a time so this is going to the Roven is what's going to solve for high density so if you if you want to take a sports team somewhere or um you're looking to to really get uh the cost of travel down to I don't know 51 cents a mile then you can use the Roven some people call it the Roba van but uh so yeah um you know one of the things that we want to do and we've seen this with the Cyber truck is we want to change the the look of the roads the future should look like the future so um speaking of robots so everything we've developed for our cars the batteries Power Electronics uh the advanced Motors gearboxes the the software the uh the AI inference computer it all actually applies to a humanoid robot so the same techniques it's just a robot with arms and legs instead of a robot with with wheels and uh We've made a lot of progress with the Optimus and uh as you can see we we started up with someone um in a robot suit uh sort of and then we've progressed dramatically year after year so if you extrapolate this you're really going to have something expect spacular something that anyone could own um so you can have your own personal R2-D2 C3PO and I think at scale the you know this would cost something like I don't know 20 $330,000 probably less less than a car is my prediction long term you know take us a minute to get to the long term but um but fundamentally at scale the Optimus robot you should be able to buy an Optimus robot for I think probably 20 to $330,000 long to him so and and and what can it do it can it'll basically do anything you want so it can um be a teacher babysit your kids it can walk your dog mow your lawn get the groceries just be your friend serve drinks um whatever you can think of it will do and yeah it's going to be awesome and I I I think this will be the biggest product ever of any kind yeah because I think everyone of the 8 billion people of Earth I think everyone's going to want their Optimist buddy and there's going to be some maybe two uh and then there will be they'll be producing products and services I I predict actually provided we to address risks of digital super intelligence uh 80% will 80% prob probability of good a good outcome look on the right side um the cup is 80% full um the uh the cost of products and services will decline dramatically and basically anyone will be able to have any products and services they they want it will be an age of abundance the likes of which people have not almost no one has envisioned it will be something special so now one of the things we wanted to show tonight was uh that Optimus is not a canned video it's not walled off the Optimus robots will walk among you please please be nice to the Optimus robots so you'll be able to walk right up to them and um they'll serve drinks at the bar and uh you'll directly I mean that's it's it's a wild experience just to have humanoid robots and it's they're there they just in front of you uh so yeah with that um let's party I love you guys too if you look at that gazebo over there let's get the party started what is baby don't hurtt baby don't hurt me don't hurt me no more I don't know why you're not there I give you my love but you don't care so what is right and what is wrong give me a What is love baby don't hurt me don't hurt me no what is baby don't hurt me don't hurt me no more w oh w okay e n I I I he right oh we hey he I one he know no she's not oh hello everyone Midwest is currently prioritizing VIP rides GA guests please make your way to our other amazing zones to board your Robo taxi experience thank you down he he he he he hey the rain the rain the rank the rank the rank he he he he hey he a a e my you w you w for e he he he he hey he"},"languages":["en"],"lang":"en","transcriptSource":"YouTube auto-generated captions (Tesla official upload)"},{"id":"tesla-ai-day-2022","type":"video","url":"https://www.youtube.com/watch?v=ODSJsviD_SU","title":"Tesla AI Day","titles":{"en":"Tesla AI Day","de":"Tesla AI Day","fr":"Tesla AI Day"},"date":"2022-09-30","summary":"Tesla AI Day 2022: the first walking prototype of the Optimus humanoid robot, plus updates on Full Self-Driving and the Dojo supercomputer.","text":"All right, welcome everybody give everyone a moment to Get back in the audience and All right great welcome to Tesla AI day 2022 We've got some really exciting things to show you I think you'll be pretty impressed I do want to set some expectations with respect to our Optimus robot as As you know last year was just a person in a robot suit But we've now we've come a long way and that's I think we you know compared to that it's gonna be very impressive and We're gonna talk about The advancements in AI for full self-driving as well as how they apply to more generally to real-world AI problems Like a humanoid robot and and even going beyond that I think there's some potential that what we're doing here at Tesla could make a meaningful contribution to AGI and And I think actually Tesla is a good Antity to do it from a governance standpoint because we're a publicly traded company with one class of stock and That means that the public controls Tesla and I think that's actually a good thing So if I go crazy you can fire me. This is important Maybe I'm not crazy. I don't know so Yeah, so we're gonna talk a lot about our progress in AI autopilot as well as progress in with dojo and Then we're gonna bring the team out and to do a long Q&A so you can ask tough questions But whatever you'd like existential questions technical questions, but we want to have As much time for Q&A as possible. So let's see you with that That's because Hey guys, I'm Milana work on autopilot and it is about and I'm Lizzie Mechanical engineer on the project as well. Okay So should we should we bring up the bot before we do that? We have one One little bonus tip for the day.\n\nThis is actually the first time we try this robot without any backup support Cranes mechanical mechanisms. No cables. Nothing. Yeah I want to do it with you guys tonight. That is the first time. Let's see.\n\nYou ready? Let's go I I I think the bug got some boobs This is essentially the simple self-driving computer that runs in your Tesla cars by the way This is the this is literally the first time the robot has operated without a tether was on stage tonight So the robot can actually do a lot more than we just showed you we just didn't want it to fall on its face So we'll we'll show you some videos now of the robot doing a bunch of other things Yeah, which are less risky. Yeah, we should close the screen guys. Yeah Yeah, we wanted to show a little bit more what we've done over the past few months with the bot and just walking around and dancing on stage Just humble beginnings, but you can see the autopilot neural networks running as it's just retrained for the bot directly on that on that new platform That's my watering can yeah when you when you see a rendered view. That's that's the robot. What's the that's the world the robot sees So it's it's very clearly identifying objects like this is the object.\n\nIt should pick up picking it up Yeah We use the same process as we did for autopilot to connect data and train neural networks that we didn't deploy on the robot That's an example that illustrates the upper body a little bit more Something that will like try to nail down in a few months over the next few months, I would say To perfection, but this is really an actual station in the Fremont factory as well that it's working at And that's not the only thing we have to show today, right? Yeah, absolutely. So That what you saw was what we call bumble sea. That's our sort of rough development robot using semi off-the-shelf actuators But we actually have gone a step further than that already the team's done an incredible job And we actually have an optimist bot with fully tesla designed and built actuators Um battery pack control system everything. Um, it it wasn't quite ready to walk But I think it will walk in a few weeks But we wanted to show you the robot The something that's actually fairly close to what we'll go into production And and show you all all the things that can do so let's bring it up All right Yeah So here you're seeing optimists with these the With the degrees of freedom that we expect to have in optimists production unit one Which is the ability to move all the fingers independently move the To have the thumb have two degrees of freedom. So it has opposable thumbs And both left and right hand so it's able to operate tools and do useful things.\n\nOur goal is to make a useful humanoid robot as quickly as possible and We've also designed it using the same discipline that we use in designing the car, which is to say to design it for All manufacturing such that it's possible to make the robot at in high volume at low cost with high reliability So that that's incredibly important. I mean, you've all seen very impressive humanoid robot demonstrations And that that's great. But what are they missing? They're missing a brain that they don't have the the intelligence to navigate the world by themselves And they're they're also very expensive Um and made in low volume. Um, whereas, uh, this this is the optimist's design to be an extremely capable robot But made in in very high volume probably ultimately millions of units And it is expected to cost much less than a car So, uh, I would say probably less than 20,000 dollars would be my guess Okay The the potential for optimist is I think appreciated by very few people As usual Tesla demos are coming in hot So Um, yeah, uh, I'm the team's put on put in and the team has put in an incredible amount of work Uh working days, you know seven days a week Running the 3am oil To to get to the demonstration today. Um, super proud of what they've done is they've really done a great job I'd just like to give a hand to the whole optimist team So, you know that now there's still a lot of work to be done to, uh refine optimists and Improve it.\n\nObviously, this is just optimist version one And that's really why we're holding this event Which is to convince some of the most talented people in the world like you guys um to Join tesla and help make it a reality and bring it to fruition at scale Such that it can help millions of people And the and the potential likes it is is really boggles the mind because you have to say like what what is an economy an economy is uh sort of productive entities times the productivity, uh capita times Productivity per capita at the point at which there is not a limitation on capita The it's not clear what an economy even means at that point. It an economy becomes quasi infinite um so What what you know taken to fruition in the hopefully benign scenario the This means a future of abundance a future where um There is no poverty where people you can have whatever you want In terms of products and services It really is a a fundamental transformation of civilization as we know it Obviously we want to make sure that transformation is a positive one and um safe And but but that's also why I think Tesla as an entity doing this being a single class of stock publicly traded owned by the public um is very important Um and should not be overlooked. I think this is essential because then if the public doesn't like what tesla's doing The public can buy shares in tesla and vote differently This is a big deal Um Like it's very important that that I can't just do what I want You know, sometimes people think that that but it's not true. Um, so You know that it's it's very important that the the corporate entity that has that that makes this happen Is something that the public can properly influence And so I think the tesla structure is is is ideal for that Um And like said that you know self-driving cars will certainly have a Tremendous impact on the world. Um, I think they will improve the productivity of transport by at least A half order of magnitude perhaps an order of magnitude perhaps more um Optimus I think has Maybe a two order of magnitude Uh potential improvement in uh economic output Like like it's it's not clear. It's not clear what the limit actually even is um so But we we need to do this in the right way we need to do it carefully and safely And ensure that the outcome is one that is beneficial to Uh civilization and and one that humanity wants Uh, I can't this is also extremely important obviously so, um And and I hope you will consider uh joining tesla to achieve those goals Um It tesla we're we really care about doing the right thing here or aspire to do the right thing and and really not Pay the road to hell with with good intentions And I think the road is road to hell is mostly paved with bad intentions, but every now and again There's a good intention in there.\n\nSo we want to do the right thing. Um, so, you know consider joining us and helping make it happen um With that let's let's uh, we want to the next phase All right, so you've seen a couple robots today. Let's do a quick timeline recap So last year we unveiled the tesla bot concept, but a concept doesn't get us very far We knew we needed a real development and integration platform to get real life learnings as quickly as possible So that robot that came out and did the little routine for you guys We had that within six months built working on software integration hardware upgrades over the months since then But in parallel we've also been designing the next generation this one over here So this guy is rooted in the the foundation of sort of the vehicle design process, you know We're leveraging all of those learnings that we already have Obviously there's a lot that's changed since last year, but there's a few things that are still the same you'll notice We still have this really detailed focus on the true human form We think that matters for a few reasons, but it's fun. We spend a lot of time thinking about how amazing the human body is We have this incredible range of motion typically really amazing strength Um a fun exercise is if you put your fingertip on the chair in front of you you'll notice that there's a huge Range of motion that you have in your shoulder and your elbow for example without moving your fingertip you can move those joints all over the place Um, but the robot, you know, its main function is to do real useful work And it maybe doesn't necessarily need all of those degrees of freedom right away So we've stripped it down to a minimum sort of 28 fundamental degrees of freedom and then of course our hands in addition to that Humans are also pretty efficient at some things and not so efficient in other times So for example, we can eat a small amount of food to sustain ourselves for several hours. That's great Uh, but when we're just kind of sitting around no offense, but we're kind of inefficient. We're just sort of burning energy So on the robot platform, what we're going to do is we're going to minimize that idle power consumption drop it as low as possible And that way we can just flip a switch and immediately the robot turns into something that does useful work So let's talk about this latest generation in some detail, shall we?\n\nSo on the screen here, you'll see in orange are actuators, which we'll get to in a little bit and in blue are electrical system So now that we have our sort of human based research and we have our first development platform We have both research and execution to draw from for this design Again, we're using that vehicle design foundation. So we're taking it from concept through design and analysis and then build and validation Along the way, we're going to optimize for things like cost and efficiency because those are critical metrics to take this product to scale eventually How are we going to do that? Well, we're going to reduce our part count and our power consumption of every element possible We're going to do things like reduce the sensing in the wiring at our extremities You can imagine a lot of mass in your hands and feet is going to be quite difficult and power consumptive to move around And we're going to centralize both our power distribution and our compute to the physical center of the platform So in the middle of our torso, actually it is the torso. We have our battery pack This is sized at 2.3 kilowatt hours, which is perfect for about a full day's worth of work What's really unique about this battery pack is it has all of the battery electronics integrated into a single pcb within the pack So that means everything from sensing to fusing Charge management and power distribution is all on one all in one place We're also leveraging both our vehicle products and our energy products To roll all of those key features into this battery. So that's streamlined manufacturing Really efficient and simple cooling methods battery management and also safety And of course we can leverage tesla's existing infrastructure and supply chain to make it So going on to sort of our brain it's not in the head, but it's pretty close Also in our torso, we have our central computer So as you know tesla already ships full self-driving computers in every vehicle we produce We want to leverage both the autopilot hardware and the software for the humanoid platform But because it's different in requirements and informed factor, we're going to change a few things first So we still are gonna it's going to do everything that a human brain does Processing vision data making split sescan decisions based on multiple sensory inputs and also communications So to support communications, it's equipped with wireless connectivity as well as audio support And then it also has hardware level security features which are important to protect both the robot and the people around the robot So now that we have our sort of core we're going to need some limbs on the sky Um, and we'd love to show you a little bit about our actuators and our fully functional hands as well But the first before we do that, I'd like to introduce Malcolm who's going to speak a little bit about our structural foundation for the robot Tesla have the capabilities to analyze highly complex systems Don't get much more complex than a crash You can see here a simulated crash from bottle three Superimposed on top of the actual physical crash It's actually incredible how um, how accurate it is Just to give you an idea of the complexity of this model It includes every not bolt and washer every spot weld and it has 35 million degrees of freedom quite amazing And it's true to say that if we didn't have models like this, we wouldn't be able to make the safest cars in the world So can we utilize our capabilities and our methods from the automotive side to influence a robot? Well, we can make a model and since we had crash software we're using the same software here We can make it fall down The purpose of this is to make sure that if it falls down ideally it doesn't but it's superficial damage We don't want it to for example break its gearbox and its arms.\n\nThat's equivalent of a dislocated shoulder of a robot Difficult and expensive to fix So we wanted to dust itself off get on with the job. It's being given We could also take the same model and we can drive the actuators using the inputs from a previously solved model Bringing it to life So this is producing the motions for the tasks we want the robot to do these tasks are picking up boxes turning squatting walking upstairs Whatever the set of tasks are we can play to the model. This is showing just simple walking We can create the stresses in all the components that helps us optimize the components These are not dancing robots these are actually the modal behavior the first five modes of the robot And typically when people make robots they make sure the first mode is up around the top single figures up towards 10 hertz Who is to do this is to make the controls of walking easier. It's very difficult to walk if you can't guarantee where your foot is wobbling around That's okay if you make one robot. We want to make thousands maybe millions We haven't got the luxury of making from carbon fiber titanium. We want to make them plastic things are not quite as stiff So we can't have these high targets.\n\nI call them dumb targets We've got to make them work at lower targets So is that it's that good to work? Well, if you think about it, sorry about this, but we're just bags of soggy jelly and bones thrown in We're not high frequency. If I start on my leg, I don't vibrate at 10 hertz We people operate at a lot of frequency. So we know the robot actually can it just makes controls harder So we take the information from this the modal Data and the stiffness and feed it into the control system that allows it to walk And Just changing tax lightly looking at the knee We could take some inspiration from biology and we can look to see what the mechanical advantage of the knee is It turns out it actually represent quite similar to four-bar link and that's quite non-linear That's not surprising really because if you think when you bend your leg down The torque on your knee is much more when it's bent than it is when it's straight So you'd expect a non-linear function and in fact the biology is non-linear. This matches it quite accurately So that's a representation the four-bar link is obviously not physically four-bar link as I said the characteristics are similar, but Me bending down that's not very scientific. Let's be a bit more scientific We've played all the tasks through the through this graph And this is showing picketing is walking squatting the tasks I said we did on the stress And that's the the torque Seen at the knee against the knee bend on the horizontal axis This is showing the requirement for the need to do all these tasks And then put a curve through it surfing over the top of the piece and that's saying this is what's required to make the robot Do these tasks?\n\nSo if we look at the four-bar link that's actually the green curve And it's saying that the non-linearity of the four-bar link is actually linearized The characteristic of the force what that really says is that's lower the force That's what makes the actuator have the lowest possible force, which is the most efficient. We want to burn energy up slowly What's the blue curve with the blue curve is actually if we didn't have a four-bar link We just had an arm sticking out of my leg here with a with an actuator on it a simple two-bar link That's the best we could do with a simple two-bar link and it shows that that would create a much more force in the actuator Which would not be efficient So what does it look like in practice? well As you'll see but it's very tightly packaged in the knee you'll see it go transparent on the second You'll see the four-bar link there is operating on the actuator. This is determined the force and the displacements on the actuator And now pass you over to Constantina to tell you a lot more detail about how these actuators are made and designed optimized. Thank you So I am I would like to talk to you about The design process and the actuator portfolio In our robot So there are many similarities between a car and the robot when it comes to powertrain design The most important thing that matters here is energy mass and cost We are carrying over most of our designing experience from the car to the robot So in the particular case you see a car with two drive units And the drive units are used in order to accelerate the car zero to 60 miles per hour time or drive a city drive site while The robot that has 28 actuators and It's not obvious. What are the tasks at actuator level?\n\nSo we have tasks that are higher level like walking or climbing stairs or carrying a heavy object Which need to be translated into joint Into joint specs therefore we use our model That generates The torque speed Trajectories for our joints which subsequently is going to be fed in our optimization model And to run through the optimization process This is one of the scenarios that the robot is capable of doing which is turning and walking So when we have this torque speed trajectory we lay it over an efficiency map of an actuator And we are able along the trajectory to generate The power consumption and the energy cumulative energy for the task versus time So this allows us to define the system cost for the particular actuator and put a simple point into the cloud Then we do this for hundreds of thousands of actuators by solving in our cluster And the red line denotes the Pareto front, which is the preferred area where we will look for optimal So the x denotes the preferred actuator design we have picked for this particular joint So now we need to do this for every joint. We have 28 joints to optimize and we parse our cloud We parse our cloud again for every joint spec and the red axis this time denotes the bespoke actuator designs for every joint The problem here Is that we have too many unique actuator designs and even if we take advantage of the symmetry Still there are too many in order to make something Mass manufacturable we need to be able to reduce the amount of unique actuator designs Therefore we run something called commonality study, which we parse our cloud again Looking this time for actuators that simultaneously meet the joint performance requirements for more than one joint at the same time So the resulting portfolio is six actuators and they show in a color map at the middle figure um And the actuators can be also viewed in this Slide we have three rotary and three linear actuators all of which have a great Output force or torque per mass The rotary actuator in particular has a mechanical clutch integrated On the high speed side angular contact ball bearing and on the high speed side And on the low speed side a cross roller bearing and the year train is a strain wave year Um, there are three integrated sensors here and the bespoke permanent magnet machine The linear actuator I'm sorry The linear actuator has planetary rollers and an inverted planetary Screw as a gear train which allows efficiency and compaction and durability So in order to demonstrate the force capability of our linear actuators, we have set up an experiment in order to test it under its limits And I will let you enjoy the video So our actuator is able to lift A half ton nine foot concert grand piano And This is a requirement it's not something nice to have Because our muscles can do the same when they are direct driven when they are directly driven our quadriceps muscles Can do the same thing it's just that the knee is an upgearing Linked system that converts the force into velocity at the end effector of our heels for purposes of giving To the human body agility So this is one of the main things that are amazing about the human body And I'm concluding my part at this point and I would like to welcome my colleague Mike who's going to talk to you about Hand design. Thank you very much Thanks for seeing us So we just saw how powerful a human and a humanoid actuator can be However, humans are also incredibly dexterous The human hand has the ability to move at 300 degrees per second There's tens of thousands of tactile sensors And it has the ability to grasp and manipulate almost every object in our daily lives For our robotic hand design, we were inspired by biology We have five fingers an opposable thumb Our fingers are driven by metallic tendons that are both flexible and strong We have the ability to complete wide aperture power grasps while also being optimized for precision gripping of small thin and delicate objects So why a human like robotic hand? Well, the main reason is that our factories in the world around us is designed to be ergonomic So what that means is that it ensures that objects in our factory are graspable But it also ensures that new objects that we may have never seen before can be grasped by the human hand And by our robotic hand as well The converse there is is pretty interesting because it's saying that these objects are designed to our hand instead of having to make changes To our hand to accompany a new object Some basic stats about our hand is that it has six actuators and 11 degrees of freedom It has an in-hand controller which drives the fingers and receives sensor feedback Sensor feedback is really important to learn a little bit more about the objects that we're grasping And also for proprioception and that's the ability for us to recognize where our hand is in space One of the important aspects of our hand is that it's adaptive This adaptability is involved essentially as complex mechanisms that allow the hand to adapt the objects that's being grasped Another important part is that we have a non-back drivable finger drive This clutching mechanism allows us to hold and transport objects without having to turn on the hand motors You just heard how we went about going we went about designing the tesla bot hardware Now I'll hand it off to Milan and our autonomy team to bring this robot to life Thanks Michael All right So all those cool things we've shown earlier in the video Were possible just in a matter of a few months. Thanks to the amazing work that we've done autopilot over the past few years Most of those components poured it quite easily over to the bot's environment If you think about it, we're just moving from a robot on wheels to a robot on legs So some of the components are pretty similar and some other require more heavy lifting So for example our computer vision neural networks Were ported directly from autopilot to the bot's situation It's exactly the same occupancy network that we'll talk into a little bit more details later with the autopilot team that is now running on the bot here in this video The only thing that changed really is the training data that we had to recollect We're also trying to find ways to improve those occupancy networks Using work made on your radiance fields to get really great volumetric Rendering of the bot's environments for example here some machinery that the bot might have to interact with Another interesting problem to think about is in indoor environments, mostly with that sense of gps signal How do you get the bot to navigate to its destination? Say for instance to find its nearest charging station So we've been training More neural networks to identify high-frequency features key points within the bot's camera streams And track them across frames over time as the bot navigates with its environment And we're using those points to get a better estimate of the bot's pose and trajectory within its environment as it's walking We also did quite some work on the simulation side and this is literally the autopilot simulator To which we've integrated the robots locomotion code and this is a video of the Motion control code running in the autopilot simulator Showing the evolution of the robot's work over time.\n\nSo as you can see we started quite slowly in April and start accelerating as we unlock more joints And deeper more advanced techniques like arms balancing over the past few months And so locomotion is specifically one component that's very different as we're moving from the car to the bot's environment And so I think it warrants a little bit more depth and I'd like my colleagues to start talking about this now Thank you Milan. Hi, everyone. I'm Felix. I'm a robotics engineer on the project and I'm going to talk about walking Walking seems easy, right? People do it every day. You don't even have to think about it But there are some aspects of walking which are challenging from engineering to technology And I think that's one of the things that makes it so much easier for me to think about it But there are some aspects of walking which are challenging from engineering perspective.\n\nFor example Physical self-awareness that means having a good representation of yourself What is the length of your limbs? What is the mass of your limbs? What is the size of your feet? All that matters Also having an energy efficient gate. You can imagine there's different styles of walking and all of them are equally efficient Most important keep balance. Don't fall And of course also coordinate the motion of all of your limbs together So now humans do all of this naturally But as engineers or roboticists we have to think about these problems And the following I'm going to show you how we address them in our locomotion planning and control stack So we start with locomotion planning And our representation of the bot that means a model of the robots kinematics dynamics and the contact properties And using that model and the desired path for the bots our locomotion planner generates reference trajectories for the entire system This means feasible trajectories with respect to the assumptions of our model The planner currently works in three stages.\n\nIt starts planning footsteps and ends with the entire motion photo system And let's dive a little bit deeper in how this works So in this video we see footsteps being planned over a planning horizon following the desired path And we start from this and add then Foot trajectories that connect these footsteps using toe-off and heel strike just as the humans Just as humans do and this gives us the largest right and less knee bend for high efficiency of the system The last stage is then finding a center of mass trajectory Which gives us a dynamically feasible motion of the entire system to keep balance As we all know plans are good, but we also have to realize them in reality. Let's say how see how we can do this Thank you Felix. Hello everyone. My name is Anand and I'm going to talk to you about controls So let's take the motion plan that Felix just talked about and put it in the real world on a real robot Let's see what happens It takes a couple steps and falls down Well, that's a little disappointing But we are missing a few key pieces here, which will make it walk Now as Felix mentioned the motion planner is using an idealized version of itself and a version of reality around it This is not exactly correct It also expresses its intention Through trajectories and wrenches wrenches of forces and torques that it wants to exert on the world to locomotive Reality is way more complex than any similar model. Also the robot is not simplified It's got vibrations and modes, compliance, sensor noise and on and on and on So what does that do to the real world when you put the bot in the real world? Well, the unexpected forces cause unmodeled dynamics, which essentially the planet doesn't know about and that causes destabilization Especially for a system that is dynamically stable like bipedal locomotion So what can we do about it?\n\nWell, we measure reality We use sensors and our understanding of the world to do state estimation And here you can see the attitude and pelvis pose, which is essentially the vestibular system in a human Along with the center of mass trajectory being tracked when the robot is walking in the office environment Now we have all the pieces we need in order to close the loop So we use our better bot model We use the understanding of reality that we've gained through state estimation And we compare what we want versus what we expect the reality expect that reality is doing to us in order to Add corrections to the behavior of the robot Here the robot certainly doesn't appreciate being poked, but it has an admirable job of staying upright The final point here is a robot that walks is not enough We need it to use its hands and arms to be useful. Let's talk about manipulation Hi everyone, my name is Eric robotics engineer on tesla bot And I want to talk about how we've made the robot manipulate things in the real world We wanted to manipulate objects while looking as natural as possible and also get there quickly So what we've done is we've broken this process down into two steps First is generating a library of natural motion references Or we could call them demonstrations and then we've adapted these motion references online to the current real world situation So let's say we have a human demonstration of picking up an object We can get a motion capture of that demonstration, which is visualized right here as A bunch of keyframes representing the location of the hands the elbows the torso We can map that to the robot using inverse kinematics And if we collect a lot of these now we have a library that we can work with But a single demonstration is not generalizable to the variation in the real world For instance, this would only work for a box in a very particular Location So what we've also done is run these Reference trajectories through a trajectory optimization program which solves for where the hand should be how the robot should balance during When it needs to adapt the motion to the real world. So for instance, if the box is In this location, then our optimizer will create this trajectory instead Next Milan's going to talk about uh, what's next for the optimist uh, tesla lie. Thanks Right, so hopefully by now you guys got a good idea of what we've been up to over the past few months Um, we started having something that's usable, but it's far from being useful. There's still a long and exciting road ahead of us um, I think the first thing within the next few weeks is to Get optimists at least apart with bumble see the other bug prototype you saw earlier and probably beyond We are also going to start focusing on the real use case at one of our factories and really going to try to try to Nail this down and I run out all the elements needed to deploy this product in the real world I was mentioning earlier, you know indoor navigation Um graceful for management or even servicing all components needed to scale this product up But um, I don't know about you, but after seeing what we've shown tonight I'm pretty sure we can get this done within the next few months or years Um, and and make this product a reality and change the entire economy Um, so I would like to thank the entire optimist team for their hard work over the past few months I think it's pretty amazing. All of this was done in barely six or eight months.\n\nThank you very much Hey everyone Hi, I'm Ashok. I lead the autopilot team alongside Milan Oh god, it's going to be so hard to top that optinist section He'll try nonetheless anyway Every tesla that has been built over the last several years We think has the hardware to make the car drive itself We have been working on the software to add higher and higher levels of autonomy This time around last year. We are roughly 2000 cars driving our fsd beta software Since then we have significantly improved the software's robustness and capability That we have now shipped it to 160,000 customers as of today This did not come for free it came from the sweat and blood of the engineering team over the last one year Um, for example, we trained 75,000 neural network models just last one year That's roughly a model every eight minutes That's you know coming out of the team and then we evaluate them on our large clusters and then we ship 281 of those models That actually improved the performance of the car And this space of innovation is happening throughout the stack The the planning software the infrastructure the tools even hiring everything is progressing to the next level The fsd beta software is quite capable of driving the car It should be able to navigate from parking lot to parking lot handling city street driving stopping for traffic lights and stop signs Negotiating with objects at intersections making turns and so on All of this comes from the Uh camera streams that go through our neural networks that run on the car itself It's not coming back to the server or anything It runs on the car and produces all the outputs uh to form the world model or on the car and the planning software drives the car based on that Today we'll go into a lot of the components that make up the system The occupancy network acts as the base geometry layer of the system This is a multi-camera video neural network That from the images predicts the full physical occupancy of the world around the robot So anything that's physically present trees walls buildings Cars balls, whatever you it predicts if it's physically present it predicts them along with their future motion On top of this base level of geometry We have more semantic layers in order to navigate the roadways. We need the lanes, of course But then the roadways have lots of different lanes and they connect in all kinds of ways So it's actually a really difficult problem for typical computer vision techniques to predict the set of lanes and their Connectivities So we reached all the way into language technologies and then pull the state of the art from other Domains are not just computer vision to make this task possible For vehicles, we need their full kinematics state to control for them All of this directly comes from neural networks video streams raw video streams come into the networks Goes through a lot of processing and then outputs the full kinematics state that positions velocities acceleration jerk all of that Directly comes out of networks with minimal post processing. That's really fascinating to me because how how much does it take? Even possible what world do we live in that this magic is possible that these networks predicts fourth derivatives of these positions and people thought We couldn't even detect these objects My opinion is that it did not come for free It it required tons of data.\n\nSo we had to be sophisticated auto labeling systems that shone through raw sensor data Run a ton of offline compute on the servers. It took a lot of time. It took a lot of time. It took a lot of time Run a ton of offline compute on the servers. It can take a few hours run expensive neural networks Distill the information into labels that train our in-car neural networks On top of this we also use our simulation system to synthetically create images and since it's a simulation We trivially have all the labels All of this goes through a well oiled data engine pipeline where we first train a baseline model with some data Ship it to the car see what the failures are and once we know the failures We mind the fleet for the cases where it fails Provide the correct labels and add the data to the training set This process systematically fixes the issues and we do this for every task that runs in the car Yeah, and to train these new massive neural networks This year we expanded our training infrastructure by roughly 40 to 50 percent So that sits us at about 14,000 GPUs today across multiple training clusters in the United States We also worked on our AI compiler which now supports new operations needed by those neural networks And map them to the the best of our underlying hardware resources And our inference engine today is capable of distributing the execution of a single neural network across two independent system on chips Essentially two independent computers interconnected within the same full self-driving computer And to make this possible we have to keep a tight control on the end-to-end latency of this new system So we deployed more advanced scheduling code across the full FSD platform All of these neural networks running in the car Together produce the vector space, which is again the model of the world around the robot or the car And then the planning system operates on top of this coming up with trajectories that avoid collisions or smooth Make progress towards the destination using a combination of model-based optimization Plus neural network that helps optimize it to be really fast Today we are really excited to present progress on all of these areas We have the engineering leads standing by to come in and explain these various blocks and these power not just the car But the same components also run on the Optimus robot that Milan showed earlier With that I welcome Paril to start talking about the planning section Hi all, I'm Paril Jain Let's use this intersection scenario today Let's use this intersection scenario to dive straight into how we do the planning and decision making in autopilot So we are approaching this intersection from a side street and we have to yield to all the crossing vehicles Right with as they are about to enter the intersection The pedestrian on the other side of the intersection decides to cross the road without a crosswalk Now we need to yield to this pedestrian Yield to the vehicles from the right and also understand the relation between the pedestrian and the vehicle on the other side of the intersection So a lot of these intra object dependencies That we need to resolve in a quick glance And humans are really good at this We look at a scene understand all the possible interactions evaluate the most promising ones And generally end up choosing a reasonable one So let's look at a few of these interactions that autopilot system evaluated We could have gone in front of this pedestrian with a very aggressive longitudinal lateral profile Now obviously we are being a jerk to the pedestrian and we would spook the pedestrian and his cute pet We could have moved forward slowly Short for a gap between the pedestrian or end the vehicle from the right Again, we are being a jerk to the vehicle coming from the right But you should not outright reject this interaction in case this is only safe interaction available Lastly the interaction we ended up choosing Stay slow initially find the reasonable gap and then finish the maneuver after all the agents pass Now evaluation of all of these interactions is not trivial Especially when you care about modeling the higher order derivatives for other agents For example, what is the longitudinal jerk required by the vehicle coming from the right when you assert in front of it? Relying purely on collision checks with marginal predictions will only get you so far because you will miss out on a lot of valid interactions This basically boils down to solving a multi-agent joint trajectory planning problem over the trajectories of ego and all the other agents Now how much ever you optimize there's going to be a limit to how fast you can run this optimization problem It will be close to close to order of 10 milliseconds even after a lot of incremental approximations Now for a typical crowded unprotected lift Say you have more than 20 objects Each object having multiple different future modes the number of relevant interaction combinations will blow up The planner needs to make a decision every 50 milliseconds.\n\nSo how do we solve this in real time? We rely on a framework what we call as interaction search, which is basically a paralyzed research over a bunch of maneuver trajectories The state space here corresponds to the kinematic state of ego, the kinematic state of other agents, their nominal future multiple multi-modal predictions and all the static entities in the scene The action space is where things get interesting We use a set of maneuver trajectory candidates to branch over a bunch of interaction decisions and also incremental goals for a longer horizon maneuver Let's walk through this research very quickly to get a sense of how it works We start with a set of vision measurements namely lanes occupancy moving objects These get represented as past attractions as well as latent features We use this to create a set of goal candidates Lanes again from the lanes network or unstructured regions which correspond to a probability mask derived from human demonstrations Once we have a bunch of these goal candidates, we create three trajectories using a combination of classical optimization approaches As well as our network planner again trained on data from the customer fleet Now once we get a bunch of these three trajectories We use them to start branching on the interactions We find the most critical interaction In our case, this would be the interaction with respect to the pedestrian Whether we assert in front of it or yield to it Obviously the option on the left is a high penalty option, it likely won't get prioritized So we branch further onto the option on the right and that's where we bring in more and more complex interactions Building this optimization problem incrementally with more and more constraints And the tree search keeps flowing, branching on more interactions, branching on more goals Now a lot of pricks here lie in evaluation of each of this node of the tree search Inside each node, initially we started with creating trajectories using classical optimization approaches Where the constraints like I described would be added incrementally And this would take close to 1 to 5 milliseconds per action Now even though this is fairly good number, when you want to evaluate more than 100% interactions, this does not scale So we ended up building lightweight queryable networks that you can run in the loop of the planner These networks are trained on human demonstrations from the fleet as well as offline solvers with relaxed time limits With this, we were able to bring the run time down to close to 100 microseconds per action Now doing this alone is not enough because you still have this massive tree search that you need to go through And you need to efficiently prune the search space So you need to do a new scoring on each of these trajectories Few of these are fairly standard, you do a bunch of collision checks, you do a bunch of comfort analysis What is the jerk and access required for a given manure The customer fleet data plays an important role here again We run two sets of again lightweight queryable networks, both really augmenting each other One of them trained from interventions from the FSD beta fleet Which gives a score on how likely is a given manure to result in interventions over the next few seconds And second, which is purely on human demonstrations, human driven data, giving a score on how close is your given selected action to a human driven trajectory The scoring helps us prune the search space, keep branching further on the interactions and focus the compute on the most promising outcomes The cool part about this architecture is that it allows us to create a cool blend between data driven approaches where you don't have to rely on a lot of hand engineered costs But also ground it in reality with physics based checks Now a lot of what I described was with respect to the agents, we could observe in the scene But the same framework extends to all of the other systems that we have We use the video feed from 8 cameras to generate the 3D occupancy of the world The blue mask here corresponds to the visibility region, we call it It basically gets blocked at the first occlusion you see in the scene We consume this visibility mask to generate the visibility of the scene We use the video feed from 8 cameras to generate the 3D occupancy of the world The blue mask here corresponds to the visibility region, we call it In the first occlusion you see in the scene, we consume this visibility mask to generate what we call as ghost objects which you can see on the top left Now if you model the spawn regions and the state transitions of this ghost objects correctly If you tune your control response as a function of their existence likelihood, you can extract some really nice human-like behaviors Now I'll pass it on to Phil to describe more on how we generate these occupancy networks Hey guys, my name is Phil, I will share the details of the occupancy network we built over the past year This network is our solution to model the physical work in 3D around our cars And it is currently not shown in our customer-facing visualization What you will see here is the raw network output from our internal lab tool The occupancy network takes video streams of all our 8 cameras as input Produces a single unified volumetric occupancy in vector space directly For every 3D location around our car, it predicts the probability of that location being occupied or not Since it has video contacts, it is capable of predicting obstacles that are occluded instantaneously For each location, it also produces a set of semantics such as curb, car, pedestrian, and road debris as color-coded here Occupancy flow is also predicted for motion Since the model is a generalized network, it does not tell static and dynamic objects explicitly It is able to produce and model the random motion such as a swarming trainer here This network is currently running in all Teslas with FSD computers And it is incredibly efficient, runs about every 10 milliseconds with our neural-line accelerator So how does this work? Let's take a look at architecture First, we rectify each camera image with a camera calibration And the images we're showing here are given to the network It's actually not the typical 8-bit RGB image As you can see from the first image on top, we're giving the 12-bit raw photo-account image to the network Since it has 4 bits more information, it has 16 times better dynamic range as well as reduced latency Since we don't have to run ISP in the loop anymore We use a set of reglets and bif-fps as a backbone to extract image space features Next, we construct a set of 3D position queries along with the image space features as keys and values fit into an attention module The output of the attention module is high-dimensional spatial features These spatial features are aligned temporally using vehicle odometry to derive motion Next, these spatial temporal features go through a set of deconvolutions to produce the final occupancy and occupancy flow output They're formed as fixed-size voxel grids, which might not be precise enough for planning on control In order to get a higher resolution, we also produce per voxel feature maps which we feed into MLP with 3D spatial point queries to get position and semantics at any arbitrary location After knowing the model better, let's take a look at another example Here we have an articulated bus parked on the right side of the road, highlighted as an L-shaped voxel here As we approach, the bus starts to move. The front of the car turns blue first, indicating the model predicts The front of the bus has a long zero occupancy flow As the bus keeps moving, the entire bus turns blue, and you can also see that the network predicts the precise curvature of the bus This is a very complicated problem for a traditional object detection network, as you'll have to see whether I'm going to use one cuboid or perhaps two to feed the curvature But for an occupancy network, since all we care about is the occupancy in the visible space, we'll be able to model the curvature precisely Besides the voxel grid, the occupancy network also produces a drivel surface The drivel surface has both 3D geometry and semantics. They are very useful for control, especially on hilly and curvy roads The surface and the voxel grid are not predicted independently. Instead, the voxel grid actually aligns with the surface implicitly Here, we are at a hill quest where you can see the 3D geometry of the surface being predicted nicely Planner can use this information to decide perhaps we need to slow down more for the hill quest And as you can also see, the voxel grid aligns with the surface consistently Besides the voxels and the surface, we're also very excited about the recent breakthrough in Neural Radiance Field or NERF We're looking into both incorporating some of the last NERF features into occupancy network training as well as using our network output as the input state for NERF As a matter of fact, Ashok is very excited about this.\n\nThis has been his personal weekend project for a while About these NERFs, because I think the academia is building out of these foundation models for language using tons of large data sets for language But I think for vision, NERFs are going to provide the foundation models for computer vision because they are grounded in geometry And geometry gives us a nice way to supervise these networks and freezes off the requirement to define an ontology And the supervision is essentially free because you just have to differentially render these images So I think in the future, this occupancy network idea where images come in and then the network produces a consistent volumetric representation of the scene That can then be differentially rendered into any image that was observed I personally think it's a future of computer vision and we do some initial work on it right now But I think in the future, both at Tesla and in academia, we will see that this combination of one-shot prediction of volumetric occupancy will be the future That's my personal bet Thanks Ashok So here's an example early result of a 3D reconstruction from our free data Instead of focusing on getting perfect RGB reproduction in image space, our primary goal here is to accurately represent the world in 3D space for driving And we want to do this for all our free data over the world in all weather and lighting conditions And obviously this is a very challenging problem and we're looking for you guys to help Finally, the occupancy network is trained with large auto-labeled data sets without any human in the loop And with that, I'll pass to Tim to talk about what it takes to train this network Thanks Phil Alright, hey everyone Let's talk about some training infrastructure So we've seen a couple of videos, no four or five I think and care more and worry more about a lot more clips on that So we've been looking at the occupancy networks just from Phil Just Phil's videos, it takes 1.4 billion frames to train that network What you just saw and if you have 100,000 GPUs, it would take one hour But if you have one GPU, it would take 100,000 hours So that is not a humane time period that you can wait for your training job to run, right? We want to ship faster than that So that means you're going to need to go parallel So you need a more compute for that That means you're going to need a supercomputer So this is why we've built in-house three supercomputers comprising of 14,000 GPUs Where we use 10,000 GPUs for training and around 4,000 GPUs for auto-labeling All these videos are stored in 30 petabytes of a distributed managed video cache You shouldn't think of our data sets as fixed Let's say as you think of your image net or something, you know, with like a million frames You should think of it as a very fluid thing So we've got half a million of these videos flowing in and out of this cluster These clusters every single day And we track 400,000 of these kind of Python video instantiations every second So that's a lot of calls We're going to need to capture that in order to govern the retention policies of this distributed video cache So underlying all of this is a huge amount of infra, all of which we build and manage in-house So you cannot just buy, you know, 14,000 GPUs and then 30 petabytes of Flash NVMe And you just put it together and let's go train It actually takes a lot of work and I'm going to go into a little bit of that What you actually typically want to do is you want to take your accelerator So that could be the GPU or dojo, which we'll talk about later And because that's the most expensive component, that's where you want to put your bottleneck And so that means that every single part of your system is going to need to outperform this accelerator And so that is really complicated That means that your storage is going to need to have the size and the bandwidth to deliver all the data down into the nodes These nodes need to have the right amount of CPU and memory capabilities to feed into your machine learning framework This machine learning framework then needs to hand it off to your GPU and then you can start training But then you need to do so across hundreds or thousands of GPU in a reliable way in lockstep And in a way that's also fast, so you're also going to need an interconnect Extremely complicated We'll talk more about dojo in a second So first I want to take you through some optimizations that we've done on our cluster So we're getting in a lot of videos and video is very much unlike, let's say, training on images or text Which I think is very well established Video is quite literally a dimension more complicated And so that's why we needed to go end to end from the storage layer down to the accelerator Optimize every single piece of that Because we train on the photon count videos that come directly from our fleet We train on those directly, we do not post-process those at all The way it's just done is we seek exactly to the frames we select for our batch We load those in including the frames that they depend on, so these are your eye frames or your key frames We package those up, move them into shared memory, move them into a double bar from the GPU And then use the hardware decoder that's only accelerated to actually decode the video So we do that on the GPU natively, and this is all in a very nice PyTorch extension Doing so unlocked more than 30% training speed increase for the occupancy networks And freed up basically a whole CPU to do any other thing You cannot just do training with just videos, of course you need some kind of a ground truth And that is actually an interesting problem as well The objective for storing your ground truth is that you want to make sure you get to your ground truth That you need in the minimal amount of file system operations And load in the minimal size of what you need in order to optimize for aggregate cross cluster throughput Because you should see a compute cluster as one big device which has internally fixed constraints and thresholds So for this we rolled out a format that is native to us that's called small We use this for our ground truth, our feature cache and any inference outputs So a lot of tensors that are in there And so just a cartoon here, let's say this is your table that you want to store Then that's how that would look out if you rolled out on disk So what you do is you take anything you'd want to index on, so for example video timestamps You put those all in the header so that in your initial header read you know exactly where to go on disk Then if you have any tensors you're going to try to transpose the dimensions to put a different dimension last as the contiguous dimension And then also try different types of compression Then you check out which one was most optimal and then store that one This is actually a huge tip if you do feature caching Unintelligible output from the machine learning network Rotate around the dimensions a little bit, you can get up to 20% increase in efficiency of storage Then when you store that we also order the columns by size So that all your small columns and small values are together So that when you seek for a single value you're likely to overlap with a read on more values which you'll use later So that you don't need to do another file system operation So I could go on and on, I just went on, touched on two projects that we have internally This is actually part of a huge continuous effort to optimize the compute that we have in-house So accumulating and aggregating through all these optimizations We now train our occupancy networks twice as fast just because it's twice as efficient And now if we add in a bunch more compute and go parallel we can now train this in hours instead of days And with that I'd like to hand it off to the biggest user of compute, John Hi everybody, my name is John Emmons, I lead the autopilot vision team I'm going to cover two topics with you today, the first is how we predict lanes And the second is how we predict the future behavior of other agents on the road In the early days of autopilot we modeled the lane detection problem as an image space instant segmentation task Our network was super simple though, in fact it was only capable of predicting lanes from a few different kinds of geometries Specifically it would segment the ego lane, it could segment adjacent lanes, and then it had some special casing for forks and merges This simplistic modeling of the problem worked for highly structured roads like highways But today we're trying to build a system that's capable of much more complex maneuvers Specifically we want to make left and right turns at intersections where the road topology can be quite a bit more complex and diverse When we try to apply this simplistic modeling of the problem here, it just totally breaks down Taking a step back for a moment, what we're trying to do here is to predict the sparse set of lane instances and their connectivity And what we want to do is to have a neural network that basically predicts this graph where the nodes are the lane segments And the edges encode the connectivity between these lanes So what we have is our lane detection neural network, it's made up of three components In the first component we have a set of convolutional layers, attention layers, and other neural network layers That encode the video streams from our eight cameras on the vehicle and produce a rich visual representation We then enhance this visual representation with a coarse road level map data Which we encode with a set of additional neural network layers that we call the lane guidance module This map is not an HD map, but it provides a lot of useful hints about the topology of lanes inside of intersections, the lane counts on various roads, and a set of other attributes that help us The first two components here produce a dense tensor that sort of encodes the world But what we really want to do is to convert this dense tensor into a sparse set of lanes and their connectivity We approach this problem like an image captioning task where the input is this dense tensor and the output text is predicted into a special language that we developed at Tesla for encoding lanes and their connectivity In this language of lanes, the words and tokens are the lane positions in 3D space In the ordering of the tokens, encrypted modifiers in the tokens encode the connected relationships between these lanes By modeling the task as a language problem, we can capitalize on recent autoregressive architectures and techniques from the language community for handling the multiple-diality of the problem We're not just solving the computer vision problem at Autopilot, we're also applying the state-of-the-art in language modeling and machine learning more generally I'm now going to dive into a little bit more detail of this language component What I have depicted on the screen here is a satellite image which sort of represents the local area around the vehicle The set of nose and edges is what we refer to as the lane graph, and it's ultimately what we want to come out of this neural network We start with a blank slate We're going to want to make our first prediction here at this green dot This green dot's position is encoded as an index into a course grid which discretizes the 3D world Now we don't predict this index directly because it would be too computationally expensive to do so There's just too many grid points and predicting a categorical distribution over this has both implications at training time and test time So instead what we do is we discretize the world coarsely first, we predict the heat map over the possible locations, and then we latch in the most probable location Condition on this, we then refine the prediction and get the precise point Now we know where the position of this token is, but we don't know it's tight In this case though, it's a beginning of a new lane So we predict it as a start token And because it's a start token, there's no additional attributes in our language We then take the predictions from this first forward pass, and we encode them using a learned positional embedding Which produces a set of tensors that we combine together Which is actually the first word in our language of lanes We add this to the first position in our sentence here We then continue this process by predicting the next lane point in a similar fashion Now this lane point is not the beginning of a new lane, it's actually a continuation of the previous lane So it's a continuation token type Now it's not enough just to know that this lane is connected to the previously predicted lane We want to encode its precise geometry, which we do by regressing a set of spline coefficients We then take this lane, we encode it again, and add it as the next word in the sentence We continue predicting these continuation lanes until we get to the end of the prediction grid We then move on to a different lane segment So you can see that cyan dot there Now it's not topologically connected to that pink point It's actually forking off of that green point there So it's got a fork type And fork tokens actually point back to previous tokens from which their fork originates So you can see here the fork point predictor is actually the index zero So it's actually referencing back to a token that is already predicted, like you would in language We continue this process over and over again until we've enumerated all of the tokens in the lane graph And then the network predicts the end of sentence token Yeah, I just wanted to note that the reason we do this is not just because we want to build something complicated It almost feels like a Turing complete machine here with neural networks though Is that we try simple approaches, for example, trying to just segment the lanes along the road or something like that But then the problem is when there's uncertainty, say you cannot see the road clearly And there could be two lanes or three lanes and you can't tell A simple segmentation-based approach would just draw both of them It's kind of a 2.5 lane situation And the post-processing algorithm would hilariously fail when the predictions are such Yeah, the problems don't end there I mean, you need to predict these connective lanes inside of intersections Which is just not possible with the approach that Ashok's mentioning Which is why we had to upgrade to this sort of approach Yeah, when it overlaps like this, segmentation would just go haywire But even if you try very hard to put them on separate layers, it's just a really hard problem But language just offers a really nice framework for getting a sample from a posterior As opposed to trying to do all of this in post-processing But this doesn't actually stop for just autopilot, right? John, this can be used for optimists Yeah, I guess they wouldn't be called lanes But you could imagine, sort of in this stage here That you might have sort of paths that sort of encode the possible places that people could walk Yeah, basically if you're in a factory or in a home setting, you can just ask the robot Okay, please route to the kitchen or please route to some location in the factory And then we predict a set of pathways that would go through the aisles, take the robot And say, okay, this is how you get to the kitchen It just really gives us a nice framework to model these different paths That simplify the navigation problem for the downstream planner Alright, so ultimately what we get from this lane detection network Is a set of lanes in their connectivity, which comes directly from the network There's no additional step here for sparsifying these dense predictions into sparse ones This is just a direct unfiltered output of the network Okay, so I talked a little bit about lanes I'm going to briefly touch on how we model and predict the future paths and other semantics on objects So I'm just going to go really quickly through two examples The video on the right here, we've got a car that's actually running a red light and turning in front of us What we do to handle situations like this is we predict a set of short time horizon future trajectories on all objects We can use these to anticipate the dangerous situation here And apply whatever breaking and steering actions required to avoid a collision In the video on the right, there's two vehicles in front of us The one on the left lane is parked, apparently it's being loaded, unloaded I don't know why the driver decided to park there But the important thing is that our neural network predicted that it was stopped Which is the red color there The vehicle in the other lane, as you notice, also is stationary But that one's obviously just waiting for that red light to turn green So even though both objects are stationary and have zero velocity It's the semantics that is really important here So that we don't get stuck behind that awkwardly parked car Predicting all of these agent attributes presents some practical problems when trying to build a real-time system We need to maximize the frame rate of our object section stack So that autopilot can quickly react to the changing environment Every millisecond really matters here To minimize the inference latency, our neural network is split into two phases In the first phase, we identified the locations in 3D space where agents exist In the second stage, we then pull out tensors at those 3D locations Append it with additional data that's on the vehicle And then we do the rest of the processing This specification step allows the neural network to focus compute on the areas that matter most Which gives us superior performance for a fraction of the latency cost So, putting it all together The autopilot vision stack predicts more than just the geometry and kinematics of the world It also predicts a rich set of semantics, which enables safe and human-like driving I'm now going to hand things off to Sri who will tell us how we run all these cool neural networks on our FSD computer Thank you Hi everyone, I'm Sri Today I'm going to give a glimpse of what it takes to run these FSD networks in the car And how do we optimize for the inference latency? Today I'm going to focus just on the FSD lanes network that John just talked about So, when we started this track, we wanted to know if we can run this FSD lanes network natively on the trip engine Which is our in-house neural network accelerator that we built in the FSD computer When we built this hardware, we kept it simple and we made sure it can do one thing ridiculously fast Dense dot products But this architecture is autoregressive and iterative Where it crunches through multiple attention-attention blocks in the inner loop Producing sparse points directly at every step So, the challenge here was how can we do this sparse point prediction and sparse computation on a dense dot product engine Let's see how we did this on the trip So, the network predicts the heat map of most probable spatial locations of the point To do this on trip, we actually built a lookup table in SRAM And we engineered the dimensions of this embedding such that we could achieve all of this thing with just matrix multiplication Not just that, we also wanted to store this embedding into a token cache So that we don't recompute this for every iteration, rather reuse it for future point prediction Again, we put some tricks here where we did all these operations just on the dot product engine It's actually cool that our team found creative ways to map all these operations on the trip engine In ways that were not even imagined when this hardware was designed But that's not the only thing we had to do to make this work We actually implemented a whole lot of operations and features to make this model compilable To improve the intate accuracy as well as to optimize performance All of these things helped us run this 75 million parameter model just under 10 millisecond of latency Consuming just 8 watts of power But this is not the only architecture running in the car There are so many other architectures, modules and networks we need to run in the car To give a sense of scale, there are about a billion parameters of all the networks combined Producing around 1000 neural network signals So we need to make sure we optimize them jointly and such that we maximize the compute utilization Throughput and minimize the latency So we built a compiler just for neural networks that shares the structure to traditional compilers As you can see, it takes the massive graph of neural nets with 150k nodes and 375k connection Takes this thing, partitions them into independent subgraphs And compiles each of those subgraphs natively for the inference devices Then we have a neural network linker which shares the structure to traditional linker Where we perform this link time optimization There we solve an offline optimization problem with compute memory and memory band with constraints So that it comes with an optimized schedule that gets executed in the car On the runtime, we designed a hybrid scheduling system which basically does heterogeneous scheduling on one SOC And distributed scheduling across both the SOCs to run these networks in a model parallel fashion To get 100 tops of compute utilization, we need to optimize across all the layers of software Right from tuning the network architecture, the compiler, all the way to implementing a low latency high bandwidth RDMA link Across both the SOCs and in fact going even deeper to understanding and optimizing the cache coherent and non-coherent data path of the accelerator in the SOC This is a lot of optimization at every level in order to make sure we get the highest frame rate and as every millisecond counts here And this is just the visualization of the neural networks that are running in the car This is our digital brain essentially As you can see these operations are nothing but just the matrix multiplication, convolution to name a few real operations running in the car To train this network with a billion parameters, you need a lot of labeled data So Egan is going to talk about how do we achieve this with the auto labeling pipeline Thank you Sri Hi everyone, I'm Egan Zhang and I'm leading a geometric vision at autopilot So yeah, let's talk about auto labeling So we have several kinds of auto labeling frameworks to support various types of networks But today I'd like to focus on the awesome lanes net here So to successfully train and generalize this network to everywhere, we think we went tens of millions of trips from probably one million intersection or even more Than how to do that So it is certainly achievable to source sufficient amount of trips because we already have, as Tim explained earlier, we already have like 500,000 trips per day cache rate However, converting all those data into a training form is a very challenging technical problem To solve this challenge, we've tried various ways of manual and auto labeling So from the first column to the second, from the second to the third, each advance provided us nearly 100x improvement in throughput But still, we run an even better auto labeling machine that can provide us good quality, diversity and scalability To meet all these requirements, despite the huge amount of engineering effort required here, we've developed a new auto labeling machine powered by multi-trip reconstruction So this can replace 5 million hours of manual labeling with just 12 hours on cluster for labeling 10,000 trips So how we solved? There are three big steps. The first step is high precision trajectory and structural recovery by multi-camera, visual, inertial, or geometry So here, all the features including ground surface are inferred from videos by neural networks, then tracked and reconstructed in the vector space So the typical trip rate of this trajectory in car is like 1.3 centimeter per meter and 0.45 milliliter per meter, which is pretty decent considering its compact compute requirement Then the recovered surface and road details are also used as a strong guidance for the later manual verification stuff This is also enabled in every FSD vehicle, so we get preprocessed trajectories and structures along with the trip data The second step is multi-trip reconstruction, which is the big and core piece of this machine So the video shows how the previously shown trip is reconstructed and aligned with other trips, basically other trips from different vehicles, not the same vehicle So this is done by multiple internal steps like course alignment, pairwise matching, joint optimization, then further surface refinement In the end, the human analyst comes in and finalizes the label So each heavy steps are already fully parallelized on the cluster, so the entire process usually takes just a couple of hours The last step is actually auto-labeling the new trips So here we use the same multi-trip alignment engine, but only between pre-built reconstruction and each new trip So it's much, much simpler than fully reconstructing all the clips altogether That's why it only takes 30 minutes per trip to auto-label instead of several hours of manual labeling And this is also the key of scalability of this machine This machine easily scales as long as we have available compute and trip data So about 50 trips were newly auto-labeled from this scene and some of them are shown here, so 53 from different vehicles So this is how we capture and transform the space-time slices of the world into the network supervision One thing I'd like to note is that Jagan just talked about how we auto-label our lanes We have auto-labels for almost every task that we do, including our planner And many of these are fully automatic, there's no humans involved For example, for objects, all the kinematics, the shapes, the futures, everything just comes from auto-labeling And the same is true for our occupancy too, and we have really just built a machine around this Yeah, so if you can go back one slide One more, it says parallelized on cluster So that sounds pretty straightforward, but it really wasn't Maybe it's fun to share how something like this comes about So a while ago we didn't have any auto-labeling at all, and then someone makes a script It starts to work, it starts working better, until you reach a volume that's pretty high And we clearly need a solution And so there were two other engineers in our team who were like, you know, that's an interesting, you know, thing What we needed to do was build a whole graph of essentially Python functions that would need to run one after the other First you pull the clip, then you do some cleaning, then you do some network inference, then another network inference Until you finally get this But so you need to do this at a large scale, so I tell them we probably need to shoot for, you know, 100,000 clips per day Or like 100,000 items, that seems good And so the engineers said, well, we can do, you know, a bit of post-gres and a bit of elbow grease, we can do it Meanwhile, we are a bit later and we're doing 20 million of these functions every single day Again, we pull in around half a million clips and on those we run a ton of functions, each of these, in a streaming fashion And so that's kind of the backend infra that's also needed to not just run training, but also auto-labeling Yeah, it really is like a factory that produces labels and production lines, yield, quality, inventory Like all of these same concepts applied to this label factory that applies for, you know, the factory for our cars That's right Okay, thanks, Tim and Ashok So, yeah, so concluding this section, I'd like to share a few more challenging and interesting examples for network for sure And even for humans, probably So from the top, there's like examples for like lack of lights, case or foggy night or roundabout and occlusions by heavy occlusions by parked cars And even rainy night with rain drops on camera lenses These are challenging, but once their original scenes are fully reconstructed by other clips, all of them can be auto-labeled So that our cars can drive even better through these challenging scenarios So, now, let me pass the mic to David to learn more about how Sim is creating the new world on top of these labels Thank you Thank you, Yegan My name is David and I'm going to talk about simulation So simulation plays a critical role in providing data that is difficult to source and or hard to label However, 3D scenes are notoriously slow to produce Take for example, the simulated scene playing behind me A complex intersection from Market Street in San Francisco It would take two weeks for artists to complete And for us, that is painfully slow However, I'm going to talk about using Yegan's automated ground truth labels along with some brand new tooling that allows us to procedurally generate this scene in many like it in just five minutes That's an amazing a thousand times faster than before So let's dive in to how a scene like this is created We start by piping the automated ground truth labels into our simulated world creator tooling inside the software Houdini Starting with road boundary labels, we can generate a solid road mesh and re-topologize it with the lane graph labels This helps inform important road details like cross-road slope and detailed material blending Next, we can use the line data and sweep geometry across its surface and project it to the road, creating lane paint decals Next, using median edges, we can spawned island geometry and populate it with randomized foliage This drastically changes the visibility of the scene Now the outside world can be generated through a series of randomized heuristics Modular building generators create visual obstructions while randomly placed objects like hydrants can change the color of the curves while trees can drop leaves below it obscuring lines or edges Next, we can bring in map data to inform positions of things like traffic traffic lights or stop signs We can trace along its normal to collect important information like number of lanes and even get accurate street names on the signs themselves Next, using lane graph, we can determine lane connectivity and spawn directional road markings on the road and their accompanying road signs And finally, with lane graph itself, we can determine lane adjacency and other useful metrics to spawn randomized traffic permutations inside our simulator And again, this is all automatic, no artist in the loop and happens within minutes And now this sets us up to do some pretty cool things Since everything is based on data and heuristics, we can start to fuzz parameters to create visual variations of the single ground truth It can be as subtle as object placement and random material swapping to more drastic changes like entirely new biomes or locations of environment like urban, suburban, or rural This allows us to create infinite, targeted permutations for specific ground truths that we need more ground truth for And all this happens within a click of a button And we can even take this one step further by altering our ground truth itself Say John wants his network to pay more attention to directional road markings to better detect an upcoming captive left turn lane We can start to procedurally alter our lane graph inside the simulator to help create entirely new flows through this intersection to help focus the network's attention to the road markings to create more accurate predictions And this is a great example of how this tooling allows us to create new data that can never be collected from the real world And the true power of this tool is in its architecture and how we can run all tasks in parallel to infinitely scale So you saw the tile creator tool in action converting the ground truth labels into their counterparts Next we can use our tile extractor tool to divide this data into geo hash tiles about 150 meter square in size We then save out that data into separate geometry and instance files This gives us a clean source of data that's easy to load and allows us to be rendering engine agnostic for the future Then using a tile loader tool we can summon any number of those cache tiles using a geo hash ID Currently we're doing about these 5x5 tiles or 3x3 usually centered around fleet hotspots or interesting lane graph locations And the tile loader also converts these tile sets into U assets for consumption by the unreal engine and gives you a finished product from what you saw in the first slide And this really sets us up for size and scale And as you can see on the map behind us we can easily generate most of San Francisco city streets And this didn't take years or even months of work but rather two weeks by one person We can continue to manage and grow all this data using our PDG network inside of the tooling This allows us to throw compute at it and regenerate all these tile sets overnight This ensures all environments are consistent, quality and features which is super important for training since new ontologies and signals are constantly released And now to come full circle, because we generated all these tile sets from ground truth data They contain all the weird intricacies from the real world We can combine that with the procedural, visual and traffic variety to create limitless, targeted data for the network to learn from And that concludes the SIM section, I'll pass it to Kate to talk about how we can use all this data to improve autopilot Thank you Thanks David, hi everyone, my name is Kate Park and I'm here to talk about the data engine Which is the process by which we improve our neural networks via data We're going to show you how we deterministically solve interventions via data And walk you through the life of this particular clip In this scenario, autopilot is approaching a turn and incorrectly predicts that crossing vehicle as stopped for traffic and thus a vehicle that we would slow down for In reality, there's nobody in the car, it's just awkwardly parked We've built this tooling to identify the mispredictions, correct the label and categorize this clip into an evaluation set This particular clip happens to be one of 126 that we've diagnosed as challenging parked cars at turns Because of this infra, we can curate this evaluation set without any engineering resources custom to this particular challenge case To actually solve that challenge case requires mining thousands of examples like it And it's something Tesla can trivially do We simply use our data sourcing infra, request data and use the tooling shown previously to correct the labels By surgically targeting the mispredictions of the current model, we're only adding the most valuable examples to our training set We surgically fix 13,900 clips and because those were examples where the current model struggles We don't even need to change the model architecture, a simple weight update with this new valuable data is enough to solve the challenge case So you see we no longer predict that crossing vehicle as stopped, as shown in orange, but parked, as shown in red In academia, we often see that people keep data constant, but at Tesla it's very much the opposite We see time and time and again that data is one of the best if not the most deterministic lever to solving these interventions We just showed you the data engine loop for one challenge case, namely these parked cars at turns But there are many challenge cases even for one signal of vehicle movement We apply this data engine loop to every single challenge case we've diagnosed, whether it's buses, curvy roads, stopped vehicles, parking lots And we don't just add data once, we do this again and again to perfect the semantic In fact, this year we updated our vehicle movement signal five times and with every weight update trained on the new data We push our vehicle movement accuracy up and up This data engine framework applies to all our signals, whether they're 3D, multi-cam video, whether the data is human labeled, auto-labeled, or simulated Whether it's an offline model or an online model And Tesla is able to do this at scale because of the fleet advantage, the infra that our NG team has built, and the labeling resources that feed our networks To train on all this data, we need a massive amount of compute, so I'll hand it off to Pete and Ganesh to talk about the Dojo supercomputing platform Thank you Thank you, Katie Thanks everybody, thanks for hanging in there, we're almost there My name is Pete Bannon, I run the custom silicon and low voltage teams at Tesla And my name is Ganesh Renke, I run the Dojo program Thank you I'm frequently asked, why is a car company building a supercomputer for training?\n\nAnd this question fundamentally misunderstands the nature of Tesla At its heart, Tesla is a hardcore technology company All across the company, people are working hard in science and engineering to advance the fundamental understanding and methods that we have available to build cars, energy solutions, robots, and anything else that we can do to improve the human condition around the world It's a super exciting thing to be a part of, and it's a privilege to run a very small piece of it in the semiconductor group Tonight we're going to talk a little bit about Dojo and give you an update on what we've been able to do over the last year But before we do that, I wanted to give a little bit of background on the initial design that we started a few years ago When we got started, the goal was to provide a substantial improvement to the training latency for our autopilot team Some of the largest neural networks they train today run for over a month, which inhibits their ability to rapidly explore alternatives and evaluate them So a 30X speedup would be really nice if we could provide it at a cost competitive and energy competitive way To do that, we wanted to build a chip with a lot of arithmetic units that we could utilize at a very high efficiency And we spent a lot of time studying whether we could do that using DRAM, various packaging ideas, all of which failed And in the end, even though it felt like an unnatural act, we decided to reject DRAM as the primary storage medium for this system And instead focus on SRAM embedded in the chip SRAM provides, unfortunately, a modest amount of capacity, but extremely high bandwidth and very low latency, and that enables us to achieve high utilization with the arithmetic units Those choices, that particular choice led to a whole bunch of other choices For example, if you want to have virtual memory, you need page tables, they take up a lot of space, we didn't have space, so no virtual memory So we also don't have interrupts, the accelerator is a bare bonds, raw piece of hardware that's presented to a compiler and the compiler is responsible for scheduling everything that happens in a deterministic way So there's no need or even desire for interrupts in the system We also chose to pursue model parallelism as a training methodology, which is not the typical situation most machines today use data parallelism, which consumes additional memory capacity, which we obviously don't have So all of those choices led us to build a machine that is pretty radically different from what's available today We also had a whole bunch of other goals, one of the most important ones was no limits So we wanted to build a compute fabric that would scale in an unbounded way for the most part, I mean obviously there's physical limits now and then But pretty much if your model was too big for the computer, you just had to go buy a bigger computer, that's what we were looking for Today the way machines are packaged, there's a pretty fixed ratio of for example GPU, CPUs and DRAM capacity and network capacity And we really wanted to disaggregate all that so that as models evolved, we could vary the ratios of those various elements and make the system more flexible to meet the needs of the autopilot team And it's so true, no limits philosophy was our guiding star all the way, all of our choices were centered around that And to the point that we didn't want traditional data center infrastructure to limit our capacity to execute these programs at speed That's why we integrated vertically our data center, the entire data center by doing a vertical integration of the data center We could extract new levels of efficiency, we could optimize power delivery, cooling and as well as system management across the whole data center stack Rather than doing box by box and integrating that, those boxes into data centers And to do this, we also wanted to integrate early to figure out limits of scale for our software workloads So we integrated Dojo environment into our autopilot software very early and we learned a lot of lessons And today Bill Chang will go over our hardware update as well as some of the challenges that we faced along the way And Rajiv Kurian will give you a glimpse of our compiler technology as well as go over some of our cool results Great Thanks Pete, thanks Ganesh I'll start tonight with a high level vision of our system that will help set the stage for the challenges and the problems we're solving And then also how software will then leverage this for performance Now our vision for Dojo is to build a single unified accelerator, a very large one Software would see a seamless compute plane with globally addressable, very fast memory and all connected together with uniform high bandwidth and low latency Now to realize this, we need to use density to achieve performance Now we leverage technology to get this density in order to break levels of hierarchy all the way from the chip to the scale out systems Now silicon technology has done this for decades Chips have followed Moore's law for density integration to get performance scaling Now a key step in realizing that vision was our training tile Probably can we integrate 25 dies at extremely high bandwidth but we can scale that to any number of additional tiles by just connecting them together Now last year we showcased our first functional training tile and at that time we already had workloads running on it And since then the team here has been working hard and diligently to deploy this at scale Now we've made amazing progress and had a lot of milestones along the way And of course we've had a lot of unexpected challenges But this is where our fail fast philosophy has allowed us to push our boundaries Now pushing density for performance presents all new challenges One area is power delivery Here we need to deliver the power to our compute die and this directly impacts our top line compute performance But we need to do this at unprecedented density We need to be able to match our die pitch with a power density of almost 1 amp per millimeter squared And because of the extreme integration this needs to be a multi-tiered vertical power solution And because there's a complex heterogeneous material stack up we have to carefully manage the material transition Especially CTE Now why does the coefficient of thermal expansion matter in this case? CTE is a fundamental material property and if it's not carefully managed that stack up would literally rip itself apart We started this effort by working with vendors to develop this power solution But we realized that we actually had to develop this in-house Now to balance schedule and risk we built quick iterations to support both our system bring up in software development And also to find the optimal design and stack up that would meet our final production goals And in the end we were able to reduce CTE over 50% and meet our performance by 3x over our initial version Now needless to say finding this optimal material stack up while maximizing performance at density is extremely difficult Now we did have unexpected challenges along the way Here's an example where we pushed the boundaries of integration that led to component failures This started when we scaled up to larger and longer workloads and then intermittently a single site on a tile would fail Now they started out as recoverable failures but as we pushed some much higher and higher power these would become permanent failures Now to understand this failure you have to understand why and how we build our power modules Solving density at every level is the cornerstone of actually achieving our system performance Now because our XY plane is used for high bandwidth communication everything else must be stacked vertically This means all other components other than our die must be integrated into our power modules Now that includes our clock and our power supplies and also our system controllers Now in this case the failures were due to losing clock output from our oscillators And after an extensive debug we found that the root cause was due to vibrations on the module from piezoelectric effects Our nearby capacitors Now singing caps are not a new phenomenon and in fact very common in power design But normally clock chips are placed in a very quiet area of the board and often not affected by power circuits But because we needed to achieve this level of integration these oscillators need to be placed in very close proximity Now due to our switching frequency and then the vibration resonance created It caused out of plane vibration on our MEMS oscillator that caused it to crack Now the solution to this problem is a multi-prong approach We can reduce the vibration by using soft terminal caps We can update our MEMS part with a lower Q factor for the out of plane direction And we can also update our switching frequency to push the resonance further away from these sensitive bands Now in addition to the density at the system level we've been making a lot of progress at the infrastructure level We knew that we had to read examine every aspect of the data center infrastructure in order to support our unprecedented power and cooling density We brought in a fully custom designed CDU to support Dojo's dense cooling requirements And the amazing part is we're able to do this at a fraction of the cost versus buying off the shelf and modifying it And since our Dojo cabinet integrates enough power and cooling to match an entire row of standard IT racks We need to carefully design our cabinet and infrastructure together And we've already gone through several iterations of this cabinet to optimize this And earlier this year we started low testing our power and cooling infrastructure And we were able to push it over 2 megawatts before we tripped our substation and got a call from the city Now last year we introduced only a couple of components of our system The custom D1 die and the training tile, but we teased the exit pod as our end goal We'll walk through the remaining parts of our system that are required to build out this exit pod Now the system tray is a key part of realizing our vision of a single accelerator It enables us to seamlessly connect tiles together, not only within the cabinet, but between cabinets We can connect these tiles at very tight spacing across the entire accelerator And this is how we achieve our uniform communication This is a laminated bus bar that allows us to integrate very high power, mechanical and thermal support, and an extremely dense integration It's 75 millimeters in height and supports 6 tiles at 135 kilograms This is the equivalent of 3 to 4 fully loaded high performance racks Next we need to feed data to the training tiles This is where we've developed the Dojo interface processor It provides our system with high bandwidth DRAM to stage our training data And it provides full memory bandwidth to our training tiles using TTP, our custom protocol that we use to communicate across our entire accelerator It also has high speed Ethernet that helps us extend this custom protocol over standard Ethernet And we provide native hardware support for this with little to no software overhead And lastly we can connect to it through a standard Gen4 PCIe interface Now we pair 20 of these cards per tray and that gives us 640 gigabytes of high bandwidth DRAM And this provides our disaggregated memory layer for our training tiles These cards are a high bandwidth ingest path both through PCIe and Ethernet They also provide a high-ratex Z-connectivity path that allows shortcuts across our large Dojo accelerator Now we actually integrate the host directly underneath our system tray These hosts provide our ingest processing and connect to our interface processors through PCIe These hosts can provide hardware video decoder support for video-based training And our user applications land on these hosts so we can provide them with the standard X86 Linux environment Now we can put two of these assemblies into one cabinet and pair it with redundant power supplies that do direct conversion of three-phase 480-volt AC power to 52-volt DC power Now by focusing on density at every level we can realize the vision of a single accelerator Now starting with the uniform nodes on our custom D1 die we can connect them together in our fully integrated training tile And then finally seamlessly connecting them across cabinet boundaries to form our Dojo accelerator And all together we can house two full accelerators in our Exapod for a combined one exa-flop of ML compute Now all together this amount of technology and integration has only ever been done a couple of times in the history of compute Next we'll see how software can leverage this to accelerate their performance Thanks Bill, my name is Rajiv and I'm going to talk some numbers So our software stack begins with the PyTorch extension that speaks to our commitment to run standard PyTorch models out of the box We're going to talk more about our JIT compiler and the ingest pipeline that feeds the hardware with data Abstractly, performance is tops times utilization times accelerator occupancy We've seen how the hardware provides peak performance is the job of the compiler to extract utilization from the hardware while code is running on it And it's the job of the ingest pipeline to make sure that data can be fed at a throughput high enough for the hardware to not ever starve So let's talk about why communication-bound models are difficult to scale But before that let's look at why ResNet 50-like models are easier to scale You start off with a single accelerator, run the forward and backward passes, followed by the optimizer Then to scale this up you run multiple copies of this on multiple accelerators And while the gradients produced by the backward pass do need to be reduced and this introduces some communication, this can be done pipeline with the backward pass This setup scales fairly well, almost linearly For models with much larger activations we run into a problem as soon as we want to run the forward pass The batch size that fits in a single accelerator is often smaller than the batch norm surface So to get around this researchers typically run this setup on multiple accelerators in sync batch norm mode This introduces latency bound communication to the critical path of the forward pass and we already have a communication bottleneck And while there are ways to get around this they usually involve tedious manual work best suited for a compiler And ultimately there's no skirting around the fact that if your state does not fit in a single accelerator you can be communication bound And even with significant efforts from our ML engineers we see such models don't scale linearly The doger system was built to make such models work at high utilization The high density integration was built to not only accelerate the compute bound portions of a model but also the latency bound portions Like a batch norm or the bandwidth bound portions like a gradient all reduced or a parameter all gathered A slice of the doger mesh can be carved out to run any model The only thing users need to do is to make the slice large enough to fit a batch norm surface for their particular model After that the partition presents itself as one large accelerator freeing the users from having to worry about the internal details of execution And as the job of the compiler to maintain this abstraction Fine grain synchronization primitives in uniform low latency makes it easy to accelerate all forms of parallelism across integration boundaries Tensors are usually stored sharded in SRAM and replicated just in time for a layer's execution We depend on the high doger bandwidth to hide this replication time Tensor replication and other data transfers are overlapped with compute and the compiler can also recompute layers when it's profitable to do so We expect most models to work out of the box As an example we took the recently released stable diffusion model and got it running on dojo in minutes Out of the box the compiler was able to map it in a model parallel manner on 25 dojo dies Here are some pictures of a Cybertruck on Mars generated by stable diffusion running on dojo Looks like it still has some ways to go before matching the Tesla design studio team So we've talked about how communication bottlenecks can hamper scalability Perhaps an asset test of a compiler and the underlying hardware is executing a cross die batch norm layer Like mentioned before this can be a serial bottleneck The communication phase of a batch norm begins with nodes computing their local mean and standard deviations Then coordinating to reduce these values, then broadcasting these values back and then they resume their work in parallel So what would an ideal batch norm look like on 25 dojo dies? Let's say the previous less activations are already split across dies We would expect the 350 nodes on each die to coordinate and produce die local mean and standard deviation values Ideally these would get further reduced with the final value ending somewhere towards the middle of the tile We would then hope to see a broadcast of this value radiating from the center Let's see how the compiler actually executes a real batch norm operation across 25 dies The communication trees were extracted from the compiler and the timing is from a real hardware one We're about to see 8,750 nodes on 25 dies coordinating to reduce and then broadcast the batch norm mean and standard deviation values Die local reduction followed by global reduction towards the middle of the tile Then the reduced value broadcast radiating from the middle accelerated by the hardware's broadcast facility This operation takes only 5 microseconds on 25 dojo dies The same operation takes 150 microseconds on 24 GPUs This is an orders of magnitude improvement over GPUs And while we talked about an already used operation in the context of a batch norm It's important to reiterate that the same advantages apply to all other communication primitives And these primitives are essential for large scale training So how about full model performance? So while we think that ResNet 50 is not a good representation of real world Tesla workloads It is a standard benchmark, so let's start there We are already able to match the 100 die for die However, perhaps a hint of dojo's capabilities is that we're able to hit this number with just a batch of 8 per die But dojo was really built to tackle larger complex models So when we set out to tackle real world workloads, we looked at the usage patterns of our current GPU cluster And two models stood out, the autolabeling networks, a class of offline models that are used to generate ground truth And the occupancy networks that you heard about The autolabeling networks are large models that have high arithmetic intensity While the occupancy networks can be ingest bound We chose these models because together they account for a large chunk of our current GPU cluster usage And they would challenge the system in different ways So how do we do on these two networks? The results we're about to see were measured on multi die systems for both the GPU and dojo, but normalized to per die numbers On our autolabeling network, we're already able to surpass the performance of an A100 With our current hardware running on our older generation VRMs On our production hardware with our newer VRMs, that translates to doubling the throughput of an A100 And our model showed that with some key compiler optimizations, we could get to more than 3x the performance of an A100 We see even bigger leaps on the occupancy network Almost 3x with our production hardware, with room for more So what does that mean for Tesla? With a current level of compiler performance, we could replace the ML compute of 1, 2, 3, 4, 5 and 6 GPU boxes with just a single dojo tile And this dojo tile costs less than one of these GPU boxes What it really means is that networks that took more than a month to train now take less than a week Alas, when we measure things, it did not turn out so well.\n\nAt the PyTorch level, we did not see our expected performance out of the gate And this timeline chart shows our problem. The teeny, tiny little green bars, that's the compile code running on the accelerator The row is mostly white space where the hardware is just waiting for data With our dense ML compute, dojo hosts effectively have 10x more ML compute than the GPU hosts. The data loader is running on this one host Simply couldn't keep up with all that ML hardware So to solve our data loader scalability issues, we knew we had to get over the limit of this single host The Tesla transport protocol moves data seamlessly across hosts, tiles and ingest processors So we extended the Tesla transport protocol to work over Ethernet. We then built the dojo network interface card, the D-NIC, to leverage TTP over Ethernet This allows any host with a D-NIC card to be able to DMA2 and from other TTP endpoints So we started with the dojo mesh, then we added a tier of data loading hosts equipped with the D-NIC card We connected these hosts to the mesh via an Ethernet switch. Now every host in this data loading tier is capable of reaching all TTP endpoints in the dojo mesh via hardware accelerated DMA After these optimizations went in, our occupancy went from 4% to 97% So the data loading sections have reduced drastically and the ML hardware has kept busy We actually expect this number to go to 100% pretty soon After these changes went in, we saw the full expected speed up from the PyTorch layer and we were back in business So we started with hardware design that breaks through traditional integration boundaries in service of our vision of a single giant accelerator We've seen how the compiler and ingest layers build on top of that hardware So after proving our performance on these complex real-world networks, we knew what our first large-scale deployment would target Our high arithmetic intensity auto-labeling networks Today that occupies 4,000 GPUs over 72 GPU racks With our dense computer and our high performance, we expect to provide the same throughput with just 4 dojo cabinets And these 4 dojo cabinets will be part of our first exapod that we plan to build by quarter one of 2023 This will more than double Tesla's auto-labeling capacity The first exapod is part of a total of 7 exapods that we plan to build in Palo Alto right here across the wall And we have a display cabinet from one of these exapods for everyone to look at 6 tiles densely packed on a tray, 54 petaflops of compute, 640 gigabytes of high bandwidth memory with power and host defeated A lot of compute And we're building out new versions of all our cluster components and constantly improving our software to hit new limits of scale We believe that we can get another 10x improvement with our next generation hardware And to realize our ambitious goals, we need the best software and hardware engineers So please come talk to us or visit tesla.com. Alright, so hopefully that was enough detail And now we can move to questions And guys, I think the team can come out on stage We really wanted to show the depth and breadth of Tesla in artificial intelligence, compute hardware, robotics actuators And try to really shift the perception of the company away from, you know, a lot of people think we're like just a car company Or we make cool cars, whatever But most people have no idea that Tesla is arguably the leader in real world AI hardware and software And that we're building what is arguably the most radical computer architecture since the Kray-1 supercomputer And I think if you're interested in developing some of the most advanced technology in the world that's going to really affect the world in a positive way Tesla's the place to be So yeah, let's fire away with some questions I think there's a mic at the front and a mic at the back Just throw mics at people Jump all for the mic Yeah, hi, thank you very much I was impressed here I was impressed very much by Optimus, but I wonder why did not driven the hand Why did you choose a tendon-driven approach for the hand?\n\nBecause tendons are not very durable And why spring-loaded? Cool, awesome, yes, that's a great question You know, when it comes to any type of actuation scheme, there's trade-offs between, you know, whether or not it's a tendon-driven system or some type of linkage-based system Keep the mic close to your mouth A little bit closer, hear me? Cool Yeah, the main reason why we went for a tendon-based system is that, you know, first we actually investigated some synthetic tendons, but we found that metallic boating cables are, you know, a lot stronger One of the advantages of these cables is that it's very good for part reduction We do want to make a lot of these hands, so having a bunch of parts, a bunch of small linkages ends up being, you know, a problem when you're making a lot of something One of the big reasons that, you know, tendons are better than linkages in a sense is that you can be anti-backlash So anti-backlash essentially, you know, allows you to not have any gaps or, you know, stuttering motion in your fingers Spring-loaded, mainly what spring-loaded allows us to do is allows us to have active opening So instead of having to have two actuators to drive the fingers closed and then open, we have the ability to, you know, have the tendon drive them closed and then the springs passively extend And this is something that's seen in our hands as well, right? We have the ability to actively flex and then we also have the ability to extend Yeah I mean, our goal with Optimus is to have a robot that is maximally useful as quickly as possible So there's a lot of ways to solve the various problems of a humanoid robot And we're probably not barking up the right tree on all the technical solutions And I should say that we're open to evolving the technical solutions that you see here over time, they're not locked in stone But we have to pick something, and we want to pick something that's going to allow us to produce the robot as quickly as possible and have it, like I said, be useful as quickly as possible We're trying to follow the goal of fastest path to a useful robot that can be made at volume And we're going to test the robot internally at Tesla in our factory and just see, like, how useful is it Because you have to have a, you've got to close the loop on reality to confirm that the robot is in fact useful And, yeah, so we're just going to use it to build things And we're confident we can do that with the hand that we have currently designed But I'm sure there'll be hand version 2, version 3, and we may change the architecture quite significantly over time Hi, the Optimus robot is really impressive, you did a great job, bipedal robots are really difficult But what I noticed might be missing from your plan is to acknowledge the utility of the human spirit And I'm wondering if Optimus will ever get a personality and be able to laugh at our jokes while it folds our clothes Yeah, absolutely. I think we want to have really fun versions of Optimus And so that Optimus can both be utilitarian and do tasks, but can also be kind of like a friend and a buddy And hang out with you, and I'm sure people will think of all sorts of creative uses for this robot And, you know, the thing, once you have the core intelligence and actuators figured out Then you can actually, you know, put all sorts of costumes, I guess, on the robot I mean, you can make the robot look, you can skin the robot in many different ways And I'm sure people will find very interesting ways to, yeah, versions of Optimus Thanks for the great presentation I wanted to know if there was an equivalent to interventions in Optimus It seems like labeling through moments where humans disagree with what's going on is important And in a humanoid robot, that might be also a desirable source of information Yeah, I think we will have ways to remote operate the robot and intervene when it does something bad Especially when we are training the robot and bringing it up And hopefully we, you know, design it in a way that we can stop the robot from, if it's going to hit something We can just, like, hold it and it will stop, it won't, like, you know, crush your hand or something And those are all intervention data Yeah, and we can learn a lot from our simulation systems, too Where we can check for collisions and supervise that those are bad actions Yeah, I mean, so Optimus, we went over time for it to be, you know, an android, the kind of android that you've seen in sci-fi movies Like Star Trek, The Next Generation, like data But obviously we could program the robot to be less robot-like and more friendly And, you know, you can obviously learn to emulate humans and feel very natural So as AI in general improves, we can add that to the robot And, you know, it should be obviously able to do simple instructions or even intuit what it is that you want So you could give it a high level instruction and then it can break that down into a series of actions And take those actions Hi, yeah, it's exciting to think that with the Optimus you will think that you can achieve orders of magnitude of improvement in economic output That's really exciting And when Tesla started, the mission was to accelerate the advent of renewable energy or sustainable transport So with the Optimus, do you still see that mission being the mission statement of Tesla or is it going to be updated with, you know, mission to accelerate the advent of, I don't know, infinite abundance or limitless economy Yeah, it is not strictly speaking, Optimus is not strictly speaking directly in line with accelerating sustainable energy To the degree that it is more efficient at getting things done than a person, it does, I guess, help with sustainable energy But I think the mission effectively does somewhat broaden with the advent of Optimus to, you know, I don't know, making the future awesome So, you know, I think you look at Optimus and I know about you, but I'm excited to see what Optimus will become And, you know, this is like, you know, if you could, I mean, you can tell like any given technology, do you want to see what it's like in a year, two years, three years, four years, five years, ten? I'd say for sure, you definitely want to see what's happened with Optimus Whereas, you know, a bunch of other technologies are, you know, sort of plateaued About name names here, but, you know, so, I think Optimus is going to be incredible in like five years, ten years like mind-blowing And I'm really interested to see that happen, and I hope you are too I have a quick question here, Justin, and I was wondering, like, are you planning to extend like conversational capabilities for the robot?\n\nAnd my second full-on question to that is, what's like the end goal? What's the end goal with Optimus? Yeah, Optimus would definitely have conversational capabilities So, you'd be able to talk to it and have a conversation, and it would feel quite natural So, from an end goal standpoint, I don't know, I think it's going to keep evolving, and I'm not sure where it ends up, but some place is interesting for sure And, you know, we always have to be careful about the, you know, don't go down the terminator path That's a, you know, I thought maybe we should start off with a video of like the terminator starting off with this, you know, skull crushing But that might be, you know, people might not get too seriously So, you know, we do want Optimus to be safe, so we are designing in safeguards where you can locally stop the robot And, you know, with like basically a localized control ROM that you can't update over the internet Which I think that's quite important, essential, frankly So, like a localized stop button or remote control, something like that, that cannot be changed But, I mean, it's definitely going to be interesting, it won't be boring Okay, yeah, I see today you have a very attractive product with Dojo and its applications So, I'm wondering what's the future for the Dojo platform? So, you know, like provide like infrastructure and service like AWS or you will like sell the chip like the NVIDIA So, basically, what's the future? Because I say you use 7nm, so the developer cost is like easily over 10 million US dollars How do you make the business like business wise? Dojo is a very big computer and actually will use a lot of power and need a lot of cooling So, I think it's probably going to make more sense to have Dojo operate in like an Amazon Web Services manner Than to try to sell it to someone else So, that would be the most efficient way to operate Dojo is just have it be a service that you can use That's available online and that where you can train your models way faster and for less money And as the world transitions to software 2.0 And that's on the bingo card Someone I know has to know to drink 5 tequila So, let's see, software 2.0 will use a lot of neural net training So, it kind of makes sense that over time as there's more neural net stuff People will want to use the fastest, lowest cost neural net training system So, I think there's a lot of opportunity in that direction Hi, my name is Ali Jahanian Thank you for this event, it's very inspirational My question is, I'm wondering what is your vision for humanoid robots that understand our emotions and art And can contribute to our creativity Well, I think you're already seeing robots that at least are able to generate very interesting art Like Dali and Dali 2 And I think we'll start seeing AI that can actually generate even movies that have coherence Like interesting movies and tell jokes So, it's quite remarkable how fast AI is advancing at many companies besides Tesla We're headed for a very interesting future And yeah, so, any guys want to comment on that?\n\nYeah, I guess the Optimus Robot can come up with physical art, not just digital art You can ask for some dance moves in text or voice and then you can produce those in the future So, it's a lot of physical art, not just digital art Oh, yeah, computers can absolutely make physical art, yeah, 100% Yeah, like dance, play soccer or whatever you... I mean, it needs to get more agile over time, for sure Thanks so much for the presentation Now, for the Tesla Autopilot slides, I noticed that the models that you were using were heavily motivated by language models And I was wondering what the history of that was and how much of an improvement it gave I thought that that was a really interesting, curious choice to use language models for the lane transitioning So, there are sort of two aspects for why we transition to language modeling So, the first... Talk loud and close Okay, got it Yeah, so the language models help us in two ways The first way is that it lets us predict lanes that we couldn't have otherwise As Ashok mentioned earlier, basically when we predicted lanes in sort of a dense 3D fashion You can only model certain kinds of lanes, but we want to get those criss-crossing connections inside of intersections It's just not possible to do that without making it a graph prediction If you try to do this with dense segmentation, it just doesn't work Also, the lane prediction is a multimodal problem Sometimes you just don't have sufficient visual information to know precisely how things look on the other side of the intersection So you need a method that can generalize and produce coherent predictions You don't want to be predicting two lanes and three lanes at the same time You want to commit to one in a general model like these language models provides that Hi Hi, my name is Giovanni Yeah, thanks for the presentation. It's really nice I have a question for FSD team For the neural networks, how do you test... How do you do unit tests, software unit tests on that? Do you have a bunch or I don't know, mid-thousands or...\n\nYes, cases where the neural network that after you train it, you have to pass it Before you release it as a product, right? Yeah, what's your software unit testing strategies for this, basically? Yeah, glad you asked. There's like a series of tests that we have defined starting from unit tests for software itself But then for the neural network models, we have VAP sets defined where you can define... If you just have a large test set, that's not enough what we find We need like sophisticated VAP sets for different failure modes And then we queate them and grow them over the time of the product So over the years, we have like hundreds of thousands of examples where we have been failing in the past That we have curated and so for any new model, we test against the entire history of these failures And then keep adding to this test set On top of this, we have shadow modes where we ship these models in silent to the car And we get data back on where they are failing or succeeding And there's an extensive QA program It's very hard to ship for regression There's like nine levels of filters before it hits customers But then we have really good infra to make this all efficient I'm one of the QA testers, so I have QA the car... Yeah, QA tester Yeah, so I'm constantly in the car just being queuing like whatever the latest alpha build is that doesn't totally crash Yeah, finds a lot of bugs Hi, great event.\n\nI have a question about foundational models for autonomous driving We have all seen that big models that really can... When you scale up with data and model parameter from GP3 to POM, it can actually now do reasoning Do you see that it's essential scaling up foundational models with data and size And then at least you can get a teacher model that potentially can solve all the problems And then you distill to a student model Is that how you see foundational models relevant for autonomous driving? That's quite similar to our auto labeling models So we don't just have models that run in the car We train models that are entirely offline that are extremely large that can't run in real time on the car So we just run those offline on the servers producing really good labels that can then train the online networks So that's one form of distillation of these teacher-student models In terms of foundation models, we are building some really, really large datasets that are multiple petabytes And we are seeing that some of these tasks work really well when we have these large datasets Kinematics, like I mentioned, video in, all the kinematics out of all the objects and up to the fourth derivative And people thought we couldn't do detection with cameras Detection, depth, velocity, acceleration And imagine how precise these have to be for these higher-order derivatives to be accurate And this all comes from these kind of large datasets and large models So we are seeing the equivalent of foundation models in our own way for geometry and kinematics and things like those Do you want to add anything, John? Yeah, I'll keep it brief Basically, whenever we train on a larger dataset, we see big improvements in our model performance And basically, whenever we initialize our networks with some pre-training steps from some other auxiliary tasks We basically see improvements The self-supervised or supervised with large datasets both help a lot Hi, so at the beginning, Elon said that Tesla was potentially interested in building artificial general intelligence systems Given the potentially transformative impact of technology like that It seems prudent to invest in technical AGI safety expertise specifically I know Tesla does a lot of technical, narrow AI safety research I was curious if Tesla was intending to try to build expertise in technical artificial general intelligence safety specifically Well, I mean, if we start looking like we're going to be making a significant contribution to artificial general intelligence Then we'll for sure invest in safety on big believer in AI safety I think there should be an AI sort of regulatory authority at the government level Just as there is a regulatory authority for anything that affects public safety So we have regulatory authority for aircraft and cars and sort of food and drugs Because they affect public safety and AI also affects public safety So I think, and this is not really something that government I think understands yet I think there should be a referee that is trying to ensure public safety for AGI And you think of like, well, what are the elements that are necessary to create AGI? Like the accessible dataset is extremely important And if you've got a large number of cars and humanoid robots processing petabytes of video data and audio data from the real world Just like humans, that might be the biggest dataset, probably is the biggest dataset Because in addition to that, you can obviously incrementally scan the internet But what the internet can't quite do is have millions or hundreds of millions of cameras in the real world Like I said, with audio and other sensors as well So I think we probably will have the most amount of data And probably the most amount of training power Therefore probably we will make a contribution to AGI Hey, I noticed the semi was back there, but we haven't talked about it too much I was just wondering for the semi truck, what are the changes you're thinking about from a sensing perspective? I imagine there's very different requirements obviously than just a car And if you don't think that's true, why is that true?\n\nNo, I think basically you can drive a car Think about what drives any vehicle, it's a biological neural net with eyes With cameras essentially What is your primary sensors are? Two cameras on a slow gimbal, a very slow gimbal That's your head So if a biological neural net with two cameras on a slow gimbal can drive a semi truck Then if you've got like eight cameras with continuous 360 degree vision Operating at a higher frame rate and a much higher reaction rate Then I think it is obvious that you should be able to drive a semi or any vehicle much better than human Hi, my name is Akshay, thank you for the event Assuming Optimus would be used for different use cases and would evolve at different speeds for these use cases Would it be possible to sort of develop and deploy different software and hardware components independently And deploy them in Optimus so that the overall feature development is faster for Optimus Okay, we did not comprehend Unfortunately our neural net did not comprehend the question Next question Hi, I want to switch the gear to the autopilot So when you guys plan to roll out the FSD beta to countries other than US and Canada And also my next question is what's the biggest bottleneck or the technology or barrier you think in the current autopilot stack And how you envision to solve that to make the autopilot is considerably better than human in terms of performance matrix Like safety assurance and the human confidence I think you also mentioned for the FSD V11 you are going to combine the highway and the city as a single stack And some architectural big improvements, can you maybe expand a bit on that, thank you Well, that's a whole bunch of questions We're hopeful to be able to, I think from a technical standpoint FSD beta should be possible to roll out FSD beta worldwide by the end of this year But for a lot of countries we need regulatory approval And so we are somewhat gated by the regulatory approval in other countries But I think from a technical standpoint it will be ready to go to a worldwide beta by the end of this year And there's quite a big improvement that we're expecting to release next month That will always be especially good at assessing the velocity of fast moving cross traffic And a bunch of other things So, anyone want to elaborate? I guess so, there used to be a lot of differences between production autopilot and the full self driving beta But those differences have been getting smaller and smaller over time I think just a few months ago we now use the same vision only object detection stack in both FSD and in the production autopilot on all vehicles There's still a few differences, the primary one being the way that we predict lanes right now So we upgraded the modeling of lanes so that it could handle these more complex geometries like I mentioned in the talk In production autopilot we still use a simpler lane model But we're extending our current FSD beta models to work in all sort of highway scenarios as well The version of FSD beta that I drive actually does have the integrated stack So it uses the FSD stack both in city streets and highway and it works quite well for me But we need to validate it in all kinds of weather like heavy rain, snow, dust And just make sure it's working better than the production stack across a wide range of environments But we're pretty close to that I think it's, I don't know, maybe, it'll definitely be before the end of the year and maybe November Yeah, in our personal drives, the FSD stack on highway drives already way better than the production stack we have And we do expect to also include the parking lot stack as a part of the FSD stack before the end of this year So that will basically bring us to, you sit in the car in the parking lot and drive till the end of the parking lot at a parking spot before the end of this year And in terms of the fundamental metric to optimize against is how many miles between a necessary intervention So just massively improving how many miles the car can drive in full autonomy before an intervention is required that is safety critical So, yeah, that's the fundamental metric that we're measuring every week and we're making radical improvements on that Hi, thank you, thank you so much for the presentation, very inspiring My name is Daisy, I actually have a non-technical question for you I'm curious, if you are back to your 20s, what are some of the things you wish you knew back then? What are some advice you would give to your younger self? Well, I'm trying to figure out something useful to say Yeah, a joint Tesla would be one thing Yeah, I think just trying to expose yourself to as many smart people as possible I don't read a lot of books You know, I did do that though So, I think there's some merit to just also not being necessarily too intense And enjoying the moment a bit more, I would say to 20-something me Just to stop and smell the roses occasionally would probably be a good idea You know, it's like when we were developing the Falcon 1 rocket on the Quageline Atoll And we had this beautiful little island that we were developing the rocket on And not once during that entire time did I even have a drink on the beach I'm like, I should have had a drink on the beach, that would have been fine Thank you very much I think you have excited all of the robotics people with Optimus This feels very much like 10 years ago in driving But as driving has proved to be harder than it actually looked 10 years ago What do we know now that we didn't 10 years ago that would make, for example, AGI on a humanoid come faster? Well, I mean, it seems to me that AGI is advancing very quickly Hardly a week goes by without some significant announcement And, yeah, I mean, at this point, like, AI seems to be able to win at almost any rule-based game It's able to create extremely impressive art Engage in conversations that are very sophisticated, you know, write essays And these just keep improving And there's so many more talented people working on AI And the hardware is getting better AI is on a super, like, a strong exponential curve of improvements Independent of what we do at Tesla And obviously we'll benefit somewhat from that exponential curve of improvement with AI Like, Tesla just also has to be very good at actuators Motors gearboxes, controllers, power electronics, batteries, sensors And, you know, really, like, I'd say the biggest difference between the robot on four wheels And the robot with arms and legs is getting the actuators right It's an actuators and sensors problem And obviously, how you control those actuators and sensors But it's, yeah, actuators and sensors and how you control the actuators I don't know, we have to have, like, the ingredients necessary to create a compelling robot And we're doing it, so...\n\nHi, Ilan You are actually bringing the humanity to the next level Literally, Tesla, and you are bringing the humanity to the next level So, you said Optimus Prime, Optimus will be used in next Tesla factory My question is, will a new Tesla factory be fully run by Optimus program? And when can general public order a humanoid? Yeah, I think it'll, you know, we're going to start Optimus with very simple tasks in the factory You know, like maybe just, like, loading a part, like you saw in the video You know, carrying a part from one place to another Or loading a part into one of our more conventional robot cells to, you know, that welds body together So we'll start, you know, just trying to, how do we make it useful at all? And then gradually expand the number of situations where it's useful And I think that number of situations where Optimus is useful will grow exponentially Like really, really fast In terms of when people can order one, I don't know, I think it's not that far away Well, I think you mean, when can people receive one? So, I don't know, I'm like, I'd say probably within three years And not more than five years Within three to five years, you could probably receive an Optimus I feel the best way to make the progress for AGI is to involve as many smart people across the world as possible And given the size and resource of Tesla compared to robot companies And given the state of humanoid research at the moment Would it make sense for the kind of Tesla to sort of open source some of the simulation hardware parts? I think Tesla can still be the dominant platformer where it can be something like an Android OS Or like an iOS stuff for the entire humanoid research Would that be something that rather than keeping the Optimus to just Tesla researchers Or the factory itself can open it and let the whole world explore humanoid research?\n\nI think we have to be careful about Optimus being potentially used in ways that are bad Because that is one of the possible things to do So I think we would provide Optimus where you can provide instructions to Optimus But where those instructions are governed by some laws of robotics that you cannot overcome So not doing harm to others and I think probably quite a few safety related things with Optimus We'll just take maybe a few more questions and then thank you all for coming Questions, one deep and one broad On the deep for Optimus, what's the current and what's the ideal controller bandwidth? And then in the broader question, there's this big advertisement for the depth and breadth of the company What is it uniquely about Tesla that enables that? Anyone want to tackle the bandwidth question? So the technical bandwidth of the... Close to your mouth and loud For the bandwidth question, you have to understand or figure out what is the task that you want it to do And if you took a frequency transform of that task, what is it that you want your limbs to do? And that's where you get your bandwidth from It's not a number that you can specifically just say you need to understand your use case And that's where the bandwidth comes from What is the broad question?\n\nThe breadth and depth thing, I can answer the breadth and depth On the bandwidth question, I think we probably will just end up increasing the bandwidth Which translates to the effective dexterity and reaction time of the robot It's safe to say it's not one hertz and maybe you don't need to go all the way to 100 hertz But maybe 10, 25, I don't know Over time, I think the bandwidth will increase quite a bit Or translate it to dexterity and latency You'd want to minimize that over time Minimize latency, maximize dexterity In terms of breadth and depth, I guess we're a pretty big company at this point So we've got a lot of different areas of expertise that we necessarily had to develop In order to make electric cars and then in order to make autonomous electric cars Tesla is like a whole series of startups basically And so far they've almost all been quite successful So we must be doing something right And I consider one of my core responsibilities in running the company Is to have an environment where great engineers can flourish And I think in a lot of companies, I don't know, maybe most companies If somebody's a really talented driven engineer, they're unable to actually Their talents are suppressed at a lot of companies And some of the companies that the engineering talent is suppressed In a way that is maybe not obviously bad But where it's just so comfortable and you paid so much money The output you actually have to produce is so low that it's like a honey trap So there's a few honey trap places in Silicon Valley Where they don't necessarily don't seem like bad places for engineers But you have to say like a good engineer went in and what did they get out And the output of that engineering talent seems very low Even though there seem to be enjoying themselves That's why I call it there's a few honey trap companies in Silicon Valley Tesla is not a honey trap that we're demanding and it's like You're going to get a lot of shit done and it's going to be really cool And it's not going to be easy But if you are a super talented engineer Your talents will be used I think to a greater degree than anywhere else You know, SpaceX also that way Hi Ilan, I have two questions So both to the autopilot team So the thing is like I have been following your progress for the past few years So today you have made changes on like the lane detection Like you said that previously you were doing instant semantic segmentation Now you guys are built transfer models for like building the lanes So what are some other common challenges which you guys are facing right now Like which you are solving in future as a curious engineer So that like we as a researcher can work on those Start working on those And the second question is like I'm really curious about the data engine Like you guys have like told a case like where the car is stopped So how are you finding cases which is very much similar to that from the data which you have So a little bit more on the data engine would be great I'll answer the first question using occupancy network as an example So what you saw in the presentation did not exist a year ago So we only spent one year on time We actually shipped more than 12 occupancy network And to have a one foundation model actually to represent the entire physical world Around everywhere and you always condition is actually really really challenging So only over a year ago we're kind of like driving a 2D world If there's a wall and if there's a curve we kind of represent with the same static edge Which is obviously you know not ideal right There's a big difference between a curve and a wall when you drive you make different choices right So after we realized that we have to go to 3D We have to basically rethink the entire problem and think about how we address that So this will be like one example of a challenges we have we have a conquer in the past year Yeah to answer the question about how we actually source examples of the tricky stopped cars There's a few ways to go about this but two examples are one we can trigger for disagreements within our signals So let's say that parked bit flickers between parked and driving We'll trigger that back and the second is we can leverage more of the shadow mode logic So if the customer ignores the car but we think we should stop for it we'll get that data back too So these are just different like various trigger logic that allows us to get those data campaigns back Hi Thank you for the amazing presentation thanks so much So there are a lot of companies that are focusing on the AGI problem And one of the reasons why it's such a hard problem is because the problem itself is so hard to define Several companies have several different definitions they focus on different things So what is Tesla how's Tesla defining the AGI problem and what are you focusing on specifically Well we're not actually specifically focused on AGI I'm simply saying that AGI is seems likely to be an emergent property of what we're doing Because we're creating the oldies autonomous cars and autonomous humanoids That are actually with a truly gigantic data stream that's coming in and being processed It's by far the most amount of real world data and data you can't get by just searching the internet Because you have to be out there in the world and interacting with people and interacting with the roads And just you know it's Earth is a big place and reality is messy and complicated So I think it's sort of like it just seems likely to be an emergent property If you've got tens or hundreds of millions of autonomous vehicles and maybe even a comparable number of humanoids Maybe more than that on the humanoid front Well that's just the most amount of data and if that video is being processed It just seems likely that the cars will definitely get way better than human drivers And the humanoid robots will become increasingly indistinguishable from humans perhaps And so then like I said you have this emergent property of AGI And arguably humans collectively are sort of a superintelligence as well Especially as we improve the data rate between humans The thing like that seems way back in the early days the internet was like the internet was like humanity acquiring a nervous system Where now all of a sudden any one element of humanity could know all of the knowledge of humans by connecting to the internet Almost all the knowledge or certainly a huge part of it Whereas previously we would exchange information by osmosis Like in order to transfer data so you would have to write a letter Someone would have to carry the letter by person to another person And then a whole bunch of things in between and then it was like Yeah I mean it's insanely slow when you think about it And even if you were in the Library of Congress you still didn't have access to all the world's information And you certainly couldn't search it and obviously very few people are in the Library of Congress So I mean one of the great sort of equality elements Like the internet has been the biggest equalizer in history in terms of access to information and knowledge And any student of history I think would agree with this Because you know you go back a thousand years there were very few books And books would be incredibly expensive but only a few people knew how to read And even a small number of people even had a book Now look at it like you can access any book instantly You can learn anything basically for free It's pretty incredible So you know I was asked recently what period of history would I prefer to be at the most And my answer was right now This is the most interesting time in history and I read a lot of history So let's do our best to keep that going And to go back to one of the earlier questions I would ask The thing that's happened over time with respect to Tesla autopilot is that the neural nets have gradually absorbed more and more software And in the limit of course you could simply take the videos as seen by the car And compare those to the steering inputs from the steering wheel and pedals Which are very simple inputs And in principle you could train with nothing in between Because that's what humans are doing with the biological neural net You could train based on video and what trains the video is the moving of the steering wheel and the pedals With no other software in between We're not there yet but it's gradually going in that direction Alright, one last question How are you going? I think we've got a question at the front here Hello, they're right there We'll do two questions, fine They're here Thanks for such a great presentation We'll do your question last Okay, cool With FSD being used by so many people How do you evaluate the company's risk tolerance in terms of performance statistics And do you think there needs to be more transparency or regulation from third parties As to what's good enough and defining thresholds for performance across many miles The number one design requirement at Tesla is safety And that goes across the board So in terms of the mechanical safety of the car We have the lowest probability of injury of any cars ever tested by the government For just a passive mechanical safety Essentially crash structure and airbags and what not We have the highest rating for active safety as well And I think it's going to get to the point where the active safety is so ridiculously good It's just absurdly better than a human And then with respect to autopilot We do publish broadly speaking the statistics on miles driven With cars that have no autonomy Tesla cars with no autonomy With hardware one, hardware two, hardware three And then the ones that are in FSD beta And we see steady improvements all along the way And sometimes there's this dichotomy of Should you wait until the car is three times safer than a person before deploying any technology But I think that's actually morally wrong At the point at which you believe that adding autonomy reduces injury and death I think you have a moral obligation to deploy it Even though you're going to get sued and blamed by a lot of people Because the people whose lives you saved don't know that their lives are saved And the people who do occasionally die or get injured Definitely know, or their state does, that there was a problem with autopilot That's why you have to look at the numbers in total miles driven How many accidents occurred, how many accidents were serious, how many fatalities And we've got well over three million cars on the road So that's a lot of miles driven every day And it's not going to be perfect But what matters is that it is very clearly safer than not deploying it Yeah So, I think, last question I think, yeah, thanks The last question here Okay, hi So, I do not work on hardware So maybe the hardware team and you guys can enlighten me Why is it required that there be symmetry in the design of Optimus? Because humans, we have handedness, right? We use some set of muscles more than others Over time there's wear and tear, right? So maybe you'll start to see some joint failures or some actuator failures more Over time, I understand that this is extremely pre-stage Also, we as humans have based so much fantasy and fiction Over superhuman capabilities Like all of us don't want to walk right over there We want to extend our arms and like we have all these, you know A lot of fantasy, fantastical designs So considering everything else that is going on In terms of batteries and intensity of compute Maybe you can leverage all those aspects into coming up with something Well, I don't know, more interesting in terms of the robot that you're building And I'm hoping you're able to explore those directions Yeah, I think it would be cool to have like, you know, make Inspector Gadget real That would be pretty sweet So, yeah, I mean, right now we just want to make basic humanoid work well And our goal is to pass this path to a useful humanoid robot I think this will ground us in reality, literally And ensure that we are doing something useful Like one of the hardest things to do is to be useful To actually, and then to have high utility under the curve Like how much help did you provide to each person on average And then how many people did you help? The total utility Like trying to actually ship useful product that people like To a large number of people is so insanely hard It boggles the mind You know, that's why I can say like, man, there's a hell of a difference between a company that has shipped product And one has not shipped product This is night and day And then even once you ship product, can you make the cost, the value of the output Worth more than the cost of the input Which is, again, insanely difficult, especially with hardware So, but I think over time I think it would be cool to do creative things And have like eight arms and whatever And have different versions And maybe, you know, there'll be some hardware Like companies that are able to add things to an optimist Like maybe we, you know, add a power port or something like that Or attach them, you can add attachments to your optimist Like you can add them to your phone There could be a lot of cool things that could be done over time And there could be maybe an ecosystem of small companies that, or big companies that Make add-ons for optimists So, with that, I'd like to thank the team for their hard work You guys are awesome And thank you all for coming And for everyone online, thanks for tuning in And I think this will be one of those great videos where you can like If you can fast forward to the bits that you find most interesting But we try to give you a tremendous amount of detail Literally so that you can look at the video at your leisure And you can focus on the parts that you find interesting and skip the other parts So, thank you all, and we'll do this, try to do this every year And we might do a monthly podcast even So, but I think it'll be great to sort of bring you along for the ride And like show you what cool things are happening And yeah, thank you Alright, thanks Thank you","textByLang":{"en":"All right, welcome everybody give everyone a moment to Get back in the audience and All right great welcome to Tesla AI day 2022 We've got some really exciting things to show you I think you'll be pretty impressed I do want to set some expectations with respect to our Optimus robot as As you know last year was just a person in a robot suit But we've now we've come a long way and that's I think we you know compared to that it's gonna be very impressive and We're gonna talk about The advancements in AI for full self-driving as well as how they apply to more generally to real-world AI problems Like a humanoid robot and and even going beyond that I think there's some potential that what we're doing here at Tesla could make a meaningful contribution to AGI and And I think actually Tesla is a good Antity to do it from a governance standpoint because we're a publicly traded company with one class of stock and That means that the public controls Tesla and I think that's actually a good thing So if I go crazy you can fire me. This is important Maybe I'm not crazy. I don't know so Yeah, so we're gonna talk a lot about our progress in AI autopilot as well as progress in with dojo and Then we're gonna bring the team out and to do a long Q&A so you can ask tough questions But whatever you'd like existential questions technical questions, but we want to have As much time for Q&A as possible. So let's see you with that That's because Hey guys, I'm Milana work on autopilot and it is about and I'm Lizzie Mechanical engineer on the project as well. Okay So should we should we bring up the bot before we do that? We have one One little bonus tip for the day.\n\nThis is actually the first time we try this robot without any backup support Cranes mechanical mechanisms. No cables. Nothing. Yeah I want to do it with you guys tonight. That is the first time. Let's see.\n\nYou ready? Let's go I I I think the bug got some boobs This is essentially the simple self-driving computer that runs in your Tesla cars by the way This is the this is literally the first time the robot has operated without a tether was on stage tonight So the robot can actually do a lot more than we just showed you we just didn't want it to fall on its face So we'll we'll show you some videos now of the robot doing a bunch of other things Yeah, which are less risky. Yeah, we should close the screen guys. Yeah Yeah, we wanted to show a little bit more what we've done over the past few months with the bot and just walking around and dancing on stage Just humble beginnings, but you can see the autopilot neural networks running as it's just retrained for the bot directly on that on that new platform That's my watering can yeah when you when you see a rendered view. That's that's the robot. What's the that's the world the robot sees So it's it's very clearly identifying objects like this is the object.\n\nIt should pick up picking it up Yeah We use the same process as we did for autopilot to connect data and train neural networks that we didn't deploy on the robot That's an example that illustrates the upper body a little bit more Something that will like try to nail down in a few months over the next few months, I would say To perfection, but this is really an actual station in the Fremont factory as well that it's working at And that's not the only thing we have to show today, right? Yeah, absolutely. So That what you saw was what we call bumble sea. That's our sort of rough development robot using semi off-the-shelf actuators But we actually have gone a step further than that already the team's done an incredible job And we actually have an optimist bot with fully tesla designed and built actuators Um battery pack control system everything. Um, it it wasn't quite ready to walk But I think it will walk in a few weeks But we wanted to show you the robot The something that's actually fairly close to what we'll go into production And and show you all all the things that can do so let's bring it up All right Yeah So here you're seeing optimists with these the With the degrees of freedom that we expect to have in optimists production unit one Which is the ability to move all the fingers independently move the To have the thumb have two degrees of freedom. So it has opposable thumbs And both left and right hand so it's able to operate tools and do useful things.\n\nOur goal is to make a useful humanoid robot as quickly as possible and We've also designed it using the same discipline that we use in designing the car, which is to say to design it for All manufacturing such that it's possible to make the robot at in high volume at low cost with high reliability So that that's incredibly important. I mean, you've all seen very impressive humanoid robot demonstrations And that that's great. But what are they missing? They're missing a brain that they don't have the the intelligence to navigate the world by themselves And they're they're also very expensive Um and made in low volume. Um, whereas, uh, this this is the optimist's design to be an extremely capable robot But made in in very high volume probably ultimately millions of units And it is expected to cost much less than a car So, uh, I would say probably less than 20,000 dollars would be my guess Okay The the potential for optimist is I think appreciated by very few people As usual Tesla demos are coming in hot So Um, yeah, uh, I'm the team's put on put in and the team has put in an incredible amount of work Uh working days, you know seven days a week Running the 3am oil To to get to the demonstration today. Um, super proud of what they've done is they've really done a great job I'd just like to give a hand to the whole optimist team So, you know that now there's still a lot of work to be done to, uh refine optimists and Improve it.\n\nObviously, this is just optimist version one And that's really why we're holding this event Which is to convince some of the most talented people in the world like you guys um to Join tesla and help make it a reality and bring it to fruition at scale Such that it can help millions of people And the and the potential likes it is is really boggles the mind because you have to say like what what is an economy an economy is uh sort of productive entities times the productivity, uh capita times Productivity per capita at the point at which there is not a limitation on capita The it's not clear what an economy even means at that point. It an economy becomes quasi infinite um so What what you know taken to fruition in the hopefully benign scenario the This means a future of abundance a future where um There is no poverty where people you can have whatever you want In terms of products and services It really is a a fundamental transformation of civilization as we know it Obviously we want to make sure that transformation is a positive one and um safe And but but that's also why I think Tesla as an entity doing this being a single class of stock publicly traded owned by the public um is very important Um and should not be overlooked. I think this is essential because then if the public doesn't like what tesla's doing The public can buy shares in tesla and vote differently This is a big deal Um Like it's very important that that I can't just do what I want You know, sometimes people think that that but it's not true. Um, so You know that it's it's very important that the the corporate entity that has that that makes this happen Is something that the public can properly influence And so I think the tesla structure is is is ideal for that Um And like said that you know self-driving cars will certainly have a Tremendous impact on the world. Um, I think they will improve the productivity of transport by at least A half order of magnitude perhaps an order of magnitude perhaps more um Optimus I think has Maybe a two order of magnitude Uh potential improvement in uh economic output Like like it's it's not clear. It's not clear what the limit actually even is um so But we we need to do this in the right way we need to do it carefully and safely And ensure that the outcome is one that is beneficial to Uh civilization and and one that humanity wants Uh, I can't this is also extremely important obviously so, um And and I hope you will consider uh joining tesla to achieve those goals Um It tesla we're we really care about doing the right thing here or aspire to do the right thing and and really not Pay the road to hell with with good intentions And I think the road is road to hell is mostly paved with bad intentions, but every now and again There's a good intention in there.\n\nSo we want to do the right thing. Um, so, you know consider joining us and helping make it happen um With that let's let's uh, we want to the next phase All right, so you've seen a couple robots today. Let's do a quick timeline recap So last year we unveiled the tesla bot concept, but a concept doesn't get us very far We knew we needed a real development and integration platform to get real life learnings as quickly as possible So that robot that came out and did the little routine for you guys We had that within six months built working on software integration hardware upgrades over the months since then But in parallel we've also been designing the next generation this one over here So this guy is rooted in the the foundation of sort of the vehicle design process, you know We're leveraging all of those learnings that we already have Obviously there's a lot that's changed since last year, but there's a few things that are still the same you'll notice We still have this really detailed focus on the true human form We think that matters for a few reasons, but it's fun. We spend a lot of time thinking about how amazing the human body is We have this incredible range of motion typically really amazing strength Um a fun exercise is if you put your fingertip on the chair in front of you you'll notice that there's a huge Range of motion that you have in your shoulder and your elbow for example without moving your fingertip you can move those joints all over the place Um, but the robot, you know, its main function is to do real useful work And it maybe doesn't necessarily need all of those degrees of freedom right away So we've stripped it down to a minimum sort of 28 fundamental degrees of freedom and then of course our hands in addition to that Humans are also pretty efficient at some things and not so efficient in other times So for example, we can eat a small amount of food to sustain ourselves for several hours. That's great Uh, but when we're just kind of sitting around no offense, but we're kind of inefficient. We're just sort of burning energy So on the robot platform, what we're going to do is we're going to minimize that idle power consumption drop it as low as possible And that way we can just flip a switch and immediately the robot turns into something that does useful work So let's talk about this latest generation in some detail, shall we?\n\nSo on the screen here, you'll see in orange are actuators, which we'll get to in a little bit and in blue are electrical system So now that we have our sort of human based research and we have our first development platform We have both research and execution to draw from for this design Again, we're using that vehicle design foundation. So we're taking it from concept through design and analysis and then build and validation Along the way, we're going to optimize for things like cost and efficiency because those are critical metrics to take this product to scale eventually How are we going to do that? Well, we're going to reduce our part count and our power consumption of every element possible We're going to do things like reduce the sensing in the wiring at our extremities You can imagine a lot of mass in your hands and feet is going to be quite difficult and power consumptive to move around And we're going to centralize both our power distribution and our compute to the physical center of the platform So in the middle of our torso, actually it is the torso. We have our battery pack This is sized at 2.3 kilowatt hours, which is perfect for about a full day's worth of work What's really unique about this battery pack is it has all of the battery electronics integrated into a single pcb within the pack So that means everything from sensing to fusing Charge management and power distribution is all on one all in one place We're also leveraging both our vehicle products and our energy products To roll all of those key features into this battery. So that's streamlined manufacturing Really efficient and simple cooling methods battery management and also safety And of course we can leverage tesla's existing infrastructure and supply chain to make it So going on to sort of our brain it's not in the head, but it's pretty close Also in our torso, we have our central computer So as you know tesla already ships full self-driving computers in every vehicle we produce We want to leverage both the autopilot hardware and the software for the humanoid platform But because it's different in requirements and informed factor, we're going to change a few things first So we still are gonna it's going to do everything that a human brain does Processing vision data making split sescan decisions based on multiple sensory inputs and also communications So to support communications, it's equipped with wireless connectivity as well as audio support And then it also has hardware level security features which are important to protect both the robot and the people around the robot So now that we have our sort of core we're going to need some limbs on the sky Um, and we'd love to show you a little bit about our actuators and our fully functional hands as well But the first before we do that, I'd like to introduce Malcolm who's going to speak a little bit about our structural foundation for the robot Tesla have the capabilities to analyze highly complex systems Don't get much more complex than a crash You can see here a simulated crash from bottle three Superimposed on top of the actual physical crash It's actually incredible how um, how accurate it is Just to give you an idea of the complexity of this model It includes every not bolt and washer every spot weld and it has 35 million degrees of freedom quite amazing And it's true to say that if we didn't have models like this, we wouldn't be able to make the safest cars in the world So can we utilize our capabilities and our methods from the automotive side to influence a robot? Well, we can make a model and since we had crash software we're using the same software here We can make it fall down The purpose of this is to make sure that if it falls down ideally it doesn't but it's superficial damage We don't want it to for example break its gearbox and its arms.\n\nThat's equivalent of a dislocated shoulder of a robot Difficult and expensive to fix So we wanted to dust itself off get on with the job. It's being given We could also take the same model and we can drive the actuators using the inputs from a previously solved model Bringing it to life So this is producing the motions for the tasks we want the robot to do these tasks are picking up boxes turning squatting walking upstairs Whatever the set of tasks are we can play to the model. This is showing just simple walking We can create the stresses in all the components that helps us optimize the components These are not dancing robots these are actually the modal behavior the first five modes of the robot And typically when people make robots they make sure the first mode is up around the top single figures up towards 10 hertz Who is to do this is to make the controls of walking easier. It's very difficult to walk if you can't guarantee where your foot is wobbling around That's okay if you make one robot. We want to make thousands maybe millions We haven't got the luxury of making from carbon fiber titanium. We want to make them plastic things are not quite as stiff So we can't have these high targets.\n\nI call them dumb targets We've got to make them work at lower targets So is that it's that good to work? Well, if you think about it, sorry about this, but we're just bags of soggy jelly and bones thrown in We're not high frequency. If I start on my leg, I don't vibrate at 10 hertz We people operate at a lot of frequency. So we know the robot actually can it just makes controls harder So we take the information from this the modal Data and the stiffness and feed it into the control system that allows it to walk And Just changing tax lightly looking at the knee We could take some inspiration from biology and we can look to see what the mechanical advantage of the knee is It turns out it actually represent quite similar to four-bar link and that's quite non-linear That's not surprising really because if you think when you bend your leg down The torque on your knee is much more when it's bent than it is when it's straight So you'd expect a non-linear function and in fact the biology is non-linear. This matches it quite accurately So that's a representation the four-bar link is obviously not physically four-bar link as I said the characteristics are similar, but Me bending down that's not very scientific. Let's be a bit more scientific We've played all the tasks through the through this graph And this is showing picketing is walking squatting the tasks I said we did on the stress And that's the the torque Seen at the knee against the knee bend on the horizontal axis This is showing the requirement for the need to do all these tasks And then put a curve through it surfing over the top of the piece and that's saying this is what's required to make the robot Do these tasks?\n\nSo if we look at the four-bar link that's actually the green curve And it's saying that the non-linearity of the four-bar link is actually linearized The characteristic of the force what that really says is that's lower the force That's what makes the actuator have the lowest possible force, which is the most efficient. We want to burn energy up slowly What's the blue curve with the blue curve is actually if we didn't have a four-bar link We just had an arm sticking out of my leg here with a with an actuator on it a simple two-bar link That's the best we could do with a simple two-bar link and it shows that that would create a much more force in the actuator Which would not be efficient So what does it look like in practice? well As you'll see but it's very tightly packaged in the knee you'll see it go transparent on the second You'll see the four-bar link there is operating on the actuator. This is determined the force and the displacements on the actuator And now pass you over to Constantina to tell you a lot more detail about how these actuators are made and designed optimized. Thank you So I am I would like to talk to you about The design process and the actuator portfolio In our robot So there are many similarities between a car and the robot when it comes to powertrain design The most important thing that matters here is energy mass and cost We are carrying over most of our designing experience from the car to the robot So in the particular case you see a car with two drive units And the drive units are used in order to accelerate the car zero to 60 miles per hour time or drive a city drive site while The robot that has 28 actuators and It's not obvious. What are the tasks at actuator level?\n\nSo we have tasks that are higher level like walking or climbing stairs or carrying a heavy object Which need to be translated into joint Into joint specs therefore we use our model That generates The torque speed Trajectories for our joints which subsequently is going to be fed in our optimization model And to run through the optimization process This is one of the scenarios that the robot is capable of doing which is turning and walking So when we have this torque speed trajectory we lay it over an efficiency map of an actuator And we are able along the trajectory to generate The power consumption and the energy cumulative energy for the task versus time So this allows us to define the system cost for the particular actuator and put a simple point into the cloud Then we do this for hundreds of thousands of actuators by solving in our cluster And the red line denotes the Pareto front, which is the preferred area where we will look for optimal So the x denotes the preferred actuator design we have picked for this particular joint So now we need to do this for every joint. We have 28 joints to optimize and we parse our cloud We parse our cloud again for every joint spec and the red axis this time denotes the bespoke actuator designs for every joint The problem here Is that we have too many unique actuator designs and even if we take advantage of the symmetry Still there are too many in order to make something Mass manufacturable we need to be able to reduce the amount of unique actuator designs Therefore we run something called commonality study, which we parse our cloud again Looking this time for actuators that simultaneously meet the joint performance requirements for more than one joint at the same time So the resulting portfolio is six actuators and they show in a color map at the middle figure um And the actuators can be also viewed in this Slide we have three rotary and three linear actuators all of which have a great Output force or torque per mass The rotary actuator in particular has a mechanical clutch integrated On the high speed side angular contact ball bearing and on the high speed side And on the low speed side a cross roller bearing and the year train is a strain wave year Um, there are three integrated sensors here and the bespoke permanent magnet machine The linear actuator I'm sorry The linear actuator has planetary rollers and an inverted planetary Screw as a gear train which allows efficiency and compaction and durability So in order to demonstrate the force capability of our linear actuators, we have set up an experiment in order to test it under its limits And I will let you enjoy the video So our actuator is able to lift A half ton nine foot concert grand piano And This is a requirement it's not something nice to have Because our muscles can do the same when they are direct driven when they are directly driven our quadriceps muscles Can do the same thing it's just that the knee is an upgearing Linked system that converts the force into velocity at the end effector of our heels for purposes of giving To the human body agility So this is one of the main things that are amazing about the human body And I'm concluding my part at this point and I would like to welcome my colleague Mike who's going to talk to you about Hand design. Thank you very much Thanks for seeing us So we just saw how powerful a human and a humanoid actuator can be However, humans are also incredibly dexterous The human hand has the ability to move at 300 degrees per second There's tens of thousands of tactile sensors And it has the ability to grasp and manipulate almost every object in our daily lives For our robotic hand design, we were inspired by biology We have five fingers an opposable thumb Our fingers are driven by metallic tendons that are both flexible and strong We have the ability to complete wide aperture power grasps while also being optimized for precision gripping of small thin and delicate objects So why a human like robotic hand? Well, the main reason is that our factories in the world around us is designed to be ergonomic So what that means is that it ensures that objects in our factory are graspable But it also ensures that new objects that we may have never seen before can be grasped by the human hand And by our robotic hand as well The converse there is is pretty interesting because it's saying that these objects are designed to our hand instead of having to make changes To our hand to accompany a new object Some basic stats about our hand is that it has six actuators and 11 degrees of freedom It has an in-hand controller which drives the fingers and receives sensor feedback Sensor feedback is really important to learn a little bit more about the objects that we're grasping And also for proprioception and that's the ability for us to recognize where our hand is in space One of the important aspects of our hand is that it's adaptive This adaptability is involved essentially as complex mechanisms that allow the hand to adapt the objects that's being grasped Another important part is that we have a non-back drivable finger drive This clutching mechanism allows us to hold and transport objects without having to turn on the hand motors You just heard how we went about going we went about designing the tesla bot hardware Now I'll hand it off to Milan and our autonomy team to bring this robot to life Thanks Michael All right So all those cool things we've shown earlier in the video Were possible just in a matter of a few months. Thanks to the amazing work that we've done autopilot over the past few years Most of those components poured it quite easily over to the bot's environment If you think about it, we're just moving from a robot on wheels to a robot on legs So some of the components are pretty similar and some other require more heavy lifting So for example our computer vision neural networks Were ported directly from autopilot to the bot's situation It's exactly the same occupancy network that we'll talk into a little bit more details later with the autopilot team that is now running on the bot here in this video The only thing that changed really is the training data that we had to recollect We're also trying to find ways to improve those occupancy networks Using work made on your radiance fields to get really great volumetric Rendering of the bot's environments for example here some machinery that the bot might have to interact with Another interesting problem to think about is in indoor environments, mostly with that sense of gps signal How do you get the bot to navigate to its destination? Say for instance to find its nearest charging station So we've been training More neural networks to identify high-frequency features key points within the bot's camera streams And track them across frames over time as the bot navigates with its environment And we're using those points to get a better estimate of the bot's pose and trajectory within its environment as it's walking We also did quite some work on the simulation side and this is literally the autopilot simulator To which we've integrated the robots locomotion code and this is a video of the Motion control code running in the autopilot simulator Showing the evolution of the robot's work over time.\n\nSo as you can see we started quite slowly in April and start accelerating as we unlock more joints And deeper more advanced techniques like arms balancing over the past few months And so locomotion is specifically one component that's very different as we're moving from the car to the bot's environment And so I think it warrants a little bit more depth and I'd like my colleagues to start talking about this now Thank you Milan. Hi, everyone. I'm Felix. I'm a robotics engineer on the project and I'm going to talk about walking Walking seems easy, right? People do it every day. You don't even have to think about it But there are some aspects of walking which are challenging from engineering to technology And I think that's one of the things that makes it so much easier for me to think about it But there are some aspects of walking which are challenging from engineering perspective.\n\nFor example Physical self-awareness that means having a good representation of yourself What is the length of your limbs? What is the mass of your limbs? What is the size of your feet? All that matters Also having an energy efficient gate. You can imagine there's different styles of walking and all of them are equally efficient Most important keep balance. Don't fall And of course also coordinate the motion of all of your limbs together So now humans do all of this naturally But as engineers or roboticists we have to think about these problems And the following I'm going to show you how we address them in our locomotion planning and control stack So we start with locomotion planning And our representation of the bot that means a model of the robots kinematics dynamics and the contact properties And using that model and the desired path for the bots our locomotion planner generates reference trajectories for the entire system This means feasible trajectories with respect to the assumptions of our model The planner currently works in three stages.\n\nIt starts planning footsteps and ends with the entire motion photo system And let's dive a little bit deeper in how this works So in this video we see footsteps being planned over a planning horizon following the desired path And we start from this and add then Foot trajectories that connect these footsteps using toe-off and heel strike just as the humans Just as humans do and this gives us the largest right and less knee bend for high efficiency of the system The last stage is then finding a center of mass trajectory Which gives us a dynamically feasible motion of the entire system to keep balance As we all know plans are good, but we also have to realize them in reality. Let's say how see how we can do this Thank you Felix. Hello everyone. My name is Anand and I'm going to talk to you about controls So let's take the motion plan that Felix just talked about and put it in the real world on a real robot Let's see what happens It takes a couple steps and falls down Well, that's a little disappointing But we are missing a few key pieces here, which will make it walk Now as Felix mentioned the motion planner is using an idealized version of itself and a version of reality around it This is not exactly correct It also expresses its intention Through trajectories and wrenches wrenches of forces and torques that it wants to exert on the world to locomotive Reality is way more complex than any similar model. Also the robot is not simplified It's got vibrations and modes, compliance, sensor noise and on and on and on So what does that do to the real world when you put the bot in the real world? Well, the unexpected forces cause unmodeled dynamics, which essentially the planet doesn't know about and that causes destabilization Especially for a system that is dynamically stable like bipedal locomotion So what can we do about it?\n\nWell, we measure reality We use sensors and our understanding of the world to do state estimation And here you can see the attitude and pelvis pose, which is essentially the vestibular system in a human Along with the center of mass trajectory being tracked when the robot is walking in the office environment Now we have all the pieces we need in order to close the loop So we use our better bot model We use the understanding of reality that we've gained through state estimation And we compare what we want versus what we expect the reality expect that reality is doing to us in order to Add corrections to the behavior of the robot Here the robot certainly doesn't appreciate being poked, but it has an admirable job of staying upright The final point here is a robot that walks is not enough We need it to use its hands and arms to be useful. Let's talk about manipulation Hi everyone, my name is Eric robotics engineer on tesla bot And I want to talk about how we've made the robot manipulate things in the real world We wanted to manipulate objects while looking as natural as possible and also get there quickly So what we've done is we've broken this process down into two steps First is generating a library of natural motion references Or we could call them demonstrations and then we've adapted these motion references online to the current real world situation So let's say we have a human demonstration of picking up an object We can get a motion capture of that demonstration, which is visualized right here as A bunch of keyframes representing the location of the hands the elbows the torso We can map that to the robot using inverse kinematics And if we collect a lot of these now we have a library that we can work with But a single demonstration is not generalizable to the variation in the real world For instance, this would only work for a box in a very particular Location So what we've also done is run these Reference trajectories through a trajectory optimization program which solves for where the hand should be how the robot should balance during When it needs to adapt the motion to the real world. So for instance, if the box is In this location, then our optimizer will create this trajectory instead Next Milan's going to talk about uh, what's next for the optimist uh, tesla lie. Thanks Right, so hopefully by now you guys got a good idea of what we've been up to over the past few months Um, we started having something that's usable, but it's far from being useful. There's still a long and exciting road ahead of us um, I think the first thing within the next few weeks is to Get optimists at least apart with bumble see the other bug prototype you saw earlier and probably beyond We are also going to start focusing on the real use case at one of our factories and really going to try to try to Nail this down and I run out all the elements needed to deploy this product in the real world I was mentioning earlier, you know indoor navigation Um graceful for management or even servicing all components needed to scale this product up But um, I don't know about you, but after seeing what we've shown tonight I'm pretty sure we can get this done within the next few months or years Um, and and make this product a reality and change the entire economy Um, so I would like to thank the entire optimist team for their hard work over the past few months I think it's pretty amazing. All of this was done in barely six or eight months.\n\nThank you very much Hey everyone Hi, I'm Ashok. I lead the autopilot team alongside Milan Oh god, it's going to be so hard to top that optinist section He'll try nonetheless anyway Every tesla that has been built over the last several years We think has the hardware to make the car drive itself We have been working on the software to add higher and higher levels of autonomy This time around last year. We are roughly 2000 cars driving our fsd beta software Since then we have significantly improved the software's robustness and capability That we have now shipped it to 160,000 customers as of today This did not come for free it came from the sweat and blood of the engineering team over the last one year Um, for example, we trained 75,000 neural network models just last one year That's roughly a model every eight minutes That's you know coming out of the team and then we evaluate them on our large clusters and then we ship 281 of those models That actually improved the performance of the car And this space of innovation is happening throughout the stack The the planning software the infrastructure the tools even hiring everything is progressing to the next level The fsd beta software is quite capable of driving the car It should be able to navigate from parking lot to parking lot handling city street driving stopping for traffic lights and stop signs Negotiating with objects at intersections making turns and so on All of this comes from the Uh camera streams that go through our neural networks that run on the car itself It's not coming back to the server or anything It runs on the car and produces all the outputs uh to form the world model or on the car and the planning software drives the car based on that Today we'll go into a lot of the components that make up the system The occupancy network acts as the base geometry layer of the system This is a multi-camera video neural network That from the images predicts the full physical occupancy of the world around the robot So anything that's physically present trees walls buildings Cars balls, whatever you it predicts if it's physically present it predicts them along with their future motion On top of this base level of geometry We have more semantic layers in order to navigate the roadways. We need the lanes, of course But then the roadways have lots of different lanes and they connect in all kinds of ways So it's actually a really difficult problem for typical computer vision techniques to predict the set of lanes and their Connectivities So we reached all the way into language technologies and then pull the state of the art from other Domains are not just computer vision to make this task possible For vehicles, we need their full kinematics state to control for them All of this directly comes from neural networks video streams raw video streams come into the networks Goes through a lot of processing and then outputs the full kinematics state that positions velocities acceleration jerk all of that Directly comes out of networks with minimal post processing. That's really fascinating to me because how how much does it take? Even possible what world do we live in that this magic is possible that these networks predicts fourth derivatives of these positions and people thought We couldn't even detect these objects My opinion is that it did not come for free It it required tons of data.\n\nSo we had to be sophisticated auto labeling systems that shone through raw sensor data Run a ton of offline compute on the servers. It took a lot of time. It took a lot of time. It took a lot of time Run a ton of offline compute on the servers. It can take a few hours run expensive neural networks Distill the information into labels that train our in-car neural networks On top of this we also use our simulation system to synthetically create images and since it's a simulation We trivially have all the labels All of this goes through a well oiled data engine pipeline where we first train a baseline model with some data Ship it to the car see what the failures are and once we know the failures We mind the fleet for the cases where it fails Provide the correct labels and add the data to the training set This process systematically fixes the issues and we do this for every task that runs in the car Yeah, and to train these new massive neural networks This year we expanded our training infrastructure by roughly 40 to 50 percent So that sits us at about 14,000 GPUs today across multiple training clusters in the United States We also worked on our AI compiler which now supports new operations needed by those neural networks And map them to the the best of our underlying hardware resources And our inference engine today is capable of distributing the execution of a single neural network across two independent system on chips Essentially two independent computers interconnected within the same full self-driving computer And to make this possible we have to keep a tight control on the end-to-end latency of this new system So we deployed more advanced scheduling code across the full FSD platform All of these neural networks running in the car Together produce the vector space, which is again the model of the world around the robot or the car And then the planning system operates on top of this coming up with trajectories that avoid collisions or smooth Make progress towards the destination using a combination of model-based optimization Plus neural network that helps optimize it to be really fast Today we are really excited to present progress on all of these areas We have the engineering leads standing by to come in and explain these various blocks and these power not just the car But the same components also run on the Optimus robot that Milan showed earlier With that I welcome Paril to start talking about the planning section Hi all, I'm Paril Jain Let's use this intersection scenario today Let's use this intersection scenario to dive straight into how we do the planning and decision making in autopilot So we are approaching this intersection from a side street and we have to yield to all the crossing vehicles Right with as they are about to enter the intersection The pedestrian on the other side of the intersection decides to cross the road without a crosswalk Now we need to yield to this pedestrian Yield to the vehicles from the right and also understand the relation between the pedestrian and the vehicle on the other side of the intersection So a lot of these intra object dependencies That we need to resolve in a quick glance And humans are really good at this We look at a scene understand all the possible interactions evaluate the most promising ones And generally end up choosing a reasonable one So let's look at a few of these interactions that autopilot system evaluated We could have gone in front of this pedestrian with a very aggressive longitudinal lateral profile Now obviously we are being a jerk to the pedestrian and we would spook the pedestrian and his cute pet We could have moved forward slowly Short for a gap between the pedestrian or end the vehicle from the right Again, we are being a jerk to the vehicle coming from the right But you should not outright reject this interaction in case this is only safe interaction available Lastly the interaction we ended up choosing Stay slow initially find the reasonable gap and then finish the maneuver after all the agents pass Now evaluation of all of these interactions is not trivial Especially when you care about modeling the higher order derivatives for other agents For example, what is the longitudinal jerk required by the vehicle coming from the right when you assert in front of it? Relying purely on collision checks with marginal predictions will only get you so far because you will miss out on a lot of valid interactions This basically boils down to solving a multi-agent joint trajectory planning problem over the trajectories of ego and all the other agents Now how much ever you optimize there's going to be a limit to how fast you can run this optimization problem It will be close to close to order of 10 milliseconds even after a lot of incremental approximations Now for a typical crowded unprotected lift Say you have more than 20 objects Each object having multiple different future modes the number of relevant interaction combinations will blow up The planner needs to make a decision every 50 milliseconds.\n\nSo how do we solve this in real time? We rely on a framework what we call as interaction search, which is basically a paralyzed research over a bunch of maneuver trajectories The state space here corresponds to the kinematic state of ego, the kinematic state of other agents, their nominal future multiple multi-modal predictions and all the static entities in the scene The action space is where things get interesting We use a set of maneuver trajectory candidates to branch over a bunch of interaction decisions and also incremental goals for a longer horizon maneuver Let's walk through this research very quickly to get a sense of how it works We start with a set of vision measurements namely lanes occupancy moving objects These get represented as past attractions as well as latent features We use this to create a set of goal candidates Lanes again from the lanes network or unstructured regions which correspond to a probability mask derived from human demonstrations Once we have a bunch of these goal candidates, we create three trajectories using a combination of classical optimization approaches As well as our network planner again trained on data from the customer fleet Now once we get a bunch of these three trajectories We use them to start branching on the interactions We find the most critical interaction In our case, this would be the interaction with respect to the pedestrian Whether we assert in front of it or yield to it Obviously the option on the left is a high penalty option, it likely won't get prioritized So we branch further onto the option on the right and that's where we bring in more and more complex interactions Building this optimization problem incrementally with more and more constraints And the tree search keeps flowing, branching on more interactions, branching on more goals Now a lot of pricks here lie in evaluation of each of this node of the tree search Inside each node, initially we started with creating trajectories using classical optimization approaches Where the constraints like I described would be added incrementally And this would take close to 1 to 5 milliseconds per action Now even though this is fairly good number, when you want to evaluate more than 100% interactions, this does not scale So we ended up building lightweight queryable networks that you can run in the loop of the planner These networks are trained on human demonstrations from the fleet as well as offline solvers with relaxed time limits With this, we were able to bring the run time down to close to 100 microseconds per action Now doing this alone is not enough because you still have this massive tree search that you need to go through And you need to efficiently prune the search space So you need to do a new scoring on each of these trajectories Few of these are fairly standard, you do a bunch of collision checks, you do a bunch of comfort analysis What is the jerk and access required for a given manure The customer fleet data plays an important role here again We run two sets of again lightweight queryable networks, both really augmenting each other One of them trained from interventions from the FSD beta fleet Which gives a score on how likely is a given manure to result in interventions over the next few seconds And second, which is purely on human demonstrations, human driven data, giving a score on how close is your given selected action to a human driven trajectory The scoring helps us prune the search space, keep branching further on the interactions and focus the compute on the most promising outcomes The cool part about this architecture is that it allows us to create a cool blend between data driven approaches where you don't have to rely on a lot of hand engineered costs But also ground it in reality with physics based checks Now a lot of what I described was with respect to the agents, we could observe in the scene But the same framework extends to all of the other systems that we have We use the video feed from 8 cameras to generate the 3D occupancy of the world The blue mask here corresponds to the visibility region, we call it It basically gets blocked at the first occlusion you see in the scene We consume this visibility mask to generate the visibility of the scene We use the video feed from 8 cameras to generate the 3D occupancy of the world The blue mask here corresponds to the visibility region, we call it In the first occlusion you see in the scene, we consume this visibility mask to generate what we call as ghost objects which you can see on the top left Now if you model the spawn regions and the state transitions of this ghost objects correctly If you tune your control response as a function of their existence likelihood, you can extract some really nice human-like behaviors Now I'll pass it on to Phil to describe more on how we generate these occupancy networks Hey guys, my name is Phil, I will share the details of the occupancy network we built over the past year This network is our solution to model the physical work in 3D around our cars And it is currently not shown in our customer-facing visualization What you will see here is the raw network output from our internal lab tool The occupancy network takes video streams of all our 8 cameras as input Produces a single unified volumetric occupancy in vector space directly For every 3D location around our car, it predicts the probability of that location being occupied or not Since it has video contacts, it is capable of predicting obstacles that are occluded instantaneously For each location, it also produces a set of semantics such as curb, car, pedestrian, and road debris as color-coded here Occupancy flow is also predicted for motion Since the model is a generalized network, it does not tell static and dynamic objects explicitly It is able to produce and model the random motion such as a swarming trainer here This network is currently running in all Teslas with FSD computers And it is incredibly efficient, runs about every 10 milliseconds with our neural-line accelerator So how does this work? Let's take a look at architecture First, we rectify each camera image with a camera calibration And the images we're showing here are given to the network It's actually not the typical 8-bit RGB image As you can see from the first image on top, we're giving the 12-bit raw photo-account image to the network Since it has 4 bits more information, it has 16 times better dynamic range as well as reduced latency Since we don't have to run ISP in the loop anymore We use a set of reglets and bif-fps as a backbone to extract image space features Next, we construct a set of 3D position queries along with the image space features as keys and values fit into an attention module The output of the attention module is high-dimensional spatial features These spatial features are aligned temporally using vehicle odometry to derive motion Next, these spatial temporal features go through a set of deconvolutions to produce the final occupancy and occupancy flow output They're formed as fixed-size voxel grids, which might not be precise enough for planning on control In order to get a higher resolution, we also produce per voxel feature maps which we feed into MLP with 3D spatial point queries to get position and semantics at any arbitrary location After knowing the model better, let's take a look at another example Here we have an articulated bus parked on the right side of the road, highlighted as an L-shaped voxel here As we approach, the bus starts to move. The front of the car turns blue first, indicating the model predicts The front of the bus has a long zero occupancy flow As the bus keeps moving, the entire bus turns blue, and you can also see that the network predicts the precise curvature of the bus This is a very complicated problem for a traditional object detection network, as you'll have to see whether I'm going to use one cuboid or perhaps two to feed the curvature But for an occupancy network, since all we care about is the occupancy in the visible space, we'll be able to model the curvature precisely Besides the voxel grid, the occupancy network also produces a drivel surface The drivel surface has both 3D geometry and semantics. They are very useful for control, especially on hilly and curvy roads The surface and the voxel grid are not predicted independently. Instead, the voxel grid actually aligns with the surface implicitly Here, we are at a hill quest where you can see the 3D geometry of the surface being predicted nicely Planner can use this information to decide perhaps we need to slow down more for the hill quest And as you can also see, the voxel grid aligns with the surface consistently Besides the voxels and the surface, we're also very excited about the recent breakthrough in Neural Radiance Field or NERF We're looking into both incorporating some of the last NERF features into occupancy network training as well as using our network output as the input state for NERF As a matter of fact, Ashok is very excited about this.\n\nThis has been his personal weekend project for a while About these NERFs, because I think the academia is building out of these foundation models for language using tons of large data sets for language But I think for vision, NERFs are going to provide the foundation models for computer vision because they are grounded in geometry And geometry gives us a nice way to supervise these networks and freezes off the requirement to define an ontology And the supervision is essentially free because you just have to differentially render these images So I think in the future, this occupancy network idea where images come in and then the network produces a consistent volumetric representation of the scene That can then be differentially rendered into any image that was observed I personally think it's a future of computer vision and we do some initial work on it right now But I think in the future, both at Tesla and in academia, we will see that this combination of one-shot prediction of volumetric occupancy will be the future That's my personal bet Thanks Ashok So here's an example early result of a 3D reconstruction from our free data Instead of focusing on getting perfect RGB reproduction in image space, our primary goal here is to accurately represent the world in 3D space for driving And we want to do this for all our free data over the world in all weather and lighting conditions And obviously this is a very challenging problem and we're looking for you guys to help Finally, the occupancy network is trained with large auto-labeled data sets without any human in the loop And with that, I'll pass to Tim to talk about what it takes to train this network Thanks Phil Alright, hey everyone Let's talk about some training infrastructure So we've seen a couple of videos, no four or five I think and care more and worry more about a lot more clips on that So we've been looking at the occupancy networks just from Phil Just Phil's videos, it takes 1.4 billion frames to train that network What you just saw and if you have 100,000 GPUs, it would take one hour But if you have one GPU, it would take 100,000 hours So that is not a humane time period that you can wait for your training job to run, right? We want to ship faster than that So that means you're going to need to go parallel So you need a more compute for that That means you're going to need a supercomputer So this is why we've built in-house three supercomputers comprising of 14,000 GPUs Where we use 10,000 GPUs for training and around 4,000 GPUs for auto-labeling All these videos are stored in 30 petabytes of a distributed managed video cache You shouldn't think of our data sets as fixed Let's say as you think of your image net or something, you know, with like a million frames You should think of it as a very fluid thing So we've got half a million of these videos flowing in and out of this cluster These clusters every single day And we track 400,000 of these kind of Python video instantiations every second So that's a lot of calls We're going to need to capture that in order to govern the retention policies of this distributed video cache So underlying all of this is a huge amount of infra, all of which we build and manage in-house So you cannot just buy, you know, 14,000 GPUs and then 30 petabytes of Flash NVMe And you just put it together and let's go train It actually takes a lot of work and I'm going to go into a little bit of that What you actually typically want to do is you want to take your accelerator So that could be the GPU or dojo, which we'll talk about later And because that's the most expensive component, that's where you want to put your bottleneck And so that means that every single part of your system is going to need to outperform this accelerator And so that is really complicated That means that your storage is going to need to have the size and the bandwidth to deliver all the data down into the nodes These nodes need to have the right amount of CPU and memory capabilities to feed into your machine learning framework This machine learning framework then needs to hand it off to your GPU and then you can start training But then you need to do so across hundreds or thousands of GPU in a reliable way in lockstep And in a way that's also fast, so you're also going to need an interconnect Extremely complicated We'll talk more about dojo in a second So first I want to take you through some optimizations that we've done on our cluster So we're getting in a lot of videos and video is very much unlike, let's say, training on images or text Which I think is very well established Video is quite literally a dimension more complicated And so that's why we needed to go end to end from the storage layer down to the accelerator Optimize every single piece of that Because we train on the photon count videos that come directly from our fleet We train on those directly, we do not post-process those at all The way it's just done is we seek exactly to the frames we select for our batch We load those in including the frames that they depend on, so these are your eye frames or your key frames We package those up, move them into shared memory, move them into a double bar from the GPU And then use the hardware decoder that's only accelerated to actually decode the video So we do that on the GPU natively, and this is all in a very nice PyTorch extension Doing so unlocked more than 30% training speed increase for the occupancy networks And freed up basically a whole CPU to do any other thing You cannot just do training with just videos, of course you need some kind of a ground truth And that is actually an interesting problem as well The objective for storing your ground truth is that you want to make sure you get to your ground truth That you need in the minimal amount of file system operations And load in the minimal size of what you need in order to optimize for aggregate cross cluster throughput Because you should see a compute cluster as one big device which has internally fixed constraints and thresholds So for this we rolled out a format that is native to us that's called small We use this for our ground truth, our feature cache and any inference outputs So a lot of tensors that are in there And so just a cartoon here, let's say this is your table that you want to store Then that's how that would look out if you rolled out on disk So what you do is you take anything you'd want to index on, so for example video timestamps You put those all in the header so that in your initial header read you know exactly where to go on disk Then if you have any tensors you're going to try to transpose the dimensions to put a different dimension last as the contiguous dimension And then also try different types of compression Then you check out which one was most optimal and then store that one This is actually a huge tip if you do feature caching Unintelligible output from the machine learning network Rotate around the dimensions a little bit, you can get up to 20% increase in efficiency of storage Then when you store that we also order the columns by size So that all your small columns and small values are together So that when you seek for a single value you're likely to overlap with a read on more values which you'll use later So that you don't need to do another file system operation So I could go on and on, I just went on, touched on two projects that we have internally This is actually part of a huge continuous effort to optimize the compute that we have in-house So accumulating and aggregating through all these optimizations We now train our occupancy networks twice as fast just because it's twice as efficient And now if we add in a bunch more compute and go parallel we can now train this in hours instead of days And with that I'd like to hand it off to the biggest user of compute, John Hi everybody, my name is John Emmons, I lead the autopilot vision team I'm going to cover two topics with you today, the first is how we predict lanes And the second is how we predict the future behavior of other agents on the road In the early days of autopilot we modeled the lane detection problem as an image space instant segmentation task Our network was super simple though, in fact it was only capable of predicting lanes from a few different kinds of geometries Specifically it would segment the ego lane, it could segment adjacent lanes, and then it had some special casing for forks and merges This simplistic modeling of the problem worked for highly structured roads like highways But today we're trying to build a system that's capable of much more complex maneuvers Specifically we want to make left and right turns at intersections where the road topology can be quite a bit more complex and diverse When we try to apply this simplistic modeling of the problem here, it just totally breaks down Taking a step back for a moment, what we're trying to do here is to predict the sparse set of lane instances and their connectivity And what we want to do is to have a neural network that basically predicts this graph where the nodes are the lane segments And the edges encode the connectivity between these lanes So what we have is our lane detection neural network, it's made up of three components In the first component we have a set of convolutional layers, attention layers, and other neural network layers That encode the video streams from our eight cameras on the vehicle and produce a rich visual representation We then enhance this visual representation with a coarse road level map data Which we encode with a set of additional neural network layers that we call the lane guidance module This map is not an HD map, but it provides a lot of useful hints about the topology of lanes inside of intersections, the lane counts on various roads, and a set of other attributes that help us The first two components here produce a dense tensor that sort of encodes the world But what we really want to do is to convert this dense tensor into a sparse set of lanes and their connectivity We approach this problem like an image captioning task where the input is this dense tensor and the output text is predicted into a special language that we developed at Tesla for encoding lanes and their connectivity In this language of lanes, the words and tokens are the lane positions in 3D space In the ordering of the tokens, encrypted modifiers in the tokens encode the connected relationships between these lanes By modeling the task as a language problem, we can capitalize on recent autoregressive architectures and techniques from the language community for handling the multiple-diality of the problem We're not just solving the computer vision problem at Autopilot, we're also applying the state-of-the-art in language modeling and machine learning more generally I'm now going to dive into a little bit more detail of this language component What I have depicted on the screen here is a satellite image which sort of represents the local area around the vehicle The set of nose and edges is what we refer to as the lane graph, and it's ultimately what we want to come out of this neural network We start with a blank slate We're going to want to make our first prediction here at this green dot This green dot's position is encoded as an index into a course grid which discretizes the 3D world Now we don't predict this index directly because it would be too computationally expensive to do so There's just too many grid points and predicting a categorical distribution over this has both implications at training time and test time So instead what we do is we discretize the world coarsely first, we predict the heat map over the possible locations, and then we latch in the most probable location Condition on this, we then refine the prediction and get the precise point Now we know where the position of this token is, but we don't know it's tight In this case though, it's a beginning of a new lane So we predict it as a start token And because it's a start token, there's no additional attributes in our language We then take the predictions from this first forward pass, and we encode them using a learned positional embedding Which produces a set of tensors that we combine together Which is actually the first word in our language of lanes We add this to the first position in our sentence here We then continue this process by predicting the next lane point in a similar fashion Now this lane point is not the beginning of a new lane, it's actually a continuation of the previous lane So it's a continuation token type Now it's not enough just to know that this lane is connected to the previously predicted lane We want to encode its precise geometry, which we do by regressing a set of spline coefficients We then take this lane, we encode it again, and add it as the next word in the sentence We continue predicting these continuation lanes until we get to the end of the prediction grid We then move on to a different lane segment So you can see that cyan dot there Now it's not topologically connected to that pink point It's actually forking off of that green point there So it's got a fork type And fork tokens actually point back to previous tokens from which their fork originates So you can see here the fork point predictor is actually the index zero So it's actually referencing back to a token that is already predicted, like you would in language We continue this process over and over again until we've enumerated all of the tokens in the lane graph And then the network predicts the end of sentence token Yeah, I just wanted to note that the reason we do this is not just because we want to build something complicated It almost feels like a Turing complete machine here with neural networks though Is that we try simple approaches, for example, trying to just segment the lanes along the road or something like that But then the problem is when there's uncertainty, say you cannot see the road clearly And there could be two lanes or three lanes and you can't tell A simple segmentation-based approach would just draw both of them It's kind of a 2.5 lane situation And the post-processing algorithm would hilariously fail when the predictions are such Yeah, the problems don't end there I mean, you need to predict these connective lanes inside of intersections Which is just not possible with the approach that Ashok's mentioning Which is why we had to upgrade to this sort of approach Yeah, when it overlaps like this, segmentation would just go haywire But even if you try very hard to put them on separate layers, it's just a really hard problem But language just offers a really nice framework for getting a sample from a posterior As opposed to trying to do all of this in post-processing But this doesn't actually stop for just autopilot, right? John, this can be used for optimists Yeah, I guess they wouldn't be called lanes But you could imagine, sort of in this stage here That you might have sort of paths that sort of encode the possible places that people could walk Yeah, basically if you're in a factory or in a home setting, you can just ask the robot Okay, please route to the kitchen or please route to some location in the factory And then we predict a set of pathways that would go through the aisles, take the robot And say, okay, this is how you get to the kitchen It just really gives us a nice framework to model these different paths That simplify the navigation problem for the downstream planner Alright, so ultimately what we get from this lane detection network Is a set of lanes in their connectivity, which comes directly from the network There's no additional step here for sparsifying these dense predictions into sparse ones This is just a direct unfiltered output of the network Okay, so I talked a little bit about lanes I'm going to briefly touch on how we model and predict the future paths and other semantics on objects So I'm just going to go really quickly through two examples The video on the right here, we've got a car that's actually running a red light and turning in front of us What we do to handle situations like this is we predict a set of short time horizon future trajectories on all objects We can use these to anticipate the dangerous situation here And apply whatever breaking and steering actions required to avoid a collision In the video on the right, there's two vehicles in front of us The one on the left lane is parked, apparently it's being loaded, unloaded I don't know why the driver decided to park there But the important thing is that our neural network predicted that it was stopped Which is the red color there The vehicle in the other lane, as you notice, also is stationary But that one's obviously just waiting for that red light to turn green So even though both objects are stationary and have zero velocity It's the semantics that is really important here So that we don't get stuck behind that awkwardly parked car Predicting all of these agent attributes presents some practical problems when trying to build a real-time system We need to maximize the frame rate of our object section stack So that autopilot can quickly react to the changing environment Every millisecond really matters here To minimize the inference latency, our neural network is split into two phases In the first phase, we identified the locations in 3D space where agents exist In the second stage, we then pull out tensors at those 3D locations Append it with additional data that's on the vehicle And then we do the rest of the processing This specification step allows the neural network to focus compute on the areas that matter most Which gives us superior performance for a fraction of the latency cost So, putting it all together The autopilot vision stack predicts more than just the geometry and kinematics of the world It also predicts a rich set of semantics, which enables safe and human-like driving I'm now going to hand things off to Sri who will tell us how we run all these cool neural networks on our FSD computer Thank you Hi everyone, I'm Sri Today I'm going to give a glimpse of what it takes to run these FSD networks in the car And how do we optimize for the inference latency? Today I'm going to focus just on the FSD lanes network that John just talked about So, when we started this track, we wanted to know if we can run this FSD lanes network natively on the trip engine Which is our in-house neural network accelerator that we built in the FSD computer When we built this hardware, we kept it simple and we made sure it can do one thing ridiculously fast Dense dot products But this architecture is autoregressive and iterative Where it crunches through multiple attention-attention blocks in the inner loop Producing sparse points directly at every step So, the challenge here was how can we do this sparse point prediction and sparse computation on a dense dot product engine Let's see how we did this on the trip So, the network predicts the heat map of most probable spatial locations of the point To do this on trip, we actually built a lookup table in SRAM And we engineered the dimensions of this embedding such that we could achieve all of this thing with just matrix multiplication Not just that, we also wanted to store this embedding into a token cache So that we don't recompute this for every iteration, rather reuse it for future point prediction Again, we put some tricks here where we did all these operations just on the dot product engine It's actually cool that our team found creative ways to map all these operations on the trip engine In ways that were not even imagined when this hardware was designed But that's not the only thing we had to do to make this work We actually implemented a whole lot of operations and features to make this model compilable To improve the intate accuracy as well as to optimize performance All of these things helped us run this 75 million parameter model just under 10 millisecond of latency Consuming just 8 watts of power But this is not the only architecture running in the car There are so many other architectures, modules and networks we need to run in the car To give a sense of scale, there are about a billion parameters of all the networks combined Producing around 1000 neural network signals So we need to make sure we optimize them jointly and such that we maximize the compute utilization Throughput and minimize the latency So we built a compiler just for neural networks that shares the structure to traditional compilers As you can see, it takes the massive graph of neural nets with 150k nodes and 375k connection Takes this thing, partitions them into independent subgraphs And compiles each of those subgraphs natively for the inference devices Then we have a neural network linker which shares the structure to traditional linker Where we perform this link time optimization There we solve an offline optimization problem with compute memory and memory band with constraints So that it comes with an optimized schedule that gets executed in the car On the runtime, we designed a hybrid scheduling system which basically does heterogeneous scheduling on one SOC And distributed scheduling across both the SOCs to run these networks in a model parallel fashion To get 100 tops of compute utilization, we need to optimize across all the layers of software Right from tuning the network architecture, the compiler, all the way to implementing a low latency high bandwidth RDMA link Across both the SOCs and in fact going even deeper to understanding and optimizing the cache coherent and non-coherent data path of the accelerator in the SOC This is a lot of optimization at every level in order to make sure we get the highest frame rate and as every millisecond counts here And this is just the visualization of the neural networks that are running in the car This is our digital brain essentially As you can see these operations are nothing but just the matrix multiplication, convolution to name a few real operations running in the car To train this network with a billion parameters, you need a lot of labeled data So Egan is going to talk about how do we achieve this with the auto labeling pipeline Thank you Sri Hi everyone, I'm Egan Zhang and I'm leading a geometric vision at autopilot So yeah, let's talk about auto labeling So we have several kinds of auto labeling frameworks to support various types of networks But today I'd like to focus on the awesome lanes net here So to successfully train and generalize this network to everywhere, we think we went tens of millions of trips from probably one million intersection or even more Than how to do that So it is certainly achievable to source sufficient amount of trips because we already have, as Tim explained earlier, we already have like 500,000 trips per day cache rate However, converting all those data into a training form is a very challenging technical problem To solve this challenge, we've tried various ways of manual and auto labeling So from the first column to the second, from the second to the third, each advance provided us nearly 100x improvement in throughput But still, we run an even better auto labeling machine that can provide us good quality, diversity and scalability To meet all these requirements, despite the huge amount of engineering effort required here, we've developed a new auto labeling machine powered by multi-trip reconstruction So this can replace 5 million hours of manual labeling with just 12 hours on cluster for labeling 10,000 trips So how we solved? There are three big steps. The first step is high precision trajectory and structural recovery by multi-camera, visual, inertial, or geometry So here, all the features including ground surface are inferred from videos by neural networks, then tracked and reconstructed in the vector space So the typical trip rate of this trajectory in car is like 1.3 centimeter per meter and 0.45 milliliter per meter, which is pretty decent considering its compact compute requirement Then the recovered surface and road details are also used as a strong guidance for the later manual verification stuff This is also enabled in every FSD vehicle, so we get preprocessed trajectories and structures along with the trip data The second step is multi-trip reconstruction, which is the big and core piece of this machine So the video shows how the previously shown trip is reconstructed and aligned with other trips, basically other trips from different vehicles, not the same vehicle So this is done by multiple internal steps like course alignment, pairwise matching, joint optimization, then further surface refinement In the end, the human analyst comes in and finalizes the label So each heavy steps are already fully parallelized on the cluster, so the entire process usually takes just a couple of hours The last step is actually auto-labeling the new trips So here we use the same multi-trip alignment engine, but only between pre-built reconstruction and each new trip So it's much, much simpler than fully reconstructing all the clips altogether That's why it only takes 30 minutes per trip to auto-label instead of several hours of manual labeling And this is also the key of scalability of this machine This machine easily scales as long as we have available compute and trip data So about 50 trips were newly auto-labeled from this scene and some of them are shown here, so 53 from different vehicles So this is how we capture and transform the space-time slices of the world into the network supervision One thing I'd like to note is that Jagan just talked about how we auto-label our lanes We have auto-labels for almost every task that we do, including our planner And many of these are fully automatic, there's no humans involved For example, for objects, all the kinematics, the shapes, the futures, everything just comes from auto-labeling And the same is true for our occupancy too, and we have really just built a machine around this Yeah, so if you can go back one slide One more, it says parallelized on cluster So that sounds pretty straightforward, but it really wasn't Maybe it's fun to share how something like this comes about So a while ago we didn't have any auto-labeling at all, and then someone makes a script It starts to work, it starts working better, until you reach a volume that's pretty high And we clearly need a solution And so there were two other engineers in our team who were like, you know, that's an interesting, you know, thing What we needed to do was build a whole graph of essentially Python functions that would need to run one after the other First you pull the clip, then you do some cleaning, then you do some network inference, then another network inference Until you finally get this But so you need to do this at a large scale, so I tell them we probably need to shoot for, you know, 100,000 clips per day Or like 100,000 items, that seems good And so the engineers said, well, we can do, you know, a bit of post-gres and a bit of elbow grease, we can do it Meanwhile, we are a bit later and we're doing 20 million of these functions every single day Again, we pull in around half a million clips and on those we run a ton of functions, each of these, in a streaming fashion And so that's kind of the backend infra that's also needed to not just run training, but also auto-labeling Yeah, it really is like a factory that produces labels and production lines, yield, quality, inventory Like all of these same concepts applied to this label factory that applies for, you know, the factory for our cars That's right Okay, thanks, Tim and Ashok So, yeah, so concluding this section, I'd like to share a few more challenging and interesting examples for network for sure And even for humans, probably So from the top, there's like examples for like lack of lights, case or foggy night or roundabout and occlusions by heavy occlusions by parked cars And even rainy night with rain drops on camera lenses These are challenging, but once their original scenes are fully reconstructed by other clips, all of them can be auto-labeled So that our cars can drive even better through these challenging scenarios So, now, let me pass the mic to David to learn more about how Sim is creating the new world on top of these labels Thank you Thank you, Yegan My name is David and I'm going to talk about simulation So simulation plays a critical role in providing data that is difficult to source and or hard to label However, 3D scenes are notoriously slow to produce Take for example, the simulated scene playing behind me A complex intersection from Market Street in San Francisco It would take two weeks for artists to complete And for us, that is painfully slow However, I'm going to talk about using Yegan's automated ground truth labels along with some brand new tooling that allows us to procedurally generate this scene in many like it in just five minutes That's an amazing a thousand times faster than before So let's dive in to how a scene like this is created We start by piping the automated ground truth labels into our simulated world creator tooling inside the software Houdini Starting with road boundary labels, we can generate a solid road mesh and re-topologize it with the lane graph labels This helps inform important road details like cross-road slope and detailed material blending Next, we can use the line data and sweep geometry across its surface and project it to the road, creating lane paint decals Next, using median edges, we can spawned island geometry and populate it with randomized foliage This drastically changes the visibility of the scene Now the outside world can be generated through a series of randomized heuristics Modular building generators create visual obstructions while randomly placed objects like hydrants can change the color of the curves while trees can drop leaves below it obscuring lines or edges Next, we can bring in map data to inform positions of things like traffic traffic lights or stop signs We can trace along its normal to collect important information like number of lanes and even get accurate street names on the signs themselves Next, using lane graph, we can determine lane connectivity and spawn directional road markings on the road and their accompanying road signs And finally, with lane graph itself, we can determine lane adjacency and other useful metrics to spawn randomized traffic permutations inside our simulator And again, this is all automatic, no artist in the loop and happens within minutes And now this sets us up to do some pretty cool things Since everything is based on data and heuristics, we can start to fuzz parameters to create visual variations of the single ground truth It can be as subtle as object placement and random material swapping to more drastic changes like entirely new biomes or locations of environment like urban, suburban, or rural This allows us to create infinite, targeted permutations for specific ground truths that we need more ground truth for And all this happens within a click of a button And we can even take this one step further by altering our ground truth itself Say John wants his network to pay more attention to directional road markings to better detect an upcoming captive left turn lane We can start to procedurally alter our lane graph inside the simulator to help create entirely new flows through this intersection to help focus the network's attention to the road markings to create more accurate predictions And this is a great example of how this tooling allows us to create new data that can never be collected from the real world And the true power of this tool is in its architecture and how we can run all tasks in parallel to infinitely scale So you saw the tile creator tool in action converting the ground truth labels into their counterparts Next we can use our tile extractor tool to divide this data into geo hash tiles about 150 meter square in size We then save out that data into separate geometry and instance files This gives us a clean source of data that's easy to load and allows us to be rendering engine agnostic for the future Then using a tile loader tool we can summon any number of those cache tiles using a geo hash ID Currently we're doing about these 5x5 tiles or 3x3 usually centered around fleet hotspots or interesting lane graph locations And the tile loader also converts these tile sets into U assets for consumption by the unreal engine and gives you a finished product from what you saw in the first slide And this really sets us up for size and scale And as you can see on the map behind us we can easily generate most of San Francisco city streets And this didn't take years or even months of work but rather two weeks by one person We can continue to manage and grow all this data using our PDG network inside of the tooling This allows us to throw compute at it and regenerate all these tile sets overnight This ensures all environments are consistent, quality and features which is super important for training since new ontologies and signals are constantly released And now to come full circle, because we generated all these tile sets from ground truth data They contain all the weird intricacies from the real world We can combine that with the procedural, visual and traffic variety to create limitless, targeted data for the network to learn from And that concludes the SIM section, I'll pass it to Kate to talk about how we can use all this data to improve autopilot Thank you Thanks David, hi everyone, my name is Kate Park and I'm here to talk about the data engine Which is the process by which we improve our neural networks via data We're going to show you how we deterministically solve interventions via data And walk you through the life of this particular clip In this scenario, autopilot is approaching a turn and incorrectly predicts that crossing vehicle as stopped for traffic and thus a vehicle that we would slow down for In reality, there's nobody in the car, it's just awkwardly parked We've built this tooling to identify the mispredictions, correct the label and categorize this clip into an evaluation set This particular clip happens to be one of 126 that we've diagnosed as challenging parked cars at turns Because of this infra, we can curate this evaluation set without any engineering resources custom to this particular challenge case To actually solve that challenge case requires mining thousands of examples like it And it's something Tesla can trivially do We simply use our data sourcing infra, request data and use the tooling shown previously to correct the labels By surgically targeting the mispredictions of the current model, we're only adding the most valuable examples to our training set We surgically fix 13,900 clips and because those were examples where the current model struggles We don't even need to change the model architecture, a simple weight update with this new valuable data is enough to solve the challenge case So you see we no longer predict that crossing vehicle as stopped, as shown in orange, but parked, as shown in red In academia, we often see that people keep data constant, but at Tesla it's very much the opposite We see time and time and again that data is one of the best if not the most deterministic lever to solving these interventions We just showed you the data engine loop for one challenge case, namely these parked cars at turns But there are many challenge cases even for one signal of vehicle movement We apply this data engine loop to every single challenge case we've diagnosed, whether it's buses, curvy roads, stopped vehicles, parking lots And we don't just add data once, we do this again and again to perfect the semantic In fact, this year we updated our vehicle movement signal five times and with every weight update trained on the new data We push our vehicle movement accuracy up and up This data engine framework applies to all our signals, whether they're 3D, multi-cam video, whether the data is human labeled, auto-labeled, or simulated Whether it's an offline model or an online model And Tesla is able to do this at scale because of the fleet advantage, the infra that our NG team has built, and the labeling resources that feed our networks To train on all this data, we need a massive amount of compute, so I'll hand it off to Pete and Ganesh to talk about the Dojo supercomputing platform Thank you Thank you, Katie Thanks everybody, thanks for hanging in there, we're almost there My name is Pete Bannon, I run the custom silicon and low voltage teams at Tesla And my name is Ganesh Renke, I run the Dojo program Thank you I'm frequently asked, why is a car company building a supercomputer for training?\n\nAnd this question fundamentally misunderstands the nature of Tesla At its heart, Tesla is a hardcore technology company All across the company, people are working hard in science and engineering to advance the fundamental understanding and methods that we have available to build cars, energy solutions, robots, and anything else that we can do to improve the human condition around the world It's a super exciting thing to be a part of, and it's a privilege to run a very small piece of it in the semiconductor group Tonight we're going to talk a little bit about Dojo and give you an update on what we've been able to do over the last year But before we do that, I wanted to give a little bit of background on the initial design that we started a few years ago When we got started, the goal was to provide a substantial improvement to the training latency for our autopilot team Some of the largest neural networks they train today run for over a month, which inhibits their ability to rapidly explore alternatives and evaluate them So a 30X speedup would be really nice if we could provide it at a cost competitive and energy competitive way To do that, we wanted to build a chip with a lot of arithmetic units that we could utilize at a very high efficiency And we spent a lot of time studying whether we could do that using DRAM, various packaging ideas, all of which failed And in the end, even though it felt like an unnatural act, we decided to reject DRAM as the primary storage medium for this system And instead focus on SRAM embedded in the chip SRAM provides, unfortunately, a modest amount of capacity, but extremely high bandwidth and very low latency, and that enables us to achieve high utilization with the arithmetic units Those choices, that particular choice led to a whole bunch of other choices For example, if you want to have virtual memory, you need page tables, they take up a lot of space, we didn't have space, so no virtual memory So we also don't have interrupts, the accelerator is a bare bonds, raw piece of hardware that's presented to a compiler and the compiler is responsible for scheduling everything that happens in a deterministic way So there's no need or even desire for interrupts in the system We also chose to pursue model parallelism as a training methodology, which is not the typical situation most machines today use data parallelism, which consumes additional memory capacity, which we obviously don't have So all of those choices led us to build a machine that is pretty radically different from what's available today We also had a whole bunch of other goals, one of the most important ones was no limits So we wanted to build a compute fabric that would scale in an unbounded way for the most part, I mean obviously there's physical limits now and then But pretty much if your model was too big for the computer, you just had to go buy a bigger computer, that's what we were looking for Today the way machines are packaged, there's a pretty fixed ratio of for example GPU, CPUs and DRAM capacity and network capacity And we really wanted to disaggregate all that so that as models evolved, we could vary the ratios of those various elements and make the system more flexible to meet the needs of the autopilot team And it's so true, no limits philosophy was our guiding star all the way, all of our choices were centered around that And to the point that we didn't want traditional data center infrastructure to limit our capacity to execute these programs at speed That's why we integrated vertically our data center, the entire data center by doing a vertical integration of the data center We could extract new levels of efficiency, we could optimize power delivery, cooling and as well as system management across the whole data center stack Rather than doing box by box and integrating that, those boxes into data centers And to do this, we also wanted to integrate early to figure out limits of scale for our software workloads So we integrated Dojo environment into our autopilot software very early and we learned a lot of lessons And today Bill Chang will go over our hardware update as well as some of the challenges that we faced along the way And Rajiv Kurian will give you a glimpse of our compiler technology as well as go over some of our cool results Great Thanks Pete, thanks Ganesh I'll start tonight with a high level vision of our system that will help set the stage for the challenges and the problems we're solving And then also how software will then leverage this for performance Now our vision for Dojo is to build a single unified accelerator, a very large one Software would see a seamless compute plane with globally addressable, very fast memory and all connected together with uniform high bandwidth and low latency Now to realize this, we need to use density to achieve performance Now we leverage technology to get this density in order to break levels of hierarchy all the way from the chip to the scale out systems Now silicon technology has done this for decades Chips have followed Moore's law for density integration to get performance scaling Now a key step in realizing that vision was our training tile Probably can we integrate 25 dies at extremely high bandwidth but we can scale that to any number of additional tiles by just connecting them together Now last year we showcased our first functional training tile and at that time we already had workloads running on it And since then the team here has been working hard and diligently to deploy this at scale Now we've made amazing progress and had a lot of milestones along the way And of course we've had a lot of unexpected challenges But this is where our fail fast philosophy has allowed us to push our boundaries Now pushing density for performance presents all new challenges One area is power delivery Here we need to deliver the power to our compute die and this directly impacts our top line compute performance But we need to do this at unprecedented density We need to be able to match our die pitch with a power density of almost 1 amp per millimeter squared And because of the extreme integration this needs to be a multi-tiered vertical power solution And because there's a complex heterogeneous material stack up we have to carefully manage the material transition Especially CTE Now why does the coefficient of thermal expansion matter in this case? CTE is a fundamental material property and if it's not carefully managed that stack up would literally rip itself apart We started this effort by working with vendors to develop this power solution But we realized that we actually had to develop this in-house Now to balance schedule and risk we built quick iterations to support both our system bring up in software development And also to find the optimal design and stack up that would meet our final production goals And in the end we were able to reduce CTE over 50% and meet our performance by 3x over our initial version Now needless to say finding this optimal material stack up while maximizing performance at density is extremely difficult Now we did have unexpected challenges along the way Here's an example where we pushed the boundaries of integration that led to component failures This started when we scaled up to larger and longer workloads and then intermittently a single site on a tile would fail Now they started out as recoverable failures but as we pushed some much higher and higher power these would become permanent failures Now to understand this failure you have to understand why and how we build our power modules Solving density at every level is the cornerstone of actually achieving our system performance Now because our XY plane is used for high bandwidth communication everything else must be stacked vertically This means all other components other than our die must be integrated into our power modules Now that includes our clock and our power supplies and also our system controllers Now in this case the failures were due to losing clock output from our oscillators And after an extensive debug we found that the root cause was due to vibrations on the module from piezoelectric effects Our nearby capacitors Now singing caps are not a new phenomenon and in fact very common in power design But normally clock chips are placed in a very quiet area of the board and often not affected by power circuits But because we needed to achieve this level of integration these oscillators need to be placed in very close proximity Now due to our switching frequency and then the vibration resonance created It caused out of plane vibration on our MEMS oscillator that caused it to crack Now the solution to this problem is a multi-prong approach We can reduce the vibration by using soft terminal caps We can update our MEMS part with a lower Q factor for the out of plane direction And we can also update our switching frequency to push the resonance further away from these sensitive bands Now in addition to the density at the system level we've been making a lot of progress at the infrastructure level We knew that we had to read examine every aspect of the data center infrastructure in order to support our unprecedented power and cooling density We brought in a fully custom designed CDU to support Dojo's dense cooling requirements And the amazing part is we're able to do this at a fraction of the cost versus buying off the shelf and modifying it And since our Dojo cabinet integrates enough power and cooling to match an entire row of standard IT racks We need to carefully design our cabinet and infrastructure together And we've already gone through several iterations of this cabinet to optimize this And earlier this year we started low testing our power and cooling infrastructure And we were able to push it over 2 megawatts before we tripped our substation and got a call from the city Now last year we introduced only a couple of components of our system The custom D1 die and the training tile, but we teased the exit pod as our end goal We'll walk through the remaining parts of our system that are required to build out this exit pod Now the system tray is a key part of realizing our vision of a single accelerator It enables us to seamlessly connect tiles together, not only within the cabinet, but between cabinets We can connect these tiles at very tight spacing across the entire accelerator And this is how we achieve our uniform communication This is a laminated bus bar that allows us to integrate very high power, mechanical and thermal support, and an extremely dense integration It's 75 millimeters in height and supports 6 tiles at 135 kilograms This is the equivalent of 3 to 4 fully loaded high performance racks Next we need to feed data to the training tiles This is where we've developed the Dojo interface processor It provides our system with high bandwidth DRAM to stage our training data And it provides full memory bandwidth to our training tiles using TTP, our custom protocol that we use to communicate across our entire accelerator It also has high speed Ethernet that helps us extend this custom protocol over standard Ethernet And we provide native hardware support for this with little to no software overhead And lastly we can connect to it through a standard Gen4 PCIe interface Now we pair 20 of these cards per tray and that gives us 640 gigabytes of high bandwidth DRAM And this provides our disaggregated memory layer for our training tiles These cards are a high bandwidth ingest path both through PCIe and Ethernet They also provide a high-ratex Z-connectivity path that allows shortcuts across our large Dojo accelerator Now we actually integrate the host directly underneath our system tray These hosts provide our ingest processing and connect to our interface processors through PCIe These hosts can provide hardware video decoder support for video-based training And our user applications land on these hosts so we can provide them with the standard X86 Linux environment Now we can put two of these assemblies into one cabinet and pair it with redundant power supplies that do direct conversion of three-phase 480-volt AC power to 52-volt DC power Now by focusing on density at every level we can realize the vision of a single accelerator Now starting with the uniform nodes on our custom D1 die we can connect them together in our fully integrated training tile And then finally seamlessly connecting them across cabinet boundaries to form our Dojo accelerator And all together we can house two full accelerators in our Exapod for a combined one exa-flop of ML compute Now all together this amount of technology and integration has only ever been done a couple of times in the history of compute Next we'll see how software can leverage this to accelerate their performance Thanks Bill, my name is Rajiv and I'm going to talk some numbers So our software stack begins with the PyTorch extension that speaks to our commitment to run standard PyTorch models out of the box We're going to talk more about our JIT compiler and the ingest pipeline that feeds the hardware with data Abstractly, performance is tops times utilization times accelerator occupancy We've seen how the hardware provides peak performance is the job of the compiler to extract utilization from the hardware while code is running on it And it's the job of the ingest pipeline to make sure that data can be fed at a throughput high enough for the hardware to not ever starve So let's talk about why communication-bound models are difficult to scale But before that let's look at why ResNet 50-like models are easier to scale You start off with a single accelerator, run the forward and backward passes, followed by the optimizer Then to scale this up you run multiple copies of this on multiple accelerators And while the gradients produced by the backward pass do need to be reduced and this introduces some communication, this can be done pipeline with the backward pass This setup scales fairly well, almost linearly For models with much larger activations we run into a problem as soon as we want to run the forward pass The batch size that fits in a single accelerator is often smaller than the batch norm surface So to get around this researchers typically run this setup on multiple accelerators in sync batch norm mode This introduces latency bound communication to the critical path of the forward pass and we already have a communication bottleneck And while there are ways to get around this they usually involve tedious manual work best suited for a compiler And ultimately there's no skirting around the fact that if your state does not fit in a single accelerator you can be communication bound And even with significant efforts from our ML engineers we see such models don't scale linearly The doger system was built to make such models work at high utilization The high density integration was built to not only accelerate the compute bound portions of a model but also the latency bound portions Like a batch norm or the bandwidth bound portions like a gradient all reduced or a parameter all gathered A slice of the doger mesh can be carved out to run any model The only thing users need to do is to make the slice large enough to fit a batch norm surface for their particular model After that the partition presents itself as one large accelerator freeing the users from having to worry about the internal details of execution And as the job of the compiler to maintain this abstraction Fine grain synchronization primitives in uniform low latency makes it easy to accelerate all forms of parallelism across integration boundaries Tensors are usually stored sharded in SRAM and replicated just in time for a layer's execution We depend on the high doger bandwidth to hide this replication time Tensor replication and other data transfers are overlapped with compute and the compiler can also recompute layers when it's profitable to do so We expect most models to work out of the box As an example we took the recently released stable diffusion model and got it running on dojo in minutes Out of the box the compiler was able to map it in a model parallel manner on 25 dojo dies Here are some pictures of a Cybertruck on Mars generated by stable diffusion running on dojo Looks like it still has some ways to go before matching the Tesla design studio team So we've talked about how communication bottlenecks can hamper scalability Perhaps an asset test of a compiler and the underlying hardware is executing a cross die batch norm layer Like mentioned before this can be a serial bottleneck The communication phase of a batch norm begins with nodes computing their local mean and standard deviations Then coordinating to reduce these values, then broadcasting these values back and then they resume their work in parallel So what would an ideal batch norm look like on 25 dojo dies? Let's say the previous less activations are already split across dies We would expect the 350 nodes on each die to coordinate and produce die local mean and standard deviation values Ideally these would get further reduced with the final value ending somewhere towards the middle of the tile We would then hope to see a broadcast of this value radiating from the center Let's see how the compiler actually executes a real batch norm operation across 25 dies The communication trees were extracted from the compiler and the timing is from a real hardware one We're about to see 8,750 nodes on 25 dies coordinating to reduce and then broadcast the batch norm mean and standard deviation values Die local reduction followed by global reduction towards the middle of the tile Then the reduced value broadcast radiating from the middle accelerated by the hardware's broadcast facility This operation takes only 5 microseconds on 25 dojo dies The same operation takes 150 microseconds on 24 GPUs This is an orders of magnitude improvement over GPUs And while we talked about an already used operation in the context of a batch norm It's important to reiterate that the same advantages apply to all other communication primitives And these primitives are essential for large scale training So how about full model performance? So while we think that ResNet 50 is not a good representation of real world Tesla workloads It is a standard benchmark, so let's start there We are already able to match the 100 die for die However, perhaps a hint of dojo's capabilities is that we're able to hit this number with just a batch of 8 per die But dojo was really built to tackle larger complex models So when we set out to tackle real world workloads, we looked at the usage patterns of our current GPU cluster And two models stood out, the autolabeling networks, a class of offline models that are used to generate ground truth And the occupancy networks that you heard about The autolabeling networks are large models that have high arithmetic intensity While the occupancy networks can be ingest bound We chose these models because together they account for a large chunk of our current GPU cluster usage And they would challenge the system in different ways So how do we do on these two networks? The results we're about to see were measured on multi die systems for both the GPU and dojo, but normalized to per die numbers On our autolabeling network, we're already able to surpass the performance of an A100 With our current hardware running on our older generation VRMs On our production hardware with our newer VRMs, that translates to doubling the throughput of an A100 And our model showed that with some key compiler optimizations, we could get to more than 3x the performance of an A100 We see even bigger leaps on the occupancy network Almost 3x with our production hardware, with room for more So what does that mean for Tesla? With a current level of compiler performance, we could replace the ML compute of 1, 2, 3, 4, 5 and 6 GPU boxes with just a single dojo tile And this dojo tile costs less than one of these GPU boxes What it really means is that networks that took more than a month to train now take less than a week Alas, when we measure things, it did not turn out so well.\n\nAt the PyTorch level, we did not see our expected performance out of the gate And this timeline chart shows our problem. The teeny, tiny little green bars, that's the compile code running on the accelerator The row is mostly white space where the hardware is just waiting for data With our dense ML compute, dojo hosts effectively have 10x more ML compute than the GPU hosts. The data loader is running on this one host Simply couldn't keep up with all that ML hardware So to solve our data loader scalability issues, we knew we had to get over the limit of this single host The Tesla transport protocol moves data seamlessly across hosts, tiles and ingest processors So we extended the Tesla transport protocol to work over Ethernet. We then built the dojo network interface card, the D-NIC, to leverage TTP over Ethernet This allows any host with a D-NIC card to be able to DMA2 and from other TTP endpoints So we started with the dojo mesh, then we added a tier of data loading hosts equipped with the D-NIC card We connected these hosts to the mesh via an Ethernet switch. Now every host in this data loading tier is capable of reaching all TTP endpoints in the dojo mesh via hardware accelerated DMA After these optimizations went in, our occupancy went from 4% to 97% So the data loading sections have reduced drastically and the ML hardware has kept busy We actually expect this number to go to 100% pretty soon After these changes went in, we saw the full expected speed up from the PyTorch layer and we were back in business So we started with hardware design that breaks through traditional integration boundaries in service of our vision of a single giant accelerator We've seen how the compiler and ingest layers build on top of that hardware So after proving our performance on these complex real-world networks, we knew what our first large-scale deployment would target Our high arithmetic intensity auto-labeling networks Today that occupies 4,000 GPUs over 72 GPU racks With our dense computer and our high performance, we expect to provide the same throughput with just 4 dojo cabinets And these 4 dojo cabinets will be part of our first exapod that we plan to build by quarter one of 2023 This will more than double Tesla's auto-labeling capacity The first exapod is part of a total of 7 exapods that we plan to build in Palo Alto right here across the wall And we have a display cabinet from one of these exapods for everyone to look at 6 tiles densely packed on a tray, 54 petaflops of compute, 640 gigabytes of high bandwidth memory with power and host defeated A lot of compute And we're building out new versions of all our cluster components and constantly improving our software to hit new limits of scale We believe that we can get another 10x improvement with our next generation hardware And to realize our ambitious goals, we need the best software and hardware engineers So please come talk to us or visit tesla.com. Alright, so hopefully that was enough detail And now we can move to questions And guys, I think the team can come out on stage We really wanted to show the depth and breadth of Tesla in artificial intelligence, compute hardware, robotics actuators And try to really shift the perception of the company away from, you know, a lot of people think we're like just a car company Or we make cool cars, whatever But most people have no idea that Tesla is arguably the leader in real world AI hardware and software And that we're building what is arguably the most radical computer architecture since the Kray-1 supercomputer And I think if you're interested in developing some of the most advanced technology in the world that's going to really affect the world in a positive way Tesla's the place to be So yeah, let's fire away with some questions I think there's a mic at the front and a mic at the back Just throw mics at people Jump all for the mic Yeah, hi, thank you very much I was impressed here I was impressed very much by Optimus, but I wonder why did not driven the hand Why did you choose a tendon-driven approach for the hand?\n\nBecause tendons are not very durable And why spring-loaded? Cool, awesome, yes, that's a great question You know, when it comes to any type of actuation scheme, there's trade-offs between, you know, whether or not it's a tendon-driven system or some type of linkage-based system Keep the mic close to your mouth A little bit closer, hear me? Cool Yeah, the main reason why we went for a tendon-based system is that, you know, first we actually investigated some synthetic tendons, but we found that metallic boating cables are, you know, a lot stronger One of the advantages of these cables is that it's very good for part reduction We do want to make a lot of these hands, so having a bunch of parts, a bunch of small linkages ends up being, you know, a problem when you're making a lot of something One of the big reasons that, you know, tendons are better than linkages in a sense is that you can be anti-backlash So anti-backlash essentially, you know, allows you to not have any gaps or, you know, stuttering motion in your fingers Spring-loaded, mainly what spring-loaded allows us to do is allows us to have active opening So instead of having to have two actuators to drive the fingers closed and then open, we have the ability to, you know, have the tendon drive them closed and then the springs passively extend And this is something that's seen in our hands as well, right? We have the ability to actively flex and then we also have the ability to extend Yeah I mean, our goal with Optimus is to have a robot that is maximally useful as quickly as possible So there's a lot of ways to solve the various problems of a humanoid robot And we're probably not barking up the right tree on all the technical solutions And I should say that we're open to evolving the technical solutions that you see here over time, they're not locked in stone But we have to pick something, and we want to pick something that's going to allow us to produce the robot as quickly as possible and have it, like I said, be useful as quickly as possible We're trying to follow the goal of fastest path to a useful robot that can be made at volume And we're going to test the robot internally at Tesla in our factory and just see, like, how useful is it Because you have to have a, you've got to close the loop on reality to confirm that the robot is in fact useful And, yeah, so we're just going to use it to build things And we're confident we can do that with the hand that we have currently designed But I'm sure there'll be hand version 2, version 3, and we may change the architecture quite significantly over time Hi, the Optimus robot is really impressive, you did a great job, bipedal robots are really difficult But what I noticed might be missing from your plan is to acknowledge the utility of the human spirit And I'm wondering if Optimus will ever get a personality and be able to laugh at our jokes while it folds our clothes Yeah, absolutely. I think we want to have really fun versions of Optimus And so that Optimus can both be utilitarian and do tasks, but can also be kind of like a friend and a buddy And hang out with you, and I'm sure people will think of all sorts of creative uses for this robot And, you know, the thing, once you have the core intelligence and actuators figured out Then you can actually, you know, put all sorts of costumes, I guess, on the robot I mean, you can make the robot look, you can skin the robot in many different ways And I'm sure people will find very interesting ways to, yeah, versions of Optimus Thanks for the great presentation I wanted to know if there was an equivalent to interventions in Optimus It seems like labeling through moments where humans disagree with what's going on is important And in a humanoid robot, that might be also a desirable source of information Yeah, I think we will have ways to remote operate the robot and intervene when it does something bad Especially when we are training the robot and bringing it up And hopefully we, you know, design it in a way that we can stop the robot from, if it's going to hit something We can just, like, hold it and it will stop, it won't, like, you know, crush your hand or something And those are all intervention data Yeah, and we can learn a lot from our simulation systems, too Where we can check for collisions and supervise that those are bad actions Yeah, I mean, so Optimus, we went over time for it to be, you know, an android, the kind of android that you've seen in sci-fi movies Like Star Trek, The Next Generation, like data But obviously we could program the robot to be less robot-like and more friendly And, you know, you can obviously learn to emulate humans and feel very natural So as AI in general improves, we can add that to the robot And, you know, it should be obviously able to do simple instructions or even intuit what it is that you want So you could give it a high level instruction and then it can break that down into a series of actions And take those actions Hi, yeah, it's exciting to think that with the Optimus you will think that you can achieve orders of magnitude of improvement in economic output That's really exciting And when Tesla started, the mission was to accelerate the advent of renewable energy or sustainable transport So with the Optimus, do you still see that mission being the mission statement of Tesla or is it going to be updated with, you know, mission to accelerate the advent of, I don't know, infinite abundance or limitless economy Yeah, it is not strictly speaking, Optimus is not strictly speaking directly in line with accelerating sustainable energy To the degree that it is more efficient at getting things done than a person, it does, I guess, help with sustainable energy But I think the mission effectively does somewhat broaden with the advent of Optimus to, you know, I don't know, making the future awesome So, you know, I think you look at Optimus and I know about you, but I'm excited to see what Optimus will become And, you know, this is like, you know, if you could, I mean, you can tell like any given technology, do you want to see what it's like in a year, two years, three years, four years, five years, ten? I'd say for sure, you definitely want to see what's happened with Optimus Whereas, you know, a bunch of other technologies are, you know, sort of plateaued About name names here, but, you know, so, I think Optimus is going to be incredible in like five years, ten years like mind-blowing And I'm really interested to see that happen, and I hope you are too I have a quick question here, Justin, and I was wondering, like, are you planning to extend like conversational capabilities for the robot?\n\nAnd my second full-on question to that is, what's like the end goal? What's the end goal with Optimus? Yeah, Optimus would definitely have conversational capabilities So, you'd be able to talk to it and have a conversation, and it would feel quite natural So, from an end goal standpoint, I don't know, I think it's going to keep evolving, and I'm not sure where it ends up, but some place is interesting for sure And, you know, we always have to be careful about the, you know, don't go down the terminator path That's a, you know, I thought maybe we should start off with a video of like the terminator starting off with this, you know, skull crushing But that might be, you know, people might not get too seriously So, you know, we do want Optimus to be safe, so we are designing in safeguards where you can locally stop the robot And, you know, with like basically a localized control ROM that you can't update over the internet Which I think that's quite important, essential, frankly So, like a localized stop button or remote control, something like that, that cannot be changed But, I mean, it's definitely going to be interesting, it won't be boring Okay, yeah, I see today you have a very attractive product with Dojo and its applications So, I'm wondering what's the future for the Dojo platform? So, you know, like provide like infrastructure and service like AWS or you will like sell the chip like the NVIDIA So, basically, what's the future? Because I say you use 7nm, so the developer cost is like easily over 10 million US dollars How do you make the business like business wise? Dojo is a very big computer and actually will use a lot of power and need a lot of cooling So, I think it's probably going to make more sense to have Dojo operate in like an Amazon Web Services manner Than to try to sell it to someone else So, that would be the most efficient way to operate Dojo is just have it be a service that you can use That's available online and that where you can train your models way faster and for less money And as the world transitions to software 2.0 And that's on the bingo card Someone I know has to know to drink 5 tequila So, let's see, software 2.0 will use a lot of neural net training So, it kind of makes sense that over time as there's more neural net stuff People will want to use the fastest, lowest cost neural net training system So, I think there's a lot of opportunity in that direction Hi, my name is Ali Jahanian Thank you for this event, it's very inspirational My question is, I'm wondering what is your vision for humanoid robots that understand our emotions and art And can contribute to our creativity Well, I think you're already seeing robots that at least are able to generate very interesting art Like Dali and Dali 2 And I think we'll start seeing AI that can actually generate even movies that have coherence Like interesting movies and tell jokes So, it's quite remarkable how fast AI is advancing at many companies besides Tesla We're headed for a very interesting future And yeah, so, any guys want to comment on that?\n\nYeah, I guess the Optimus Robot can come up with physical art, not just digital art You can ask for some dance moves in text or voice and then you can produce those in the future So, it's a lot of physical art, not just digital art Oh, yeah, computers can absolutely make physical art, yeah, 100% Yeah, like dance, play soccer or whatever you... I mean, it needs to get more agile over time, for sure Thanks so much for the presentation Now, for the Tesla Autopilot slides, I noticed that the models that you were using were heavily motivated by language models And I was wondering what the history of that was and how much of an improvement it gave I thought that that was a really interesting, curious choice to use language models for the lane transitioning So, there are sort of two aspects for why we transition to language modeling So, the first... Talk loud and close Okay, got it Yeah, so the language models help us in two ways The first way is that it lets us predict lanes that we couldn't have otherwise As Ashok mentioned earlier, basically when we predicted lanes in sort of a dense 3D fashion You can only model certain kinds of lanes, but we want to get those criss-crossing connections inside of intersections It's just not possible to do that without making it a graph prediction If you try to do this with dense segmentation, it just doesn't work Also, the lane prediction is a multimodal problem Sometimes you just don't have sufficient visual information to know precisely how things look on the other side of the intersection So you need a method that can generalize and produce coherent predictions You don't want to be predicting two lanes and three lanes at the same time You want to commit to one in a general model like these language models provides that Hi Hi, my name is Giovanni Yeah, thanks for the presentation. It's really nice I have a question for FSD team For the neural networks, how do you test... How do you do unit tests, software unit tests on that? Do you have a bunch or I don't know, mid-thousands or...\n\nYes, cases where the neural network that after you train it, you have to pass it Before you release it as a product, right? Yeah, what's your software unit testing strategies for this, basically? Yeah, glad you asked. There's like a series of tests that we have defined starting from unit tests for software itself But then for the neural network models, we have VAP sets defined where you can define... If you just have a large test set, that's not enough what we find We need like sophisticated VAP sets for different failure modes And then we queate them and grow them over the time of the product So over the years, we have like hundreds of thousands of examples where we have been failing in the past That we have curated and so for any new model, we test against the entire history of these failures And then keep adding to this test set On top of this, we have shadow modes where we ship these models in silent to the car And we get data back on where they are failing or succeeding And there's an extensive QA program It's very hard to ship for regression There's like nine levels of filters before it hits customers But then we have really good infra to make this all efficient I'm one of the QA testers, so I have QA the car... Yeah, QA tester Yeah, so I'm constantly in the car just being queuing like whatever the latest alpha build is that doesn't totally crash Yeah, finds a lot of bugs Hi, great event.\n\nI have a question about foundational models for autonomous driving We have all seen that big models that really can... When you scale up with data and model parameter from GP3 to POM, it can actually now do reasoning Do you see that it's essential scaling up foundational models with data and size And then at least you can get a teacher model that potentially can solve all the problems And then you distill to a student model Is that how you see foundational models relevant for autonomous driving? That's quite similar to our auto labeling models So we don't just have models that run in the car We train models that are entirely offline that are extremely large that can't run in real time on the car So we just run those offline on the servers producing really good labels that can then train the online networks So that's one form of distillation of these teacher-student models In terms of foundation models, we are building some really, really large datasets that are multiple petabytes And we are seeing that some of these tasks work really well when we have these large datasets Kinematics, like I mentioned, video in, all the kinematics out of all the objects and up to the fourth derivative And people thought we couldn't do detection with cameras Detection, depth, velocity, acceleration And imagine how precise these have to be for these higher-order derivatives to be accurate And this all comes from these kind of large datasets and large models So we are seeing the equivalent of foundation models in our own way for geometry and kinematics and things like those Do you want to add anything, John? Yeah, I'll keep it brief Basically, whenever we train on a larger dataset, we see big improvements in our model performance And basically, whenever we initialize our networks with some pre-training steps from some other auxiliary tasks We basically see improvements The self-supervised or supervised with large datasets both help a lot Hi, so at the beginning, Elon said that Tesla was potentially interested in building artificial general intelligence systems Given the potentially transformative impact of technology like that It seems prudent to invest in technical AGI safety expertise specifically I know Tesla does a lot of technical, narrow AI safety research I was curious if Tesla was intending to try to build expertise in technical artificial general intelligence safety specifically Well, I mean, if we start looking like we're going to be making a significant contribution to artificial general intelligence Then we'll for sure invest in safety on big believer in AI safety I think there should be an AI sort of regulatory authority at the government level Just as there is a regulatory authority for anything that affects public safety So we have regulatory authority for aircraft and cars and sort of food and drugs Because they affect public safety and AI also affects public safety So I think, and this is not really something that government I think understands yet I think there should be a referee that is trying to ensure public safety for AGI And you think of like, well, what are the elements that are necessary to create AGI? Like the accessible dataset is extremely important And if you've got a large number of cars and humanoid robots processing petabytes of video data and audio data from the real world Just like humans, that might be the biggest dataset, probably is the biggest dataset Because in addition to that, you can obviously incrementally scan the internet But what the internet can't quite do is have millions or hundreds of millions of cameras in the real world Like I said, with audio and other sensors as well So I think we probably will have the most amount of data And probably the most amount of training power Therefore probably we will make a contribution to AGI Hey, I noticed the semi was back there, but we haven't talked about it too much I was just wondering for the semi truck, what are the changes you're thinking about from a sensing perspective? I imagine there's very different requirements obviously than just a car And if you don't think that's true, why is that true?\n\nNo, I think basically you can drive a car Think about what drives any vehicle, it's a biological neural net with eyes With cameras essentially What is your primary sensors are? Two cameras on a slow gimbal, a very slow gimbal That's your head So if a biological neural net with two cameras on a slow gimbal can drive a semi truck Then if you've got like eight cameras with continuous 360 degree vision Operating at a higher frame rate and a much higher reaction rate Then I think it is obvious that you should be able to drive a semi or any vehicle much better than human Hi, my name is Akshay, thank you for the event Assuming Optimus would be used for different use cases and would evolve at different speeds for these use cases Would it be possible to sort of develop and deploy different software and hardware components independently And deploy them in Optimus so that the overall feature development is faster for Optimus Okay, we did not comprehend Unfortunately our neural net did not comprehend the question Next question Hi, I want to switch the gear to the autopilot So when you guys plan to roll out the FSD beta to countries other than US and Canada And also my next question is what's the biggest bottleneck or the technology or barrier you think in the current autopilot stack And how you envision to solve that to make the autopilot is considerably better than human in terms of performance matrix Like safety assurance and the human confidence I think you also mentioned for the FSD V11 you are going to combine the highway and the city as a single stack And some architectural big improvements, can you maybe expand a bit on that, thank you Well, that's a whole bunch of questions We're hopeful to be able to, I think from a technical standpoint FSD beta should be possible to roll out FSD beta worldwide by the end of this year But for a lot of countries we need regulatory approval And so we are somewhat gated by the regulatory approval in other countries But I think from a technical standpoint it will be ready to go to a worldwide beta by the end of this year And there's quite a big improvement that we're expecting to release next month That will always be especially good at assessing the velocity of fast moving cross traffic And a bunch of other things So, anyone want to elaborate? I guess so, there used to be a lot of differences between production autopilot and the full self driving beta But those differences have been getting smaller and smaller over time I think just a few months ago we now use the same vision only object detection stack in both FSD and in the production autopilot on all vehicles There's still a few differences, the primary one being the way that we predict lanes right now So we upgraded the modeling of lanes so that it could handle these more complex geometries like I mentioned in the talk In production autopilot we still use a simpler lane model But we're extending our current FSD beta models to work in all sort of highway scenarios as well The version of FSD beta that I drive actually does have the integrated stack So it uses the FSD stack both in city streets and highway and it works quite well for me But we need to validate it in all kinds of weather like heavy rain, snow, dust And just make sure it's working better than the production stack across a wide range of environments But we're pretty close to that I think it's, I don't know, maybe, it'll definitely be before the end of the year and maybe November Yeah, in our personal drives, the FSD stack on highway drives already way better than the production stack we have And we do expect to also include the parking lot stack as a part of the FSD stack before the end of this year So that will basically bring us to, you sit in the car in the parking lot and drive till the end of the parking lot at a parking spot before the end of this year And in terms of the fundamental metric to optimize against is how many miles between a necessary intervention So just massively improving how many miles the car can drive in full autonomy before an intervention is required that is safety critical So, yeah, that's the fundamental metric that we're measuring every week and we're making radical improvements on that Hi, thank you, thank you so much for the presentation, very inspiring My name is Daisy, I actually have a non-technical question for you I'm curious, if you are back to your 20s, what are some of the things you wish you knew back then? What are some advice you would give to your younger self? Well, I'm trying to figure out something useful to say Yeah, a joint Tesla would be one thing Yeah, I think just trying to expose yourself to as many smart people as possible I don't read a lot of books You know, I did do that though So, I think there's some merit to just also not being necessarily too intense And enjoying the moment a bit more, I would say to 20-something me Just to stop and smell the roses occasionally would probably be a good idea You know, it's like when we were developing the Falcon 1 rocket on the Quageline Atoll And we had this beautiful little island that we were developing the rocket on And not once during that entire time did I even have a drink on the beach I'm like, I should have had a drink on the beach, that would have been fine Thank you very much I think you have excited all of the robotics people with Optimus This feels very much like 10 years ago in driving But as driving has proved to be harder than it actually looked 10 years ago What do we know now that we didn't 10 years ago that would make, for example, AGI on a humanoid come faster? Well, I mean, it seems to me that AGI is advancing very quickly Hardly a week goes by without some significant announcement And, yeah, I mean, at this point, like, AI seems to be able to win at almost any rule-based game It's able to create extremely impressive art Engage in conversations that are very sophisticated, you know, write essays And these just keep improving And there's so many more talented people working on AI And the hardware is getting better AI is on a super, like, a strong exponential curve of improvements Independent of what we do at Tesla And obviously we'll benefit somewhat from that exponential curve of improvement with AI Like, Tesla just also has to be very good at actuators Motors gearboxes, controllers, power electronics, batteries, sensors And, you know, really, like, I'd say the biggest difference between the robot on four wheels And the robot with arms and legs is getting the actuators right It's an actuators and sensors problem And obviously, how you control those actuators and sensors But it's, yeah, actuators and sensors and how you control the actuators I don't know, we have to have, like, the ingredients necessary to create a compelling robot And we're doing it, so...\n\nHi, Ilan You are actually bringing the humanity to the next level Literally, Tesla, and you are bringing the humanity to the next level So, you said Optimus Prime, Optimus will be used in next Tesla factory My question is, will a new Tesla factory be fully run by Optimus program? And when can general public order a humanoid? Yeah, I think it'll, you know, we're going to start Optimus with very simple tasks in the factory You know, like maybe just, like, loading a part, like you saw in the video You know, carrying a part from one place to another Or loading a part into one of our more conventional robot cells to, you know, that welds body together So we'll start, you know, just trying to, how do we make it useful at all? And then gradually expand the number of situations where it's useful And I think that number of situations where Optimus is useful will grow exponentially Like really, really fast In terms of when people can order one, I don't know, I think it's not that far away Well, I think you mean, when can people receive one? So, I don't know, I'm like, I'd say probably within three years And not more than five years Within three to five years, you could probably receive an Optimus I feel the best way to make the progress for AGI is to involve as many smart people across the world as possible And given the size and resource of Tesla compared to robot companies And given the state of humanoid research at the moment Would it make sense for the kind of Tesla to sort of open source some of the simulation hardware parts? I think Tesla can still be the dominant platformer where it can be something like an Android OS Or like an iOS stuff for the entire humanoid research Would that be something that rather than keeping the Optimus to just Tesla researchers Or the factory itself can open it and let the whole world explore humanoid research?\n\nI think we have to be careful about Optimus being potentially used in ways that are bad Because that is one of the possible things to do So I think we would provide Optimus where you can provide instructions to Optimus But where those instructions are governed by some laws of robotics that you cannot overcome So not doing harm to others and I think probably quite a few safety related things with Optimus We'll just take maybe a few more questions and then thank you all for coming Questions, one deep and one broad On the deep for Optimus, what's the current and what's the ideal controller bandwidth? And then in the broader question, there's this big advertisement for the depth and breadth of the company What is it uniquely about Tesla that enables that? Anyone want to tackle the bandwidth question? So the technical bandwidth of the... Close to your mouth and loud For the bandwidth question, you have to understand or figure out what is the task that you want it to do And if you took a frequency transform of that task, what is it that you want your limbs to do? And that's where you get your bandwidth from It's not a number that you can specifically just say you need to understand your use case And that's where the bandwidth comes from What is the broad question?\n\nThe breadth and depth thing, I can answer the breadth and depth On the bandwidth question, I think we probably will just end up increasing the bandwidth Which translates to the effective dexterity and reaction time of the robot It's safe to say it's not one hertz and maybe you don't need to go all the way to 100 hertz But maybe 10, 25, I don't know Over time, I think the bandwidth will increase quite a bit Or translate it to dexterity and latency You'd want to minimize that over time Minimize latency, maximize dexterity In terms of breadth and depth, I guess we're a pretty big company at this point So we've got a lot of different areas of expertise that we necessarily had to develop In order to make electric cars and then in order to make autonomous electric cars Tesla is like a whole series of startups basically And so far they've almost all been quite successful So we must be doing something right And I consider one of my core responsibilities in running the company Is to have an environment where great engineers can flourish And I think in a lot of companies, I don't know, maybe most companies If somebody's a really talented driven engineer, they're unable to actually Their talents are suppressed at a lot of companies And some of the companies that the engineering talent is suppressed In a way that is maybe not obviously bad But where it's just so comfortable and you paid so much money The output you actually have to produce is so low that it's like a honey trap So there's a few honey trap places in Silicon Valley Where they don't necessarily don't seem like bad places for engineers But you have to say like a good engineer went in and what did they get out And the output of that engineering talent seems very low Even though there seem to be enjoying themselves That's why I call it there's a few honey trap companies in Silicon Valley Tesla is not a honey trap that we're demanding and it's like You're going to get a lot of shit done and it's going to be really cool And it's not going to be easy But if you are a super talented engineer Your talents will be used I think to a greater degree than anywhere else You know, SpaceX also that way Hi Ilan, I have two questions So both to the autopilot team So the thing is like I have been following your progress for the past few years So today you have made changes on like the lane detection Like you said that previously you were doing instant semantic segmentation Now you guys are built transfer models for like building the lanes So what are some other common challenges which you guys are facing right now Like which you are solving in future as a curious engineer So that like we as a researcher can work on those Start working on those And the second question is like I'm really curious about the data engine Like you guys have like told a case like where the car is stopped So how are you finding cases which is very much similar to that from the data which you have So a little bit more on the data engine would be great I'll answer the first question using occupancy network as an example So what you saw in the presentation did not exist a year ago So we only spent one year on time We actually shipped more than 12 occupancy network And to have a one foundation model actually to represent the entire physical world Around everywhere and you always condition is actually really really challenging So only over a year ago we're kind of like driving a 2D world If there's a wall and if there's a curve we kind of represent with the same static edge Which is obviously you know not ideal right There's a big difference between a curve and a wall when you drive you make different choices right So after we realized that we have to go to 3D We have to basically rethink the entire problem and think about how we address that So this will be like one example of a challenges we have we have a conquer in the past year Yeah to answer the question about how we actually source examples of the tricky stopped cars There's a few ways to go about this but two examples are one we can trigger for disagreements within our signals So let's say that parked bit flickers between parked and driving We'll trigger that back and the second is we can leverage more of the shadow mode logic So if the customer ignores the car but we think we should stop for it we'll get that data back too So these are just different like various trigger logic that allows us to get those data campaigns back Hi Thank you for the amazing presentation thanks so much So there are a lot of companies that are focusing on the AGI problem And one of the reasons why it's such a hard problem is because the problem itself is so hard to define Several companies have several different definitions they focus on different things So what is Tesla how's Tesla defining the AGI problem and what are you focusing on specifically Well we're not actually specifically focused on AGI I'm simply saying that AGI is seems likely to be an emergent property of what we're doing Because we're creating the oldies autonomous cars and autonomous humanoids That are actually with a truly gigantic data stream that's coming in and being processed It's by far the most amount of real world data and data you can't get by just searching the internet Because you have to be out there in the world and interacting with people and interacting with the roads And just you know it's Earth is a big place and reality is messy and complicated So I think it's sort of like it just seems likely to be an emergent property If you've got tens or hundreds of millions of autonomous vehicles and maybe even a comparable number of humanoids Maybe more than that on the humanoid front Well that's just the most amount of data and if that video is being processed It just seems likely that the cars will definitely get way better than human drivers And the humanoid robots will become increasingly indistinguishable from humans perhaps And so then like I said you have this emergent property of AGI And arguably humans collectively are sort of a superintelligence as well Especially as we improve the data rate between humans The thing like that seems way back in the early days the internet was like the internet was like humanity acquiring a nervous system Where now all of a sudden any one element of humanity could know all of the knowledge of humans by connecting to the internet Almost all the knowledge or certainly a huge part of it Whereas previously we would exchange information by osmosis Like in order to transfer data so you would have to write a letter Someone would have to carry the letter by person to another person And then a whole bunch of things in between and then it was like Yeah I mean it's insanely slow when you think about it And even if you were in the Library of Congress you still didn't have access to all the world's information And you certainly couldn't search it and obviously very few people are in the Library of Congress So I mean one of the great sort of equality elements Like the internet has been the biggest equalizer in history in terms of access to information and knowledge And any student of history I think would agree with this Because you know you go back a thousand years there were very few books And books would be incredibly expensive but only a few people knew how to read And even a small number of people even had a book Now look at it like you can access any book instantly You can learn anything basically for free It's pretty incredible So you know I was asked recently what period of history would I prefer to be at the most And my answer was right now This is the most interesting time in history and I read a lot of history So let's do our best to keep that going And to go back to one of the earlier questions I would ask The thing that's happened over time with respect to Tesla autopilot is that the neural nets have gradually absorbed more and more software And in the limit of course you could simply take the videos as seen by the car And compare those to the steering inputs from the steering wheel and pedals Which are very simple inputs And in principle you could train with nothing in between Because that's what humans are doing with the biological neural net You could train based on video and what trains the video is the moving of the steering wheel and the pedals With no other software in between We're not there yet but it's gradually going in that direction Alright, one last question How are you going? I think we've got a question at the front here Hello, they're right there We'll do two questions, fine They're here Thanks for such a great presentation We'll do your question last Okay, cool With FSD being used by so many people How do you evaluate the company's risk tolerance in terms of performance statistics And do you think there needs to be more transparency or regulation from third parties As to what's good enough and defining thresholds for performance across many miles The number one design requirement at Tesla is safety And that goes across the board So in terms of the mechanical safety of the car We have the lowest probability of injury of any cars ever tested by the government For just a passive mechanical safety Essentially crash structure and airbags and what not We have the highest rating for active safety as well And I think it's going to get to the point where the active safety is so ridiculously good It's just absurdly better than a human And then with respect to autopilot We do publish broadly speaking the statistics on miles driven With cars that have no autonomy Tesla cars with no autonomy With hardware one, hardware two, hardware three And then the ones that are in FSD beta And we see steady improvements all along the way And sometimes there's this dichotomy of Should you wait until the car is three times safer than a person before deploying any technology But I think that's actually morally wrong At the point at which you believe that adding autonomy reduces injury and death I think you have a moral obligation to deploy it Even though you're going to get sued and blamed by a lot of people Because the people whose lives you saved don't know that their lives are saved And the people who do occasionally die or get injured Definitely know, or their state does, that there was a problem with autopilot That's why you have to look at the numbers in total miles driven How many accidents occurred, how many accidents were serious, how many fatalities And we've got well over three million cars on the road So that's a lot of miles driven every day And it's not going to be perfect But what matters is that it is very clearly safer than not deploying it Yeah So, I think, last question I think, yeah, thanks The last question here Okay, hi So, I do not work on hardware So maybe the hardware team and you guys can enlighten me Why is it required that there be symmetry in the design of Optimus? Because humans, we have handedness, right? We use some set of muscles more than others Over time there's wear and tear, right? So maybe you'll start to see some joint failures or some actuator failures more Over time, I understand that this is extremely pre-stage Also, we as humans have based so much fantasy and fiction Over superhuman capabilities Like all of us don't want to walk right over there We want to extend our arms and like we have all these, you know A lot of fantasy, fantastical designs So considering everything else that is going on In terms of batteries and intensity of compute Maybe you can leverage all those aspects into coming up with something Well, I don't know, more interesting in terms of the robot that you're building And I'm hoping you're able to explore those directions Yeah, I think it would be cool to have like, you know, make Inspector Gadget real That would be pretty sweet So, yeah, I mean, right now we just want to make basic humanoid work well And our goal is to pass this path to a useful humanoid robot I think this will ground us in reality, literally And ensure that we are doing something useful Like one of the hardest things to do is to be useful To actually, and then to have high utility under the curve Like how much help did you provide to each person on average And then how many people did you help? The total utility Like trying to actually ship useful product that people like To a large number of people is so insanely hard It boggles the mind You know, that's why I can say like, man, there's a hell of a difference between a company that has shipped product And one has not shipped product This is night and day And then even once you ship product, can you make the cost, the value of the output Worth more than the cost of the input Which is, again, insanely difficult, especially with hardware So, but I think over time I think it would be cool to do creative things And have like eight arms and whatever And have different versions And maybe, you know, there'll be some hardware Like companies that are able to add things to an optimist Like maybe we, you know, add a power port or something like that Or attach them, you can add attachments to your optimist Like you can add them to your phone There could be a lot of cool things that could be done over time And there could be maybe an ecosystem of small companies that, or big companies that Make add-ons for optimists So, with that, I'd like to thank the team for their hard work You guys are awesome And thank you all for coming And for everyone online, thanks for tuning in And I think this will be one of those great videos where you can like If you can fast forward to the bits that you find most interesting But we try to give you a tremendous amount of detail Literally so that you can look at the video at your leisure And you can focus on the parts that you find interesting and skip the other parts So, thank you all, and we'll do this, try to do this every year And we might do a monthly podcast even So, but I think it'll be great to sort of bring you along for the ride And like show you what cool things are happening And yeah, thank you Alright, thanks Thank you"},"languages":["en"],"lang":"en","transcriptSource":"https://gist.github.com/L0rdCha0s/de22ae0c7e7a7a70b37ac9c1262e27e1"},{"id":"tesla-battery-day-2020","type":"video","url":"https://www.youtube.com/watch?v=l6T9xIeZTds","title":"Tesla Battery Day","titles":{"en":"Tesla Battery Day","de":"Tesla Battery Day","fr":"Tesla Battery Day"},"date":"2020-09-22","summary":"Tesla's Battery Day: the new 4680 cell, structural battery pack and a roadmap to cheaper, longer-range electric cars, with Elon Musk and Drew Baglino.","text":"Al Prescott: (41:03) Good afternoon, everyone. Welcome to Tesla's 2020 Annual Meeting of Stockholders. We're really excited that you could be here with us today. My name is Al Prescott. I'm Tesla's vice president of legal. Al Prescott: (41:15) There'll be two parts of today's meeting. First, the former part of the meeting we'll get out of the way, which we'll cover the seven items that stockholders have been asked to vote on. After the voting, I'll introduce Tesla's co-founder and CEO, Elon Musk, who will give a presentation about the company update and year in review. And then following the conclusion of the stockholder meeting, we'll start our separate Battery Day event. Al Prescott: (41:40) At this time, I'd like to thank the members of the Tesla team and our board, especially those who were able to make it out here in person today, as well as to our representative from PricewaterhouseCoopers, Tesla's independent auditor who is also here. But before we begin, I'd like to introduce you to Robyn Denholm, the chairwoman of Tesla, who would like to say a few words remotely. Robyn Denholm: (42:12) Thank you, Al. Hello everyone and welcome to the 2020 Tesla Shareholder Meeting. A special welcome to the many Tesla shareholders that have joined us today in person as well as online from across the country and around the globe. Robyn Denholm: (42:29) I wanted to start today's proceedings by thanking you, our shareholders, for your tremendous support over the last year. And especially to those of you who have been with us through our journey over the past 10 years, since the company's IPO in 2010. While we have stayed true to our mission of accelerating the world's transition to sustainable energy, in many ways, our company has evolved beyond recognition over the past decade. And that is a great thing. In fact, the pace of developments and the evolution of Tesla has further accelerated over the past 15 months since I last addressed you in June of 2019. You'll hear more about many of the specific achievements from Elon later in the agenda. Robyn Denholm: (43:16) But I would like to take this opportunity to thank all of our Tesla employees across the globe who have done a tremendous job of executing and staying focused on delivering for our customers and shareholders, as the world has gone through one of the most challenging periods in our lifetimes. As a board, we have always taken a long-term view. We have made decisions and supported decisions made by the management team that may not have seemed obvious at the time, but are delivering and will continue to deliver breakthrough results. But it's also important to remember why we do this. As a company, we are focused on addressing one of the biggest environmental challenges of our generation, how to accelerate the world's transition to sustainable energy. Robyn Denholm: (44:08) The last year in particular has seen a tremendous increase in momentum in the movement to sustainable energy from both shareholders and the general public. So in addition to developing amazing clean transportation and energy products, we are doing our part by contributing the right facts and information to this important issue. And we released an extended version of our impact report in April of 2020. In issues version, we have covered in great detail, many areas that are important to our shareholders and our customers alike, such as our environmental impact, greenhouse and other noxious gas elimination, our supply chain efforts, especially in cobalt, and our culture and people focus. We hope that by continuing to put this data out there, we will underscore to the world the importance and impact that we are having as a company. Robyn Denholm: (45:09) Lastly, continuous feedback and input from our shareholders is essential for us to do our jobs. And I would like to thank you for your support in this regard. Many of you have provided me and the team with ideas and insights that we as a board take into consideration as we evolve our governance and company practices. It's especially crucial to the board members as we pride ourselves in adaptability and the diversity of thought and experience that we collectively represent on the board. Robyn Denholm: (45:41) This brings me to my final two things today, as today is his last shareholder meeting, on behalf of the board, I would like to sincerely thank Steve Jurvetson for over a decade of service to Tesla, the board, and our shareholders. You will be missed. Finally, I would like to introduce to you our newest member of the board, Hiro Mizuno, who until recently led the largest pension fund in the world. He brings a wealth of experience to the board, but let me hand over to Hiro to say a few words. Hiro ... Hiro Mizuno: (46:17) Thank you, Robyn. Ladies and gentlemen, welcome to Tesla Annual Shareholders Meeting. It is my real pleasure to virtually meet you, Tesla shareholders, people who believe in Tesla's mission and its growth opportunities. I spent all my career in finance and asset management in Tokyo, New York, London, and the Silicon Valley. Hiro Mizuno: (46:42) Until recently, I was a chief investment officer of GPIF $1.5 trillion Japanese public pension fund. And one of my priorities as the investment chief was to promote responsible investments, which aim to make financial returns while pursuing ESG agenda, such as environment and social issues. I believe in the market where ESG is becoming mainstream. Purpose or mission driven businesses will gain long-term investors support. Hiro Mizuno: (47:19) This is why I was interested in Tesla, where our mission is to accelerate the world's transition to sustainable energy. I'm very excited to join the Tesla team on the journey and hope that [inaudible 00:47:34] Tesla deliver what investors expect by further enhancing its environmental and social impact. Once again, Tesla shareholders, thanks for your support. I'm looking forward to seeing you in person next year. Thank you. Al Prescott: (47:54) Thanks Robyn and Hiro. I will now call the meeting to order. Please refer to the meeting agenda that has been provided to you and posted also to our virtual meeting site. The time is now 1:49 PM Pacific Time. And I declare that the polls are now open. Al Prescott: (48:12) We've already received voting proxies from stockholders over the past few weeks, meaning that almost all of the votes that will be counted were already submitted before the meeting. However, if you wish to vote now or to change your prior vote, you may do so through the virtual meeting site. For those that are here in person today, ballots and ballot boxes were available to you at check-in. Al Prescott: (48:37) Tesla's board of directors has appointed Computershare Trust Company to serve as inspector of elections for the meeting. Computershare has taken and signed an oath as inspector of election and has certified that starting on August 13th, 2020, the proxy material, or a notice of internet availability of the proxy material were mailed or provided to all Tesla stockholders of record as of July 31, 2020. Al Prescott: (49:04) We have a majority of the outstanding shares represented at the meeting. So I declare that there is now a quorum present and that we may proceed with the meeting. The items on the agenda are as follows; the election of three class one directors, Elon Musk, Robyn Denholm, and Hiromichi Mizuno to each serve for or term of three years. Two, to approve Tesla's executive compensation on an advisory basis. And three, to ratify the appointment of PricewaterhouseCoopers, LLP as Tesla's independent, registered public accounting firm for the fiscal year of 2020. Tesla's board has recommended that our stockholders vote for each of the director nominees and for each of those proposals. Al Prescott: (49:59) In addition, we have also received four stockholder proposals as described in the proxy statement. I would like to remind our stockholders that Tesla's board has prepared a statement in opposition to each of these proposals, which appear in the proxy. The first stockholder proposal is an advisory vote regarding paid advertising. Our board has recommended that our stockholders vote against this stockholder proposal. This stockholder proposal comes to us from James Danforth. Al Prescott: (50:33) However, Mr. Danforth has notified us that neither he nor his representative will be presenting the proposal at the meeting today. So we will continue. The second stockholder proposal is an advisory vote regarding simple majority voting and our governing documents. Our board has recommended that our stockholders vote against this stockholder proposal. The proposal comes from James McRitchie, who is on the line to present the proposal today. Mr. McRitchie, I would like to invite you now to present. You will have three minutes. James McRitchie: (51:14) I'd like to thank the board for holding such an innovative hybrid meeting during these difficult times. Proposal number five basically asks for a majority voting standard to amend bylaw. I first introduced a proposal on this subject at the 2014 Tesla meeting. Super majority provisions generally use to entrench incumbent directors and managers. Academic research finds that reducing such devices is associated with higher returns. James McRitchie: (51:46) The board's opposition statement argues they tried to adopt a [inaudible 00:51:51] party standard last year, but shareholders rejected it. However, 99.6% of shares voted for the proposal. Only 0.4% voted against it. The problem was that a little more than 35% of shares went unvoted. The vast majority of retail shareholders often don't bother to vote. Since only 65% of shares were voted, we didn't achieve the 66.67% necessary to overturn the current super majority bylaw. James McRitchie: (52:32) It appears the proposal failed primarily for three reasons. One, the board put forth less than robust arguments in favor. Two, they added confusion with another proposal to reclassify the board, not into a single class, that's the norm, but into two classes, elected in altering years. Third, the board also failed to make a substantial effort to solicit votes in favor. Also, please consider this proposal in context with other poor corporate governance provisions at Tesla. First, shareholders can only remove directors for cause. What that basically means is the director has to be caught in criminal activity for shareholders to remove them. Second, because the board is divided into three classes, shareholders can only hold individual directors accountable every three years. And third, shareholders cannot call special meetings, nor can they act by written consent. I hope you will agree. Corporations should not be democratic-free zones. Vote for proposal number five so that 33% of shares cannot overrule the wishes of 67%. Thank you. Al Prescott: (53:54) Thank you, Mr. McRitchie. We'll now move on to our third stockholder proposal, which is an advisory vote regarding reporting on employee arbitrations. Our board has recommended that our stockholders vote against this stockholder proposal. This proposal comes from Nia Impact Capital, whose representative Kelly Hull is on the line to present the proposal today. Ms. Hull, I'd like to invite you to go ahead and present. You will have three minutes. Dr. Kristin Hull: (54:28) Hello. My name is Dr. Kristin Hull, and I'm the founder and CEO of Nia Impact Capital. I formally move [inaudible 00:20:36]. This resolution requests that Tesla board of directors overseeing the preparation of a report on the impact of the use of mandatory arbitration on Tesla's employees and on its work place culture. The report will evaluate the association of Tesla's current use of arbitration with the prevalence of both harassment and discrimination in its workplace and on employee's ability to [inaudible 00:21:01], should harassment or discrimination occur. Dr. Kristin Hull: (55:05) This proposal speaks to the widespread experience of discrimination in the workplace by Black, Latinx, and female employees, despite this discrimination being unlawful under the Civil Rights Act of 1964. Tesla has faced a number of serious allegations of racism and sexism at its Buffalo and Fremont plant. Companies that allow bias discrimination and harassment in their workplaces are at risk for unnecessary legal brand financial and human capital issues. Dr. Kristin Hull: (55:36) Support of this resolution is warranted for the following five reasons. One, research shows that companies benefit from diverse and inclusive workplaces. Two, corporate policies that allow harassment and discrimination risk investors capital. Three, the use of arbitration exposes investors to an unknown level of risk. Four, broad concerns exist with respect to fair treatment in Tesla workplace. And Tesla employees have alleged harassment and discrimination on their basically both race and gender. [inaudible 00:56:13] Tesla, a company investors love for its innovation, leadership, and [inaudible 00:56:18] is increasingly lagging behind its peers in its [inaudible 00:56:22] related to workplace diversity, equity, and inclusion. Dr. Kristin Hull: (56:26) Unlike the forward thinking and innovation in its extraordinary product lines, Tesla has not challenged proactive leadership and building a positive company culture or in addressing concerns about its workplace practices. In these material issues, Tesla lags behind its technology and automotive competitors. Dr. Kristin Hull: (56:48) The use of arbitration limits employees remedy for wrongdoing, precludes employees from stewing in court, and often keeps underlying facts, misconduct or case outcomes secret, therefore preventing employees from learning about and acting on shared concerns. Dr. Kristin Hull: (57:05) Simply stated, arbitration allows that corporate behavior like bias, harassment, and discrimination to continue to keep hidden from employees and investors. To maintain Tesla's [inaudible 00:57:17], it is essential that the board seriously assess the implications of the use of arbitration and that Tesla begins to seriously the need to ensure a fair, equitable, positive, and inclusive workplace. Thank you. Al Prescott: (57:34) Thank you, Ms. Hull. Our fourth and final proposal is an advisory vote regarding reporting on human rights. Our board has recommended that stockholders vote against this proposal. This proposal comes to us from the Sisters of Good Shepherd, New York province, whose representative, Terrence Collingsworth is on the line to present today. Mr. Collingsworth, I would like to invite you to speak now. You have three minutes for your proposal. Terry Collingsworth: (58:09) Thank you. I'm Terry Collingsworth, executive director of the International Rights Advocates. I'm here representing the Sisters of Good Shepherd New York province to present item seven on human rights disclosure, which calls upon Tesla to issue a report to describe board oversight of human rights and its human rights due diligence process, including systems to provide meaningful remedies when human rights impacts occur. Terry Collingsworth: (58:40) Tesla faces serious human rights issues and failure to establish a culture of respect for human rights will expose Tesla to new liability issues and significant reputational injury, all of which will have a material impact on the company and its shareholders. The need to set a new course for human rights compliance at Tesla is glaring. Terry Collingsworth: (59:05) Here are five examples of human rights violations occurring now in Tesla's operations: racism, sexual harassment, and disregard for human safety and dignity harm workers at the Gigafactory 2 in Buffalo, New York, every single day. And those workers urge you to remember their experiences in your vote. Terry Collingsworth: (59:28) Tesla is experienced serious labor relations issues at its production facilities and is actively discouraging union organizing. Workers are being exposed to COVID-19 and then are facing retaliation when they ask for greater protections. There are numerous worker health and safety violations as well as wage and hour issues. Terry Collingsworth: (59:50) And finally, there are serious, even deadly, human rights violations occurring in Tesla's global supply chains. On this last issue, my organization brought the pending suit against Tesla for using cobalt mined in the Democratic Republic of Congo by young children. I personally met young boys who lost limbs or were paralyzed in cobalt tunnel collapses. Tesla sources cobalt from these very mines. And its claimed to have quote, \"Zero tolerance for child labor,\" in its supplier code of conduct is simply not true. Tesla is not only tolerating child labor in its cobalt supply chain, it is tolerating the death and maiming of young child minors. Terry Collingsworth: (01:00:42) This demonstrates why the company must circle back and begin a process to report on its treatment of human rights issues as requested in this proposal. I think consumers will have zero tolerance for a company that is exposed as being indifferent to killing and maiming child minors. We are hopeful that Tesla's innovative spirit can be brought to bear on making human rights a priority at the company. Terry Collingsworth: (01:01:12) For example, if the Elon Musk cared about implementing a zero tolerance child labor policy, instead of having a useless paper policy, Tesla could employ satellites or drones at every mine it sources from to actually monitor child labor. I encourage all Tesla shareholders to vote for item seven, human rights disclosure. Thank you for your attention. Al Prescott: (01:01:40) Thank you Mr. Collingsworth. At this time, I'd like to thank our stockholders for all of their active participation in today's meeting and for those who just presented on the line. I'd also like to read some of the comments that have been submitted by you over the course of the meeting. The first comment comes from Michael [Overbaugh 00:01:02:01]. \"I take great pride in the fact that we haven't had to stoop to the level of what advertising represents to get where we are today. I'd hate to give into that kind of temptation now, when we're so close to becoming a household name that's based solely on our merit alone. But if assets do end up having to be set aside for marketing, I'd like to suggest that rather than shoving ads down the customer's throat, we established some sort of hardcore nationwide campaign and event with the goal of getting as many people as possible behind the wheel of a Tesla for an introduction drive. It's well-known how far just doing that alone goes to converting people into fans.\" Al Prescott: (01:02:46) \"A line I recently ran across says, 'You can talk all about the specs as much as you want, but when it comes to buying a car, what ultimately puts butts in seats is the feeling that the vehicle gives you.' By demonstrating that Tesla clearly has both the specs and the feeling, what more needs saying?\" Al Prescott: (01:03:08) Our second comment comes from the United Steel workers on behalf of the Clean Air Now Coalition of Western New York by Sabrina Lu. And it reads as follows, \"Proposal six and seven up for vote this year are the results of widespread concern about mistreatment of Tesla workers at US factories and across the supply chain. It is clear that Tesla is not interested in addressing the harm they have caused to their workers as their board is advising shareholders to vote against the proposal. We're urging all shareholders to vote in favor of proposal six and seven. And on behalf of our workers at the United Steelworkers here in Western New York and for Tesla employees across the country and across the global supply chain, while this doesn't repair the harm, that's already been caused to countless employees, nor repair harm to children and communities forced into slave labor in the DRC, they represent steps towards a more just workplace at Tesla.\" Al Prescott: (01:04:19) This concludes all of the comments. Thank you all for your participation in the comments. We'll now have a final opportunity for any of you to submit proxies in order for them to be counted. So I'll pause and wait for a moment for you to do that. Al Prescott: (01:04:51) Okay. I declare that the polls are now closed. So based on the proxies that we have previously received, I'd like to announce on a preliminary basis that our stockholders have approved the recommendations of Tesla's board on all agenda items, other than the stockholder proposal for an advisory vote regarding simple majority voting in our governing documents. After the final tabulation is completed, we'll formally announce the results of the voting by filing a form 8-K with the SEC within four business days of today. This now concludes the official business of Tesla's 2020 annual stockholders meeting, which is now adjourned. Al Prescott: (01:05:37) Next, we will have a company update and a year in review presented by Elon. And then we will start our Battery Day Event. During the course of those following sessions, we may discuss our business outlook and make forward looking statements. Such statements or predictions based on our current expectations. Actual events or results could materially differ due to a number of risks and uncertainties, including those disclosed in our most recent 10-Q filed with the SEC. These forward looking statements represent our views. As of today. They shouldn't be relied on after today and we disclaim any obligation to update them after today as well. Al Prescott: (01:06:23) We will now continue with the company update and year in review. And it's my pleasure to introduce Tesla co-founder and CEO, Mr. Elon Musk. Elon Musk: (01:06:45) Everyone. Well, I mean, this is definitely a new approach. We've got the Tesla drive in movie theater, basically. It's good to see everyone. It's a little hard to read the room with everyone being in cars, but it's the only way we can do it. So hopefully it's cool. And hopefully you can hear me. Can you guys hear me? Elon Musk: (01:07:08) Okay. All right. Great. Elon Musk: (01:07:12) Well, thanks for coming. I think it's been an incredible year and I'd like to just thank you for your support through tough times, good times. It's been great. Really appreciate everyone who's put their heart and money into Tesla and I think it's worked out pretty well. This has been a good year. And I think there's many good years to come. So I'll go through the shareholder presentation fairly quickly because the real main event here is Battery Day. And really, I'm just going through a recap of what's happened over the past a year or so. Elon Musk: (01:07:58) I think starting from in terms of our ability to create a ... Elon Musk: (01:08:03) In terms of our ability to create a factory, huge kudos to the Tesla Shanghai team for being able to go from literally a dirt pile to volume production in 15 months. It's like, damn. Yeah. And I think something that's really quite noteworthy here is Tesla's the only foreign manufacturer to have a hundred percent owned factory in China. So this is often not well understood or not appreciated, but to have the only hundred percent owned foreign factory in China is a really big deal, and it's paying huge dividends here. So we really wouldn't have the results that we have had this year without the great efforts of the Tesla China team, so I'm super appreciative of that, and we'll see the Shanghai factory continue to scale quite a bit from where it is right now. I think we really could expect that to be, over time, a factory that produces over a million vehicles a year. Elon Musk: (01:09:16) Yeah, it's cool. So let's see. So we also reached in the past year of volume production of the Model Y, and this was the smoothest launch that we've ever had, so I think we're definitely getting better at a new vehicle launches and building factories and scaling production. As you've heard me say before, the hardest thing is scaling production, especially of a new technology. It's insanely difficult. Making a prototype is relatively easy, and if I think, like, what is the real achievement of Tesla in sort of car company terms, it's like it wasn't making sort of exciting prototypes. It was that Tesla was really the first company in about a century in the U.S., the first U.S. company in the U.S. to reach volume production and be sustainably profitable. The crazy thing is this has really not happened in a hundred years. That's the actual super hard part, and we now have four vehicles in volume production, S3XY. Also, the toughest joke I think maybe ever. It was a very difficult joke to make. Elon Musk: (01:10:38) So we also introduced the lowest cost solar in the U.S. It's only a dollar 49 a watt, and we really just simplified the whole value chain, so reduced sales and advertising, got rid of a bunch of unnecessary costs, and really are just relying upon the fact that it's just the lowest cost, most efficient solar in the U.S., providing both a retrofit and the solar glass roof, which I think is a really great product. A hard product to make work, but it will be a major pipeline in the future. Elon Musk: (01:11:13) And we also got four consecutive quarters of gap profitability, which was very difficult. Yeah. And certainly a testament to the hard work of people at Tesla. I mean, to do this in extremely difficult times against a wide range of adverse circumstances was insanely hard, but we got it done, and I think the future is looking I think, very promising from a sort of an annual profitability standpoint. So in order to sort of do well financially, you really need economies of scale, and you need ideally the best technology, and I think we've had the best technology for a while, but now we are also achieving economies of scale, and we're also rapidly improving autonomy, which is a massive value add to each car. So, I think the value of Tesla is going to be like total, just on the vehicle side, total vehicles produced times the value of autonomy. That's a way to think about the future value of Tesla. Elon Musk: (01:12:35) We also have consistent free cashflow generation. This is really important for growth, and a key element here is tightening up the time from when a car is ordered to when it is built and delivered. So for a company that is growing rapidly, it's extremely important to tighten the supply chain and to have, from when parts arrive, put it into a car very quickly and deliver the car very quickly to the customer. And if you can do that inside soft of your payables timeline, then the faster you grow, the more cash you have. Or conversely, if you're unable to do it within your payables timeline, the faster you grow, the less money you will have, which is obviously bad for capital intensive situation. So just tightening up and having the parts move very quickly to the factory, put it in a car, get it to a customer makes a massive difference to cashflow generation. Elon Musk: (01:13:34) I mean, that's why it's extremely important to have a factory in each continent, because if you don't at least have a factory in the continent, it isn't impossible to achieve this. So having a factory in China, that's able to serve China, and then soon many other countries in the region will be key to us tightening that total sort of chain of cashflow, and essentially the faster we grow, the more cash. This is really important. That's also why it's important to have Giga Berlin complete, because then we'll have a factory in China, a factory in the U.S. and soon a second factory in the U.S. in Austin, and a factory in Europe. Elon Musk: (01:14:18) I mean, even if for Giga Texas in Austin, even if we had exactly the same cost as in California, it would still be advantageous to do it there because it's roughly two-thirds of the way across the U.S., so in terms of delivering cars to the central U.S. and to the East Coast, it's just faster, it costs less, and it fundamentally improves our economics. So I think this is also maybe something that's not fully appreciated of just how important it is to have a factory at least on the continent or reasonably close to where the end customers is, so you can tighten that whole chain. Elon Musk: (01:14:56) Industry performance. While the rest of industry is, has gone down, Tesla has gone up, I think this speaks to ... Thanks. And so I'd like to thank all the customers for taking a chance on Tesla and buying our product and really hope you're enjoying it. This is really, our sales, as [inaudible 01:15:21] was saying, it really grew by word of mouth, so this is really, I think it's very pure in the sense that it's growing on the basis of existing owners recommending it to others to new customers. This is, really, I think, a good way to grow. Elon Musk: (01:15:40) So, and then in 2019 we had 50% growth, and I think we'll do really pretty well in 2020. Probably somewhere between 30-40% growth, despite a lot of very difficult circumstances. I mean, there's so many. Pandemic, the wildfires. It's a whole bunch of difficult production issues, but thanks to the hard work of the Tesla team and a lot of innovative approaches to overcoming issues, we're able to still see significant growth in one of the most difficult. In fact, I'd say probably the most difficult year of Tesla's existence. Elon Musk: (01:16:25) We also published our extended impact report. At Tesla, we try very hard to do the right thing. If what I think does not happen, it's just because we maybe made a mistake or weren't aware of it, but we always try to do the right thing to the best of our ability, and then we published the extended impact report to show just a self-examination of, okay, what are we doing, right? What are we doing wrong? What can we do better in the future? We're definitely trying to accomplish the most good, and so if we occasionally make a mistake, we work quickly to fix it and do the right thing. So it's worth looking at the average life cycle of emissions in the U.S. and just how much better a Tesla is or electric car than any kind of gasoline car, and what we'll talk about in the Battery Day is also just how much the grids around the world, and actually especially in the U.S., are greening. It's actually much faster than I think people realize, the U.S. is moving towards sustainable energy. And so as we move more and more to sustainable energy, then effectively you end up building the solar factories and the car factories themselves with solar or with sustainable energy. Over time, you will even mine with sustainable energy, and eventually it will get to an effective emissions of zero, so that's where things will end up. Yeah. Elon Musk: (01:17:59) So we also have safety at the core of our design. The Tesla cars are the safest cars ever designed. We have the lowest probability of injury of any cars ever tested by the U.S. government, And that's just passive safety. When you add active safety into that, it's even better, so it's really ... If safety is important to you, which obviously it is, the safest car you could drive is a Tesla. So I think some people aren't aware of this, but it's really safety is paramount. It is actually the number one design objective when we build a Tesla is safety. Elon Musk: (01:18:41) Our factories are also becoming safer, and if you look at the sort of accidents per vehicle, total vehicle made it's dramatically better than in the past, and it's already better than industry average, and we're confident we can get it to the best in the auto industry. Autopilot functionality continues to improve, and you can see it in the safety report that we publish every quarter. It's just getting better and better. The U.S. average for collisions is at roughly 2.1 per million miles, and with autopilot engaged, it's 0.3. I mean, this is a profound difference, really massive, and this will get even better. So we're confident that over time we can get the probability of an accident, especially the probability of injury, to 10 times better than the industry average, like an order of magnitude better. So that's just a lot of lives saved and a lot of injuries avoided, so that's a huge priority for us. Elon Musk: (01:19:50) Yeah, the autopilot front, I think it's hard for people to judge the progress of autopilot. I'm driving ... As a matter of course, I've always done this. I drive the bleeding edge alpha build of autopilot, and so I sort of have insight into what is going on. Previously about a couple of years ago, we were kind of stuck in a local maximum, so we're improving, but the improvements kind of started tailing off and just not getting where they needed to be. I call this sort of getting trapped in a local maximum, and so we had to do a fundamental rewrite of the entire autopilot software stack and all of the labeling software as well. Elon Musk: (01:20:40) So we are now labeling in 3D video, so this is hugely different from the previously where we were labeling essentially a bunch of single images from the eight cameras, and they would be labeled at different times by different people, and some of the labels, you literally can't tell what it is you're labeling. So it basically made it sort of in some cases impossible to label, and the labels had a lot of errors. Now with our new labeling tools, we label it in video, so we actually label entire video segments in the system, so you get basically a surround video thing to label with the surround video and with time. So it's now taking all cameras simultaneously and looking at how the image has changed over time and labeling that, and then the sophistication of the neural nets in the car and the overall logic in the car has improved dramatically. Elon Musk: (01:21:44) I think we'll hopefully release a private beta of autopilot, of the full self-driving version of autopilot in, I think, a month or so, and then people will really understand just the magnitude of the change. It's profound. So, yeah. Anyway, so you'll see it. It's just like a hell of a step change, but because we had to rewrite everything, labeling software, just the entire code base, it took us quite a while. The sort of new ... I call it like 4D in the sense that it's three dimensions plus time. It's just taken us a while to rewrite everything, and so you'll see what it's like. It's amazing. Yeah. It's just clearly going to work. Elon Musk: (01:22:42) At Tesla, the core competencies, we've got engineering, obviously, but also manufacturing. I think manufacturing is underappreciated in general, and the difficulty of designing the machine that makes the machine is vastly harder than the machine itself. So the designing, like making a Model 3 or Model Y or Cybertruck truck prototype is really quite trivial compared to designing the factory that makes it, especially if it's new technology, and you want to use new manufacturing methods. It's just at least 10 to 100 times harder to do the factory than the prototype, and that's why you see a lot of companies out there or startups they'll bring out a prototype, but they just can't get it over the hump for who manufacturing, because manufacturing of new technology especially is the hardest thing by fa. Basically, the prototype is at best 10% of the difficulty and probably closer to 1%. Elon Musk: (01:23:50) And then software. Tesla is both a hardware and a software company, so a huge percentage of our engineers are actually software engineers, and you can think of our car as kind of like a laptop on wheels, so software is incredibly important. Actually, not just in the car, but also in the factory. So the factory software is extremely important. Just software in general. I mean, these are fundamental. These are the three critical areas that are needed to make for an awesome company. So, yeah. Elon Musk: (01:24:29) So we have ... Now we'll soon have three new factories incremental on ... Well, we have one already. On three different continents. Shanghai, we're expanding the Shanghai with the second phase. Berlin is making rapid progress, and Texas is making even faster progress. So, yeah. With each factory, what we're trying to do is also improve the manufacturing technology, so in some cases like the Model Y made in Berlin might look the same, but it actually is made in a much more efficient way. Yeah, we'll talk about that later in the battery presentation. Elon Musk: (01:25:15) Yeah, we launched Megapack. It's three megawatt hours all in one energy storage solution, so it's been great overall. Yeah. All right. And I think that's basically it, right? All right, thank you. All right. Well, thanks, everyone, for coming, and we'll be back in a little bit to go through the battery stuff, and there's a little bit more. In addition to the battery stuff, we've got a few extras as well. So I think you'll really like what we have to say on batteries. Elon Musk: (01:25:53) The battery stuff we're going to talk about is truly revolutionary and essential to Tesla's goal. The fundamental good of Tesla, it's like, if you look back in history and say, \"What good did Tesla do?\" The good will by how many years did we accelerate sustainable energy? That's the true metric of success. It matters if sustainable energy happens faster or slower, and so that's really how I think about Tesla and how we should assess our progress. By how many years did we accelerate sustainable energy? And what we're going to talk about with batteries and a few other things will really explain how we're going to make a step change improvement in the acceleration of sustainable energy. Thank you. [inaudible 00:18:44]. Speaker 1: (01:26:50) Hi, folks. That was great. We're going to take a short break before we begin the Battery Day event, so stay tuned. If you're local and here in the audience today, you can feel free to get out of the cars and stretch your legs, but try to stay near the cars, because we're going to begin properly in a little bit. See you soon. (silence). Elon Musk: (01:27:06) [inaudible 01:40:28]. Drew Baglino: (01:40:32) Hello, everyone. Elon Musk: (01:40:37) Great. Should you start? Drew Baglino: (01:40:38) Sure. Thanks, Elon. Hi. I'm Drew Baglino, SVP of Powertrain and Energy Engineering at Tesla, and I'm incredibly excited to talk about what we've been doing with batteries here at Tesla. Elon Musk: (01:40:48) Great. So let's see. You've got the clicker? Drew Baglino: (01:40:53) I've got the clicker, yeah. Elon Musk: (01:40:54) Okay. Let's ... Yeah. I'll take it at first, perhaps. Drew Baglino: (01:40:57) Sure. Elon Musk: (01:40:58) So obviously the issues we're facing are very serious with climate change, and we're experiencing these issues on a day-to-day basis. It's incredibly important that we accelerate the advent of sustainable energy. Time really matters. This presentation is about accelerating the time to sustainable energy. Elon Musk: (01:41:23) So the past five years were the hottest on record. We have what looks like a wall for CO2 PPM. It's obviously ... This time is not like the past. It's really important that we take action. Running this climate experiment is insane, so ... Drew Baglino: (01:41:46) Especially when it's just a transitory one, anyway. Elon Musk: (01:41:49) Yes. Drew Baglino: (01:41:50) We're going to run out of these fossil fuels. Let's just move to the future and not run this experiment any longer. Yeah. Elon Musk: (01:41:55) Talk a bit louder. Drew Baglino: (01:41:56) You got it. Elon Musk: (01:41:57) Okay. So anyway, there is a lot of good ... Elon Musk: (01:42:03) There is a lot of good news though. A lot of people may not be aware that that wind and solar comprise 75% of new electricity capacity in the US this year. So this is really major. So the grid is going sustainable very quickly. Now, it's also worth noting that the length of time that power plants lasts is on the order of 25 years. So even if a hundred percent of energy generation was sustainable, it will still take 25 years to convert the grid. And it's also worth noting that in the past 10 years, power production from coal has dropped in half. So it went from 46% of electricity in 2010 to 23% in 2020. So this is a massive improvement. So good things are happening on a lot of levels. We just need to go faster. Elon Musk: (01:43:06) So Tesla's contribution, we've delivered over a million electric vehicles, 26 billion electric miles driven, and many gigawatt hours of stationary batteries, 17 terawatt hours of solar generated. So I think solar is sometimes underweighted at Tesla, but it is a massive part of our future. The three parts of a sustainable energy future are sustainable energy generation, storage, and electric vehicles. So we intend to play a significant role in all three. So to accelerate the transition to sustainable energy, we must produce more EVs that need to be affordable and a lot more energy storage, while building factories faster and with far less investment. So goal number one is a terawatt hour scale battery production. So tera is the new giga. And a terawatt is a thousand times more than a gigawatt. So we used to talk in terms of gigawatts, in the future, we'll be talking in terms of terawatt hours. So this is what's needed in order to transition the world to sustainability. Drew Baglino: (01:44:24) Yeah, and you can see it's... We're talking about a hundred X growth in batteries for electric vehicles to achieve this mission. And we are going to get there. It's just a matter of how fast. And our intention is to accelerate it. Elon Musk: (01:44:38) Yeah, you basically need on the order of roughly 10 terawatt hours a year of battery production to transition the global fleet of vehicles to electric. Drew Baglino: (01:44:48) And the average vehicle lasts 15 years. So we're talking about 150 terawatt hours give or take to transition the whole electric, all vehicles of all types, to electric. Elon Musk: (01:45:00) Yeah. So it's a lot of batteries, basically. And so- Drew Baglino: (01:45:07) Yeah. And then on the grid side, we have a similar mountain to climb, 1600 times growth from today's grid batteries to go a hundred percent renewable on the grid and to take all of the existing heating fossil fuel uses in homes and businesses, a hundred percent electric. Elon Musk: (01:45:24) Yeah. And this number I think might grow even more. As the world economy matures, and as countries with high populations industrialize, we could see this number be even more. But let's say it's like roughly 20 to 25 terawatt hours per year sustained for 15 to 25 years to transition the world to renewable. This is a lot. Drew Baglino: (01:45:53) Yeah. Elon Musk: (01:45:55) So today's batteries cannot scale fast enough. They're just too small. For Giga Nevada, 150 gigawatt hours per year is what we probably expect to make out of there. But this is really pretty small in the grand scheme of things. That's only 0.15 terawatt hours. And it costs too much. Drew Baglino: (01:46:16) We would need 135 fully built out in Nevada Giga factories to achieve 20 terawatt hours a year. It's not scalable enough of a solution. We need a dramatic rethink of the cell manufacturing system to scale as fast as we can and should. Elon Musk: (01:46:32) Yeah, and I think we should view this as more than just a question of money. Money is sort of an ethereal thing, but it's really the amount of effort. You have a certain amount of effort in terms of people and machines, and depending on how efficient that effort is, for a given amount of effort, you want the most amount of batteries. So it's not just the question of well, if we have $2 trillion, tomorrow you could make this. It's not that easy. You actually need to organize a massive number of people, build a lot of machines, build the machines that make the machines. And so it's incredibly important to have that effort yield the most number of batteries. Elon Musk: (01:47:16) So, and then goal two, obviously we need to make more affordable cars. I think one of the things that troubles me the most is that we don't yet have a truly affordable car, and that is something that we will make in the future. But in order to do that, we've got to get the cost of batteries down. We've got to make, and we've got to be better at manufacturing. And we need to do something about this curve. The curve of the cost per kilowatt hour of batteries is not improving fast enough. So we've given this a lot of thought over many years to say, okay, how can we radically improve the cost per kilowatt hour curve? It's been somewhat flattening out actually in recent years. Drew Baglino: (01:48:02) Yeah. I mean, early growth was promising, but you can see we're kind of plateauing. So that's what's motivating us to rethink how cells are produced and designed. Elon Musk: (01:48:10) Yeah, exactly. So yeah. And EV market share is growing, but EVs still aren't accessible to all. And you can see, as you Drew were saying, it's like starting to flatten out a little bit because the rate of improvement of the affordability of cars is just not fast enough. So that's why we've got Battery Day. Drew Baglino: (01:48:33) Yeah. To make the best cars in the world, we designed vehicles in factories from the ground up. Next. And now we do this for batteries as well. Elon Musk: (01:48:45) Yeah. It's weird, the slides don't show up quite right here. What shows up on the screen is not quite what shows up there. Drew Baglino: (01:48:55) Oh, okay. Elon Musk: (01:48:56) It's different. Drew Baglino: (01:48:57) Yeah. I think it's because that's... Yeah. Elon Musk: (01:48:59) That one's current, supposed to be current. Whatever. Drew Baglino: (01:49:02) So let's get started. We have a plan to have the cost per kilowatt hour. And it's not a plan that rests on a single innovation, some research project that will never see the light of day. It's a plan that has taken creative engineering and industrialization across every facet of what makes a cell into a battery pack, from raw material to the finished thing. And we're going to go through that plan with you today, step-by-step, and build up how we get to these goals and how we accelerate this transition and make our vehicles and our grid batteries more affordable. Elon Musk: (01:49:45) Yeah. I mean, we basically thought through every element of the battery, or almost every element. There are a few more elements that we won't get to today, but we will get to in the future. Drew Baglino: (01:49:53) Yes. So first before we get too far into it, let's talk about what is in a battery cell. We've got the cap and the can, negative and positive terminals of the cell. When you open that cell, you've got a tab connected to those terminals, what we call the jelly roll, which is the wound electrodes on the inside. You can actually see what this looks like as you unwind it. This is over a meter long in a typical 2170 cell. So it's quite a long winding process. And you can see the tab still there. And then to explain what's actually going on here, we've identified, we've got anode, cathode, separator, positive and negative terminal. Drew Baglino: (01:50:37) Watch what happens as we, there we go, discharge the cell. Got lithium moving from anode to cathode. And then the reverse when we charge the cell, lithium moving from cathode to anode across the separator. This is the basic of what makes all lithium-ion batteries, no matter what the form factor is. And when we look at what's happened today, at least in our products, we've moved from the 18650 form factor to the 2170 form factor through great collaboration with our partners, Panasonic, new partners like LG and CATL and probably others in the future. Elon Musk: (01:51:20) Actually, slight note on why is the one called 18650, although not on the slide, versus the 2170, is that the first two digits refer to the diameter, and the second two digits refer to the length. So that helps explain what's up with these weird numbers. But nobody could explain to me why there was an extra zero. So I, so I said, \"Okay, well, we're deleting the zero that nobody can explain in future form factors.\" So that's why it's technically, it's like the 18650 bizarrely, but going forward it's the 2170, because we just got rid of the extra zero because it's pointless. Drew Baglino: (01:51:56) And this was a evolutionary step going from 1865 to 2170, bringing 50% more energy into the cell. But when we look to the ideal cell design, if we were to do it ourselves, we need to go beyond just what we're looking at us in front of us and study the full spectrum of options. So as you can see, we kind of swept the key figures of merit, how much we can reduce the cost and how much vehicle range increases as we change the outer diameter of the cell. We found a sweet spot somewhere around 46 millimeters. But it's not just about a bigger form factor. Anybody could make a bigger form factor. Elon Musk: (01:52:37) Any fool, any fool could make a bigger form factor. Are we not any fool? Drew Baglino: (01:52:42) Yeah, exactly. There are problems as you make cells larger. In fact, supercharging and thermals in general become really challenging as you make bigger cells. And this was the challenge that our team set our sights on to overcome. And we did, we came up with this tabless architecture that maybe you've heard about, that basically removes the thermal problem from the equation and allows us to go to the absolute lowest cost form factor and the simplest manufacturing process. And this is what we mean when we talk about tabless. It's kind of a beautiful thing. Elon Musk: (01:53:22) Yeah. That's what these t-shirts mean, but it's very esoteric. It was like, nobody could figure it out. Drew Baglino: (01:53:26) Yeah, we basically took the existing foils, laser pattered them, and enabled dozens of connections into the active material through this shingled spiral you can see with simpler manufacturing, fewer parts, 50 millimeter versus 250 millimeter electrical path length, which is how we get all the thermal benefits. Elon Musk: (01:53:46) Yeah. This is important to appreciate. Basically the distance that that electron has to travel, it's just much less. So you actually have a shorter path length in a large tabless cell than you have in the smaller cell with tabs. This is a big deal. So even though the cell is bigger, it actually has more power. The power to weight ratio is actually better than the smaller cell with tabs. This is, again, this is quite hard to do it. Nobody's done it before and it really took a tremendous amount of effort within Tesla engineering to figure out how do we make a frigging tabless cell and have it actually work and then connect that to the top cap. There's a whole bunch of things that we're keeping a little secret sauce here that we're not telling everything, but- Drew Baglino: (01:54:40) Sometimes what's elegant and simple is still hard. And it took us a lot of trials, but we're happy where we ended up. Elon Musk: (01:54:46) Yeah. I mean, everything is simple in recollection, after you... it's hard until it's discovered and then it's simple. So anyway, there's a lot of really cool things going on that enable tabless. And it was really due to a really great engineering team. Drew and the rest of the team had done amazing work in achieving this tabless construction. I think it may sort of sound a bit silly to some people, but for people that really know cells, this is a massive breakthrough. Drew Baglino: (01:55:19) For cylindricals to be able to get rid of the tabs dramatically simplifies winding and coding. And has an awesome thermal and performance benefit. Elon Musk: (01:55:28) Yeah. Just to elaborate on that a bit, it's like when the cell is going through the system, it has to keep stopping where all the tabs are. So you can't do a continuous motion production if you have tabs. You have to keep stopping and then there's a rate at which you can start and stop and accelerate again and it really slows down the rate of production. And then sometimes you get the tabs wrong and you also lose a little bit of active area. It's really a huge pain in the ass to have tabs from a production standpoint. Drew Baglino: (01:56:03) Yes. And so when we put it all together and go to our new 80 millimeter length, 4680 we call this a new cell design, we get five times the energy with six times the power and enable 16% range increase, just form factor alone. Elon Musk: (01:56:23) Yeah. So these... Yeah. It's pretty great. And just to clarify, when we see these plus 16% or whatever the percentage rate increase is, these are the amounts due just to that particular innovation. So we'll list a whole bunch of innovations and then when you add them up, you get a total improvement in energy density and cost. But these numbers are what refer to just this thing. Drew Baglino: (01:56:56) Yeah. And I want to stress, this is not just a concept or a rendering. We're starting to ramp up manufacturing of these cells at our pilot 10 gigawatt hour production facility, just around the corner. Elon Musk: (01:57:08) Yeah. So. Yeah. It's a video of some of what's going on in the plant. Now. I mean, to be clear, it will take about a year to reach the 10 gigawatt hour capacity. So this is important to appreciate. When you build a factory, there's a certain capacity that you design to, and then it takes some period of time to actually achieve that capacity. So I would say it's probably about a year before we get to the 10 gigawatt hour annualized rate with the pilot plant. And this is just a pilot plant. The actual production plants will be more on the order of maybe 200 gigawatt hours, maybe more over time. Drew Baglino: (01:58:00) And... Thank you. But let's stack up everything we just saw at the cell level. So just the cell form factor change enables a 14% dollar per kilowatt hour reduction, just that cell form factor change. And now that you've been teased on this factory, we're going to go on and walk step-by-step through that factory and discuss a series of innovations there. When thinking about the ideal cell factory, we have inspirations behind us in the paper and bottling industry, where from humble beginnings, over a century of innovation has enabled mass scale, continuous motion, unbelievably low manufacturing costs. And when we think about the lithium-ion industry, which is really only in its third decade of high volume production, it has so far to go to achieve similar scale and simplicity. And that was the inspiration that we set out to the team as we thought about how to marry cell design and manufacturing in the best possible factory. Drew Baglino: (01:59:05) And let's talk a little bit about what's in a cell factory. First, there's an electrode process where the active materials are coated into films onto foils. Then those coated foils are wound in the winding process we just talked about where if you do have tabs, you have to start and stop a lot. Then the jelly roll is assembled into the can, sealed, filled with electrolyte, and then sent to formation where the cell is charged for the first time and where the sort of the electrochemistry is set and the quality of the cell is verified. And we set out at every step of this process to try to take that inspiration we just showed and think about how we make those processes fundamentally better and more scalable. And one of the most important processes is where it all begins, the wet process of the electrode coding. And just to give you all a sense of scale, I'm going to walk through what's in that wet process. Drew Baglino: (02:00:09) You've got mixing where the powders are mixed with either a water or a solvent, solvents for the cathode. That mix then goes into a large coat and dry oven where the slurry is coated onto the foil, huge ovens, tens of meters long, dried, and that solvent then has to be recovered. You can see the solvent recovery system. And then finally the coated foil is compressed to the final density. And when you're looking at this, you're like, wow, that's a lot of equipment for one step, especially when you consider that little spec next to the coating oven is a person. This is serious iron involved in making batteries. Wouldn't it be great if we could skip that solvent step, which is one of those dig a ditch, and then fill it kind of things where you put the solvent in and then take it out and recycle it, and just go straight to a dry mix to coat? And that's what the dry process really is about. And in the most basic form, you can see it here on a benchtop, literally powder into film, as simple as that. Elon Musk: (02:01:25) I mean, it's hard actually, just to be clear. If this was easy, everyone would do it. It's not like dry coating electrode is actually easy. It's actually very hard to do what appears to be a simple thing. And it's worth noting, we did acquire Maxwell a little over a year ago, I guess, and certainly a good company and everything, but the dry coating they had was like, it's like sort of, I would call proof of concept. Since the acquisition. We've actually ramped the machine that does dry coating four times. So revision full post acquisition of the machine, and there's still a lot of work to do. So I would not say this is completely in the bag. It's still a lot of work to do. And as you grow, as you scale, go from benchtop to lab to pilot to volume production, there are actually major issues that you encounter at every level. It's not like you make something work on your bench and bingo, now you can make a bazillion of it. Drew Baglino: (02:02:26) Absolutely. Elon Musk: (02:02:26) It's insanely difficult to scale up. Yeah. Drew Baglino: (02:02:31) Yeah, but if you do scale it up, what you saw before becomes this. So you can see the motivation. A 10 times reduction in footprint, a 10 times reduction in energy and a massive reduction in investment. But as Elon was saying, simple is hard. Elon Musk: (02:02:48) Yeah. I mean, to be clear, I would like to not say that right now, it just totally working. It's close to working, but it's not, even now at the pilot plant level, it is close to working. It's fair to say probably it does work, but with not a good, not a high yield. Drew Baglino: (02:03:07) Yeah. We're still ironing out the kinks, but we've made tens of thousands of cells, thousands of kilometers of electrode. I mean, we are on the fourth generation of the equipment so we've learned a lot along the way. I mean, it is super demanding because every atom has its place if you want to deliver the energy density and the cycle life and the supercharging. But we're confident that we will get there, but it will be a lot of work along that. Elon Musk: (02:03:29) There's a clear path to success, but a ton of work between here and there. But this is a really profound improvement. Again, for people that know battery manufacturing, this is gigantic. We'll probably be on machine revision six or seven by the time we do large scale production. The rate at which the machines are being improved is extremely rapid. Literally every three or four months, there's a new rev. Drew Baglino: (02:03:55) Yeah. And beyond the electrode, we continue to innovate on every other process steps. So let's talk a little bit about assembly, which is next. The key to a high-performing assembly line is accomplishing processes while in motion, continuous motion. And thinking of the line as a highway, max velocity down the highway, no start and stop, no city driving. Elon Musk: (02:04:24) Exactly, no stop lights and traffic lights sort of thing. You want the highway. Drew Baglino: (02:04:28) You want the highway. And together with our internal design team that makes this equipment and designs this equipment, we coupled thinking about how to make the best cell with thinking about how to make the best equipment so that we could accomplish the fastest parts per minute rates on all of these tools. And through all of that development, we were able to get to the point where we can implement assembly lines, one line, 20 gigawatt hours, seven times increase in output per line. And when you're thinking about scalability and pure effort, having one line be seven X the capability is just effort multiplying. Elon Musk: (02:05:10) Yeah. So you can sort of think about the sort of the fundamental physics of a factory or something. I think it's actually quite a lot like the rocket equation where you've got basically the rocket equation you've got your exhaust velocity and then the log of [end 02:05:24] masses. So it's basically saying how fast are things going and what percentage of the factory volume is doing useful work? And conveyance does not count as useful work. Drew Baglino: (02:05:34) Only the value added steps. Elon Musk: (02:05:37) Yeah. If you break the factory down into cubic meter sections and say... or smaller. Could be like one liter sections, and say, \"Is a majority of this volume of doing useful work?\" You'd be astounded at how bad most factories are. They'd be like maybe two or 3%, including our factory in Fremont. So I think it's possible to get to at least 10 times that of volumetric efficiency. So more like 30%ish, maybe more, and be 10x better, which means the factory can be 10 times smaller. And then the other thing is how fast are things going through the factory? It's like speed and density. A factory that's moving at say twice the speed of another factory is equivalent to two factories basically. And the company that will be successful is the company that with one factory can accomplish what other companies take two or three or four factories to do. So this is what we're trying to do here is say, okay, how do we, with one factory achieve what maybe five or even 10 factories would normally be required to achieve? Drew Baglino: (02:06:43) And the vertical integration with the machine design teams at Grohmann and Hibar and others allows us to really accomplish that because we don't have any of these edge conditions between one piece of equipment and another, we can design the entire machine to be one machine and remove all of these unnecessary steps. Elon Musk: (02:07:03) Yeah. I mean, basically Tesla is aiming to be the best at manufacturing of any company on Earth. This is the thing that's actually most important in the long run I think, just from a company standpoint and from basically achieving sustainability as fast as possible. But I think also for long-term competitiveness, eventually every car company will have long range electric cars. Eventually every company will have autonomy, I think, but not every company will be a great at manufacturing. Tesla will be absolutely head and shoulders above anyone else in manufacturing, that is our goal. Drew Baglino: (02:07:44) Manufacturing is hard and hard problems are fun to solve. Okay. Now let's talk about formation. In a typical cell factory, formation represents 25% of the investment. And what is formation? Is it's charging and discharging cells and verifying the quality of the cell. It turns out we've charged and discharged billions and billions of cells in our vehicles so we know a thing or two about that. The typical formation set up is you charge and discharge each cell individually. In our car, we charge thousands of cells at once. And we took our principal and our power electronics, leveraging Powerwall vehicle battery management systems and others to dramatically improve the formation equipment cost-effectiveness and density. 86% reduction in formation investment, 75% reduction in footprint. You want to take this one? Elon Musk: (02:08:43) Sure. So essentially what this translates to based on what we know today is about a 75% reduction in the investment per kilowatt hour. Or gigawatt hour. It's just basically four times better than the current state of the art to the best of our knowledge. And I think there's probably room to improve even beyond that. Drew Baglino: (02:09:04) Definitely. Elon Musk: (02:09:05) Definitely. Yeah. So we're able to, from a volume standpoint, actually get what, in a smaller form factor than Giga Nevada, we're able to get many times the cell output. So you can see basically we can get a terawatt hour in less space than it took to make a gigawatt hour, 150 gigawatt hours. So this is pretty profound. I would actually not have thought this was possible several years ago, that we could actually get to a terawatt scale in less space than what we currently envisioned for doing 150 gigawatt hours. Drew Baglino: (02:09:48) Yes. Simpler accelerates terawatt scale. And that's what we need to do to accelerate our mission. And as Elon said, we're going to try to even improve on this as we push towards our goals, which are... Elon Musk: (02:10:02) Yeah. So this is just talking about Tesla internal cell production. As I tweeted out earlier, we will continue to use our cell suppliers, Panasonic and LG and CATL. And so this is a hundred gigawatt hours supplemental to what we buy from suppliers. And yeah, essentially, this does reduce our weighted average cost of a sale, but it allows us to make a lot more cars and a lot more stationary storage. And then long-term, we're expecting to make on the order of a 3000 gigawatt hours or three terawatt hours per year. I think we've got a good chance of achieving this actually before 2030, but I'm highly confident that we could do it by 2030. Drew Baglino: (02:10:58) When you look at the size of that factory on the previous page, it really shows how enabling all of these advancements are in achieving a three terawatt hour goal by 2030. And not only is all of that manufacturing innovation fantastic for enabling scale, it's also an additional 18% reduction in dollar per kilowatt hour at the battery pack level. Elon Musk: (02:11:18) But wait, there's more. Drew Baglino: (02:11:19) But wait, there's more. So we have a manufacturing system, we've got a cell design. What are the active materials we're going to put in that cell design? Let's talk about the anode first. Let's talk about silicon. Why is silicon awesome? It's awesome because it's the most abundant element in the Earth's crust after oxygen, which means it's everywhere. It's sand. Elon Musk: (02:11:44) Sand is silicon dioxide. Drew Baglino: (02:11:47) And it happens to store nine times more lithium than graphite, which is the typical anode material in lithium-ion batteries today. So why isn't everybody using it? The main reason is because the challenge with silicon is that it expands four X when fully charged with lithium. And basically all of that expansion stress on the particle, the particles start cracking, they start electrically isolating, you lose capacity. The energy retention of the battery starts to fade. And it also gumps up with a passivation layer that has to keep reforming as the particles expand. Elon Musk: (02:12:19) Yeah. Basically with silicon, the cookie crumbles and gets gooey. That's basically what happens. Drew Baglino: (02:12:24) Good analogy. And current approaches to solve this, which exist, I mean, we have silicon in the cars that you're all in right now, involved highly engineered, expensive materials in the scheme of things. Now they're still great and they enable some of the benefits of silicon. They just don't enable all of it and they're not scalable enough. And you can see some of the things that maybe you've heard of, SIO, silicon with carbon, or silicon nanowires. That's kind of the space right now. What we're proposing is a step change in capability and a step change in cost. And what that really is is to just go to the raw metallurgical silicon itself, don't engineer the base metal, just start with that and design for it to expand in how you think of the particle in the electrode design and how you coat it. Elon Musk: (02:13:14) Yeah. And I'm not sure if you saw this. Basically a dollar per kilowatt hours basically. If you use simple silicon, it's dramatically less than even the silicon that is currently used in the batteries that are made today, and you can use a lot more of it. Drew Baglino: (02:13:29) The anode would cost, yeah, with this silicon, the anode costs a dollar and 20 cents a kilowatt hour. Elon Musk: (02:13:38) Yeah. Drew Baglino: (02:13:41) And how does it work? Start with raw metallurgical silicon, stabilize the surface with an elastic ion conducting polymer coating that is applied through a very scalable approach. No chemical vapor deposition, no highly engineered high capacity, high cap ex solutions, and then integrate it into the electrode through a robust network formed out of a highly elastic binder. And in the end, by leveraging this silicon to its potential, we can increase the range of our vehicles by an additional 20%. Just this improvement. Elon Musk: (02:14:13) Yeah. It gets cheaper and longer range. Okay. Drew Baglino: (02:14:20) Yeah. And when we take that anode cost reduction, we're looking at another 5% dollar per kilowatt hour reduction at the battery pack level. And there's more. Let's talk about cathodes. What is a battery cathode? Cathodes are like bookshelves where the metal, the nickel, the cobalt, the manganese or aluminum is like the shelf, and the lithium is the book. And really what sets apart these different metals is how many books of lithium they can fit on the shelves and how sturdy the shelves are. Cobalt is a- Elon Musk: (02:14:53) Sorry, I was going to say it's tough to exactly figure out what the right analogy is to explain a cathode and anode. But a bookshelf is probably a pretty good one in the sense that you need a stable structure to contain the ions. So you want a structure that does not crumble or get gooey, or basically that that holds its shape in both the cathode and the anode. As you're moving these ions back and forth, it needs to retain its structure. So if it doesn't retain a structure, then you lose cycle life and your battery capacity drops very quickly. Drew Baglino: (02:15:30) Absolutely. Yeah. I totally agree. And I think people are always talking about like, oh, what's the catheter going to be? Is it [NCA 02:15:38] or whatever? The thing to consider is just fundamentally what the nickel, the metals are capable of. And that's what we have on the chart here. Dollar per kilowatt hour cathode of just the metal using just LME, London Metal Exchange prices, versus the energy density of just the cathode. And you can see nickel is the cheapest and the highest energy density. And that's why increasing nickel is a goal of ours and really everybody's in the energy- Drew Baglino: (02:16:03) Why increasing nickel is a goal of ours, and really everybody's in the battery industry. But one of the reasons why cobalt is even used at all is because it is a very stable bookshelf. And the challenge with going to pure nickel is stabilizing that bookshelf with only nickel. And that's what we've been working on with our high nickel cathode development, which has zero cobalt in it, leveraging novel coatings and dopants We can get a 15% reduction in cathode dollar per kilowatt hour. Elon Musk: (02:16:31) Yeah. Big deal. Drew Baglino: (02:16:38) But it's not just about nickel. You want a? Elon Musk: (02:16:41) Yeah. Sure. So in order to scale, we really need to make sure that we're not constrained by total nickel availability. I actually spoke with the CEOs of the biggest mining companies in the world and said, \"Please make more nickel, this is very important.\" And so I think they are going to make more nickel. I think we need to have a kind of a three-tiered approach to batteries. Elon Musk: (02:17:07) So starting with iron, that's kind of like a medium range, and then nickel manganese as sort of a medium plus intermediate and then high nickel for long range applications like Cyber Truck and the semi. Something like a semi-truck, it's extremely important to have high energy density in order to get long range. And just to give sort of iron up a bit more time, if you look at [inaudible 02:17:37] per kilogram at the cathode level of iron, it looks like nickel's twice as good, but when you fully consider it at the pack level, everything else taken into account, nickel is about maybe 50 or 60% better than iron. Elon Musk: (02:17:52) So iron is a little better than it would seem, when you look at it at the pack level fully considered. It's not as good as nickel, nickel's like 50 to 60% better, but it's actually pretty good. Good for stationary storage and for medium range applications where energy density is not paramount. And then, like I said, for intermediate, it's kind of a nickel manganese, and it's a relatively straightforward to do a cathode that's two-thirds nickel one third manganese, which would then allow us to make 50% more cell volume with the same amount of nickel. Drew Baglino: (02:18:32) And with very little energy trade-off. Just enough to have, you still want to use 100% nickel for something like a semi-truck, but really not much of a sacrifice at all. Elon Musk: (02:18:41) Yeah. Drew Baglino: (02:18:44) And beyond the metals, because a lot of people spend time talking about the metals. Actually the cathode process itself is a big target. 35% of the cathode dollar per kilowatt hour is just in transferring it into its final form. And so we see that as a big target. And we decided to take that on. Drew Baglino: (02:19:03) Here's a view of the traditional cathode process. Effectively, if you start at the left and you have the metal from the mine, the first thing that happens is the metal from the mine is changed into an intermediate thing called a metal sulfate, because that just happened to be what chemists wanted a long time ago. And then when you're making the cathode you have to take this intermediate thing called the metal sulfate fate, add chemicals, add a whole bunch of water, a whole bunch of stuff happens in the middle, and at the end you get that little bit of cathode and a whole bunch of wastewater and byproducts. Elon Musk: (02:19:34) It's insanely complicated. It's a small world journey of, \"I am a nickel atom, what happens to me?\" And it is crazy. You're going around the world three times, there's the moral equivalent of digging the ditch, filling the ditch and digging the ditch again, it's total madness basically. And these things just grew up, they're just kind of like legacy things, it's like how it was done before and then they connected the dots but really didn't think of the whole thing from a first principle standpoint saying, \"How do we get from the nickel ore in the ground to the finished nickel product for a battery?\" So we've looked at the entire value chain and said, \"How can we make this as simple as possible?\" Drew Baglino: (02:20:18) And that's what we're proposing here with our process. As you can see, a whole lot less is going on here. We get rid of the intermediate, metal, water, final product cathode, recirculate the water, no wastewater at all. And when you summarize all of that it's a 66% reduction in CapEx investment, a 76 reduction in process costs and zero waste water. Much more scalable solution. Drew Baglino: (02:20:48) And then when you think about the fact that now we're actually just directly consuming the raw metal nickel powder, it dramatically simplifies the metal refining part of the whole process. So we can eliminate billions in battery grade nickel intermediate production. It's not needed at all. And we can also use that same process we showed on the previous page to directly consume the metal powder coming out of recycled electric vehicle and grid storage batteries. So this process enables both simpler mining and simpler recycling. Drew Baglino: (02:21:21) And now that we have this process, obviously we're going to go and start building our own cathode facility in North America and leveraging all of the North American resources that exist for nickel and lithium. And just doing that, just localizing our cathode supply chain and production, we can reduce miles traveled by all the materials that end up in the cathode by 80%, which is huge for cost. Elon Musk: (02:21:45) Yeah. To be clear, cathode production would be part of our the Tesla cell production plant. So it would just be basically raw materials coming from the mine and from raw materials in the mine out comes a battery. Drew Baglino: (02:21:59) And on that note, the way the lithium ends up in the cell is through the cathode. So then we should obviously on-site lithium conversion as well, which is what we will do using a new process that we're going to pioneer. That's a sulfate-free process again, skip the intermediate, 33% reduction in lithium cost, a hundred percent electric facility co-located with the cathode plant. Elon Musk: (02:22:25) So it's important to note that there is a massive amount of lithium on earth. So lithium is not like oil. There's a massive amount of it, pretty much everywhere. In fact is there's enough lithium in the United States to convert the entire United States fleet to electric, all the cars in the United States. Like 300 million or something like that. Every vehicle in the United States can be converted to electric using only lithium that is available in the United States. Drew Baglino: (02:22:55) Discovered today. Elon Musk: (02:22:57) Yeah, what we already know is exist. Drew Baglino: (02:22:58) People really haven't even been looking. Elon Musk: (02:22:59) Yeah, people haven't been trying because it's just widely available. But it is important to say, \"Okay, what is the smartest way to take ores and extract the lithium and do so in an environmentally friendly way?\" And we actually discovered... Again, looking at it a first principles physics standpoint, instead of just the way it's always been done, is we found that we can actually use table salt, sodium chloride, to basically extract the lithium from the ores. Nobody's done this before, to the best of my knowledge, nobody's done this. And all the elements are reusable, it's a very sustainable way of obtaining lithium. And we actually got rights to a lithium clay deposit in Nevada. Drew Baglino: (02:23:51) Over 10,000 acres. Elon Musk: (02:23:52) Over 10,000 acres. And then the nature of the mining is actually also very environmentally sensitive in that we sort of take a chunk of dirt out of the ground, remove the lithium, and then put the chunk of dirt back where it was. So it will look pretty much the same as before, it will not look like terrible. And yeah, it'll be nice. Drew Baglino: (02:24:13) Simply mix clay with salt, put it in water, salt comes out with the lithium, done. Elon Musk: (02:24:18) Yeah. It's pretty crazy. Drew Baglino: (02:24:19) Yeah. So we're really excited about this and there really is enough lithium in Nevada alone to electrify the entire US fleet. Elon Musk: (02:24:27) Yeah, that's true. Actually, just what's in Nevada. Basically, there's so much damn lithium on Earth it's crazy. It's one of the most common elements on the planet. Drew Baglino: (02:24:39) And eventually, as we said at the beginning, when we get to this steady state 20 terawatt hours per year of production, we will transfer the entire non-renewable fleet of both power plants, home heating and industry heating and vehicles to electric. And at that point, we have an awesome resource in those batteries to recycle, to make new batteries. So we don't need to do any more mining at that point. And you can see why. The difference in the value of the material coming back from the vehicle versus the ground, you'd always go to the vehicle. And we recycle a hundred percent of our vehicle batteries today. And actually, we are starting our pilot full-scale recycling production at Gigafactory Reno next quarter to continue to develop this process as our recycling returns grow. Elon Musk: (02:25:30) To date, it's been done by third parties, but we think we can recycle the batteries more effectively, especially since our batteries, we're making the same battery as the thing we're recycling. Whereas third party recyclers have to consider batteries of all kinds. Drew Baglino: (02:25:46) Yeah. And just to think about what this actually means, the recycling resource is always 10 or greater years delayed because batteries last a really long time, but eventually it is the way that all resources will be made available. And that's why we're investing in this recycling facility in Nevada. Elon Musk: (02:26:04) Yeah. Long-term, new batteries will come from old batteries once the fleet reaches steady state. Drew Baglino: (02:26:11) Right. Okay. So we just talked about scaling cathode and recycling, all of the benefits that you just saw are added to this benefit of a 12% reduction in dollars per kilowatt hour at the battery pack level, almost at our half of the cost goal, but there's one more section. Take it away, Elon. Elon Musk: (02:26:31) So there's an architecture that we've been wanting to do at Tesla for a long time, and we've finally figured it out. And I think it's the way that all electric cars in the future will ultimately be made, it's the right way to do things. Elon Musk: (02:26:52) So it starts with having a single piece casting for the front body and the rear body. And in order to do this, we commissioned the largest casting machine that has ever been made. And it's currently working just over the road at our Fremont plant. It's pretty sweet. Currently making the entire rear section of the car as a single piece, high pressure die-cast aluminum. And in order to do this, we actually had to develop our own alloy because we wanted a high strength casting alloy that did not require coatings or heat treatment. This is a big deal for castings. Especially with a large casting. If you heat treat it afterwards it tends to deform, it kind of does this like potato chip thing. So it's very hard to keep a large casting to have its shape. Elon Musk: (02:27:47) So in order to achieve this, there was no alloy that existed that could do this, so we developed our own alloy, a special allow of aluminum, that has high strength without heat treat and is very castable. So that's a great achievement of our materials team. In fact, in general, we've got a lot of advanced materials coming for Tesla, new alloys and materials that have never existed before. Elon Musk: (02:28:10) So, you're basically making the front and rear of the car is a single piece and that then interfaces to what we call it, the structural battery. Where the battery for the first time will have dual use. The battery will both have the use as an energy device and as structure. This is absolutely the way things are done. In the early days of aircraft they would carry the fuel tanks as cargo. So the fuel tanks actually were quite difficult to carry. They're basically worse than cargo, you had to add to kind of bolt them down. It was very difficult. And then somebody said, \"Hey, what if we just make the fuel tank in wing shape?\" So all modern airplanes, your wing is just a fuel tank in wing shape. This is absolutely the way to do it. And then the fuel tanks serves this dual structure, and it's no longer cargo. It's fundamental to the structure of the aircraft. This was a major breakthrough. We're doing the same for cars. Elon Musk: (02:29:26) So this is really quite profound. Effectively the non cell portion of the battery has negative mass. So we saved more mass than the rest of the vehicle than the non cell portion of the battery. So it's like, \"How do you really minimize the mass of a battery? Make it negative. Make the non cell portion of battery pack negative.\" So it also allows us to pack the cells more densely because we do not have intermediate structure in the battery pack. So instead of having these supports and stabilizers and stringers and structural elements in the battery, we now have a lot more space in the battery because the pack itself is structural. Elon Musk: (02:30:10) What we do essentially, instead of having just a filler that is a flame retardant, which is currently what is in the 3NY battery packs, we have a filler that is a structural adhesive, as well as flame-retardant. So it effectively glues the cells to the top and bottom sheet. And this allows you to do shear transfer between upper and lower sheet. Just like if you have a formula one craft or a racing boat, and you have carbon fiber face sheets and aluminum honeycomb between them, this gives you incredible stiffness and it's really the way that any super fast thing works is you create basically a honeycomb sandwich with two face sheets. Elon Musk: (02:30:58) This is actually even better than what aircraft do. Because aircraft do not do this. They can't do this because fuel is liquid. So in our case the batteries are solid. So we can actually use the steel shell case of the battery to transfer shear from the upper and lower face sheet, which makes for an incredibly stiff structure, even stiffer than a regular car. In fact, if this was a convertible that had no upper structure, that convertible will be stiffer than a regular car. So it's just really major. Elon Musk: (02:31:38) So it improves the mass efficiency of the battery. And then those castings are also quite important because you want to transfer load into the structural battery pack in a very smooth, continuous way. So you don't put arbitrary point loads into the battery. So you want to sort of feather the load out from the front and rear into the structural battery. It also allows us to move the cells closer to the center of the car, because we don't have the... In the top one we've got all the supports and stuff, so the volumetric efficiency of the structural pack is as much better than a non-structural pack. And we're going to actually bring the cells closer to the center and because they're closer to the center it reduces the probability of a side impact potentially contacting the cells because in any kind of side impact has to go further in order to reach the cells. Elon Musk: (02:32:36) It also proves what's called the polar moment of inertia. Which is if you think of when there's a ice skater arms out or arms in. Arms in, you rotate faster. So if you can bring things closer to the center, you reduce the polar moment of inertia and that means the car maneuvers better. It just feels better. You won't know why, but it just feels more agile. So it's really cool. This is really major. Elon Musk: (02:33:03) Like it says, so 10% mass reduction in the body of the car, 14% range increase, 370 fewer parts. I really think that long-term in any cars that do not take this architecture will not be competitive, Drew Baglino: (02:33:22) And it's not just at the product level, a better product. But in the factory, it's a massive simplification. You saw the part removal, it's casting machines, it's the structural battery pack. So we're looking at over 50% reduction in investment per gigawatt hour, 35% reduction in floor space. And we'll continue to improve that as we make the vehicle factory of the future. Elon Musk: (02:33:45) Yeah. So major improvements on all fronts from the cell all the way to the vehicle. Drew Baglino: (02:33:52) And in addition to the improvements we just said on enabling additional range and improving the structural performance of the vehicle, it is worth another 7% dollar per kilowatt hour reduction at the battery pack level, bring our total reductions now to 56% dollars per kilowatt. Drew Baglino: (02:34:17) All right. So stacking it up. We're not just talking about cost or range. We've got to look at all the facets. So range increase, we're unlocking up to 54% increase in range for our vehicles and energy density for our energy products. 56% reduction in dollars per kilowatt hour at the battery pack level, and a 69% reduction in investment per gigawatt hour, which is the true enabler when we talk back about how do we achieve this scale problem here. Elon Musk: (02:34:47) Yeah. So I think it's pretty nice that investment per gigawatt hour reduction is 69%. I mean, who would have thought? Drew Baglino: (02:34:57) Yeah, just happened to come out that way. Elon Musk: (02:35:03) I mean, 0.420%, of course. So what this enables us to do is achieve a new trajectory in the reduction of cell cost. And now to be clear, it will take us probably a year to 18 months to start realizing these advantages and to fully realize the advantages probably it's about three years or thereabouts. So if we could do this instantly we would, but it just really bodes well for the future and means that the long-term scaling of Tesla and the sustainable energy products that we make will be massively increased. So, what tends to happen as companies get bigger is things tend to slow down, actually they're going to speed up. Drew Baglino: (02:36:00) And they have to speed up if we're going to accelerate the transition to sustainable energy. Elon Musk: (02:36:04) Yeah. Long-term we want to try to replace at least 1% of the total vehicle fleet on Earth, which is about 2 billion vehicles. So long-term, we want to try and make about 20 million vehicles a year. Drew Baglino: (02:36:25) But I think it's important to point out that when we talked about three terawatt hours by 2030, the problem is a 20 terawatt hour problem. So everybody needs to be accelerating their efforts to accomplish these objectives. It doesn't matter where you are in the value chain. There is a ton to do, you need to rethink from first principles how you do it, so that you can scale to meet all of our objectives. Elon Musk: (02:36:47) Yep. Drew Baglino: (02:36:49) And, Elon. Elon Musk: (02:36:50) Sure. Drew Baglino: (02:36:53) What does this mean... Elon Musk: (02:36:55) What does this mean for our future products? So we're confident that long-term we can design and manufacturer a compelling $25,000 electric vehicle. This has always been our dream from the beginning of the company. I even wrote a blog piece about it because our first car was an expensive sports car, then it was a slightly less expensive sedan, and then finally sort of a, I don't know, mass market premium, the Model 3 and Model Y. But it was always our goal to try to make an affordable electric car. And I think probably, like I said, about three years from now, we're confident we can make a very compelling $25,000 electric vehicle that's also fully autonomous. Drew Baglino: (02:37:48) And when you think about the $25,000 price point, you have to consider how much less expensive it is to own an electric vehicle. So actually it becomes even more affordable at that $25,000 price point. Elon Musk: (02:38:02) Yeah. So we have and extreme performance and range. And we should probably talk about, more or less, Plaid. What about that? So, yeah. Anyway, we took the latest Plaid out to Laguna Seco on Sunday, it got a minute 30, and we think probably there's another three seconds or more to take off that time. So we're confident the Model Plaid will achieve the best track time of any production vehicle ever, of any kind, two-door or otherwise. Elon Musk: (02:39:15) And you can order it now. And it's available basically in the next year. And now we'll move to Q&A. Drew Baglino: (02:39:26) Absolutely. Elon Musk: (02:39:27) So we'll invite a few people on stage. Drew Baglino: (02:39:31) Come on up team. Elon Musk: (02:39:32) This is just a small portion of the team, but I thought it'd be great to show you some more of the team and when we do Q&A we can give various people different questions to answer. Drew Baglino: (02:39:49) Sounds great. Actually, I don't know how we're getting the questions. Elon Musk: (02:39:54) Actually, I don't know either. You can maybe get out of the car for two seconds and yell it at us. How are we getting the questions? Speaker 2: (02:40:07) [inaudible 02:40:08]. Drew Baglino: (02:40:09) Oh, there are mics. Wait for the mic. Elon Musk: (02:40:11) Oh, there are mics. Okay, great, great. Drew Baglino: (02:40:14) All right. Elon Musk: (02:40:16) Okay. We'll definitely needs to give people mics cause otherwise there's no way. Sorry? All right. We're going to pass some mics out. Oh, we don't have a name for the $25,000 car yet. Drew Baglino: (02:40:45) That's a great question, though. Speaker 3: (02:40:45) Elon, you talked about in Berlin that you were going to [inaudible 02:40:45] manufacturing [inaudible 02:40:44]. Elon Musk: (02:40:45) Yes, we will be manufacturing cells in Berlin. Yep. Drew Baglino: (02:40:52) Thermal management system? Speaker 4: (02:40:53) [inaudible 02:40:56]. Drew Baglino: (02:40:55) For homes. Elon Musk: (02:40:56) Oh, you mean like the home HVA? Yeah. That's a pet project that I'd love to get going on. I don't know, maybe we'll start working on that next year. Because I just think, man, you could really make a way better home HVAC system that's really quiet and super efficient, super energy efficient, and also has a way better filter for particles. And it works very reliably, and we've already developed that for the car. So the heat pump in the Model Y is really pretty spectacular. It's tiny, it's efficient, it has to last for 15 years, it's got to work in all kinds of conditions from the coldest winter to the hottest summer. So we've actually already done a massive amount of the work necessary for a really kick-ass home HVAC. Elon Musk: (02:41:53) And they could also stack them. So if you want to say, depending upon the size of your house or whatever, how much you need, you can just basically stack them and just have a very compelling, super efficient home HVAC. And then you could also communicate with the car and it'll know when you're coming home. So it's like, \"Oh, I don't need to keep the house cold all day, I'll just cool it down because I knew you were coming home.\" So the pack can communicate with the car and just really dial it into when you actually need cooling and heating. It'll be great. Drew Baglino: (02:42:25) Fun product. Who's next? Eli: (02:42:30) Hello? Hey guys, Eli here from Tesla Owners Club, my Tesla adventure. Just quick question, so I'm a huge fan of car camping in my Tesla with my dream case, my all time favorite activity, is it going to be possible to get climate control to the back of the cyber truck? Because that would be the ultimate camping machine if we can get all night climate control. Elon Musk: (02:42:51) We'll try to do that. Yeah. I agree, that would be really cool. Yeah. Drew Baglino: (02:42:59) All right. Who's next? Speaker 5: (02:43:00) Hello, longtime fan, Elon, great guy. Just a question, how does the ICE industry look like in the future? Elon Musk: (02:43:12) Well, I don't think there will be at ICE industry longterm. Well, I guess there might be like a few things that it's a like curious thing. There's still like some steam engines made somewhere, but they're just basically sort of quirky collector's items. I mean, that will be the future of the internal combustion engine car. Ryan McCaffrey: (02:43:36) Hi, Elon, to your left here in the white Model y. Ryan McCaffrey from the Ride the Lightning Tesla podcast. Curious about cyber truck, it was interesting to see where you had it in on the battery technology front. I'm sort of curious what you see for it in the production front. Trucks are so popular in America, do you see its volume equaling the 3 or the Y in the future? And also, were Tesla's able to legally be sold in Texas as part of the Giga Texas deal? Elon Musk: (02:44:09) Well, it's hard to say what the volume exactly would be for the cyber truck. The orders are gigantic. We have like, I don't know, well over half a million orders, I think maybe six or 600,000. It's a lot, basically, we stopped counting. So I think there's probably room for, I don't know, at least like a unit volume of like 250 to 300,000 a year, maybe more. Now, we are designing the cyber truck to meet the American spec. Because if you try to design a car to meet the super set of all global requirements you can't make the cyber truck, it's impossible. So it really is designed for the American market, but this is the biggest market. Our North American market is the biggest market for pickup trucks by far or large pickup trucks. Elon Musk: (02:44:59) And then I think we'll probably make an international version of cyber truck that'll be kind of smaller, kind of like a tight Wolverine package. It'll still be cooler, but it'll be smaller because you just can't make a giant truck like that for most markets. So, yeah, but it's going to be great. Elon Musk: (02:45:17) And I don't know. I think probably we'll be able to sell directly in Texas. We do pretty well right now, but it is a bit weird not being able to actually conclude a transaction in Texas, but it's got to be like a click on a server based in California. But weirdly we can do leasing in Texas, but not selling. Hopefully that'll get cleared up in the future. Speaker 6: (02:45:42) Elon, great job with everything that you're doing. It's Ross Gerber from Gerber Kawasaki. Your team's amazing. What I'm most curious about, these innovations are incredible but on my drive up here fully on autopilot for 400 miles, the entire state is brown and this is ultimately about climate. Has there been some analysis done if all these things are achieved, what will its direct impact be on climate? Elon Musk: (02:46:10) I think it will have a very significant impact because it will stop the CO2 PPM from growing as it is every year. I should say, I try to view the whole climate thing as a science question as much as possible. Science, you always question your hypothesis, is it true? Is not true? Or assign a probability to a given hypothesis. And I should say that my original interest in electric vehicles predates the climate issue. When I was in high school, I thought, \"Man, if we don't figure out electric cars, the whole economy's going to collapse when we run out of oil.\" So we better figure out electric cars and sustainable energy or civilization's going to crumble. Elon Musk: (02:46:57) And then it was only later that the significance of the climate risk became apparent. And we were also able, using tracking and other types of technology to access a lot more fossil fuels than previously thought, which is helpful for lowering the cost of gasoline, but it's pretty bad for the total tonnage of CO2 that you could put in the atmosphere. It's now greatly beyond what people previously thought. As we were just going through this presentation, it is a absolutely monumental task to accelerate the advent of sustainable energy. The entire global economy is still more than 99% dependent on, or call it roughly 99% dependent on, fossil fuels. So although electric cars get a lot of press right now, as a percentage of the total global fleet it's practically nothing. I would say yes, less than 1% of the global fleet is electric right now. Because of two billion cars and trucks and whatnot in use. So there's a massive amount of work ahead. Just insane, like hard to comprehend how much work is ahead to get the new vehicle production to be sustainable, to massively increase the amount of stationary storage, which is critical because renewable energy is intermittent, wind, and solar is intermittent, sometimes the wind doesn't blow and this obviously sun doesn't shine at night, so you got to have batteries, a massive, massive number of batteries. Drew Baglino: (02:48:44) Yeah, it's hard to measure in direct impact, but it's an experiment that we shouldn't be performing. And the sooner we can end the experiment the sooner we can kind of move on in a fully sustainable way that is actually lower cost. I think the thing that people haven't fully internalized is once we do get to the 25K car, the ownership cost of that car is incredibly lower than the prior car. And then on the solar side and wind, with the cost of solar wind coming down and with batteries coming down with them, the actual cost of energy on the grid is going down. So we're sort of moving towards a sustainable lower cost future. So there's not like a sacrifice. Elon Musk: (02:49:21) That's true. It is a false dichotomy to say that it's either prosperity or sustainability. This is often used by oil and gas to say like, \"Oh, well, do you want people to lose their jobs? Do you want to lower people's standards of living? Do you want to make all these economic sacrifices really in order to have sustainability?\" And the reality, as Drew was saying, is that sustainable energy is going to be lower cost, not higher cost than fossil fuels. Speaker 7: (02:49:52) Elon, quick question for you, right here in front. First, thanks for having everyone. I was telling a friend, the one company to go work for that's going to have the biggest structural... Speaker 8: (02:50:03) And the one company to go work for, that's going to have the biggest structural impact over the next 10 years at scale, it's probably Tesla. So kudos to everyone at Tesla for what they've done to this point and going forward. The two questions for you, as you've looked at the auto in the storage markets, I know you've talked about it at kind of 50/50 longterm, but it seems like a lot of the battery cost curve achievements that you presented today, really make some of these storage opportunities much more feasible over the next five years. And so I guess the first part of the question is, does your calculus upon learning and improving these things, change on that 50/50 mix, or is there a role where storage becomes bigger? And then the second part of the question, with all these huge grand visions, who's going to be with Tesla from a corporate perspective, accomplishing these things? Obviously, Tesla can't do it alone, but when you look at some of the traditional auto industry or power, et cetera, I don't see a lot of other Tesla's. Elon Musk: (02:50:59) Well actually, there's a lot of companies in China that I think are doing great work with electric vehicles and also with stationary storage, although we don't see that much in the US yet, but I think probably we will in the future. I don't know, obviously we're doing everything we can to encourage other companies to move to sustainable transport and also make stationary storage batteries. We made our patents freely available, we really try to tell these companies, \"Hey, you really need to do this, or you won't exist in the future,\" but they don't believe it. So we've talked until we're blue in the face. What are we supposed to do? But we really are hopeful that other companies will also do what we're doing and that will make a sustainable future come sooner. Drew Baglino: (02:51:53) From a fundamental market size perspective, we did the first ground up work to show the size of the market in terawatt hours and they are roughly 50/50. 10 terawatt hours for transportation, 10 terawatt hours for the grid. And part of that is because the grid batteries, because when you're making a power plant, you're making a large investment, our 25-year assets are greater. If the grid batteries were 10-year kind of things, the grid market would be bigger, but because it's a longer duration asset, they're roughly the same size. Speaker 9: (02:52:31) Thinking long-term, is there any other segments that this new battery will be able to disrupt or electrify, beyond just the initial Model 2 or cheaper sedan? Like a Boring Company loop, plane- Elon Musk: (02:52:44) Where are you? Are you there? Speaker 9: (02:52:45) What's up? Right here. Elon Musk: (02:52:46) Okay, great. It's like ventriloquism here, we just get the sound out of the speaker and can't tell where the heck it's coming from. Speaker 9: (02:52:55) Yeah. Any hints or is the model too such a big deal because it decreases the cost of transportation, that that is really the disruption, or should we get hyped that this new cost curve opens up different vehicle categories, like a high passenger density bus, Boring loop, boat, plane? Elon Musk: (02:53:12) Well, I mean, there are batteries in limited production right now, that do exceed 400 watt hours per kilogram, which I think is about the number you need for a decent range, medium range aircraft. And I think our batteries will, over time, start to approach the 400 watt hours per kilogram range as well. So yeah, I mean, I think over time, we'll see all modes of transport, with the ironic exception of rockets, transition to sustainability or to electric basically. On the rocket front, what we're planning to do is, about 80% of Starship is liquid oxygen and we're actually already running a power line to be able to use wind power to create the liquid oxygen. So we're making some decent progress on sustainability on the rocket front, but there's just no way to have an electric rocket. And it's important for the future of life and consciousness, that we become a multi planet species, so got to keep doing that. Josh Phillips: (02:54:21) Hi Elon, Josh Phillips here, retail investor. I have a question in regards to the lithium and nickel industries and the likely price spikes and shortages of high grade materials the EV industry is likely to see if they don't act fast to address future supply. Tesla have clearly made the right moves that are necessary, but there's a real worry that the potential supply issues and price spikes will create a drag on the rest of the EV industry and therefore a drag on global EV adoption. What advice would you give to the EV and mining industries to quickly solve this looming hurdle? Because for a sustainable energy future, the spice must flow. Thank you. Elon Musk: (02:55:07) Yeah, indeed. The spice must flow. The new spice. I don't know. I'm not sure. I guess we can try to basically overdo it in cell production and perhaps supply cells to others, but we do see the fundamental constraint, as total cell production. That's why we're putting so much effort into making cells and kind of trying to reinvent every aspect of cell production, from mining the ore, to a complete battery pack, because it's the fundamental constraint. We're not getting into the cell business just for the hell of it, it's because it's the fundamental constraint, it's the thing that is the limiting factor for rapid growth. But we could certainly try to overdo it on cell production and perhaps sell cells to others, although we are going at absolute top speed, so it's not like we're holding it back. Elon Musk: (02:56:15) I think just making really efficient cars that have lower drag coefficient, low rolling resistance, efficient powertrains, I mean, that's kind of what we've done in order to make iron phosphate still have a good range. So the iron phosphate's a lower energy density solution, but while there are some limitations on the total amount of nickel produced every year, there's really no limit on the iron. There's so much iron it's ridiculous. So you can really scale up iron phosphate at a raw materials basis, more than you can nickel. Drew Baglino: (02:57:00) And just to point out, when we were walking through this presentation, we intentionally separated all the different aspects. The benefits of structural battery, apply to an iron based cathode in the same way they apply to a nickel based cathode. So you get longer range, iron base vehicles. And also the silicon benefit can apply to the iron based vehicles as well. So we can do a lot to extend the range of an iron based vehicle, which is why it's a key part of the roadmap going forward. And then I invited Turner up here to talk about what the mining industry can do. Turner: (02:57:31) Yeah. Diversification on the cathode side, is obviously massive and EVs are all about efficiency. And so for the EV industry, for the vehicle industry, we need to see powertrain efficiency really increase, all other companies, matching Tesla powertrain efficiency, so that everyone can have that diversified cathode approach, where LFP is used in medium range, and even really make a 300 mile vehicle with LFP. And really the goal that we were trying to present here, was a model for vertical integration, strategic vertical integration, that a lot of different people can do. What we need to see is vertical integration that shortens the process path, from mine to cathode. And what we're doing here is novel and we're trying to push the industry in that direction. So we're presenting a model here that anyone can can follow. Elon Musk: (02:58:27) Yeah. In fact, if there's anything that you guys want to comment on, feel free to step forward and say something. Speaker 10: (02:58:34) I think the key is to be smart about your chemistry choices, your materials choices. Elon Musk: (02:58:38) Talk louder. Speaker 10: (02:58:38) Yeah. If you're smart about your materials choices, the spice will continue to flow. You don't need to use the same kind everywhere. It's about strategically planning it out and for miners, I think we are incentivizing them quite a bit, to ramp up their production. Drew Baglino: (02:58:57) Yeah. And actually we had good calls, they're all motivated. I think, they've been sort of sitting back being like, \"Are you going to grow like crazy?\" And we're like, \"Yeah, we're going to grow like crazy.\" And then I think this indicates we're going to grow like crazy and that's what the miners want to hear and then they'll go make the investments. Ben Limpic: (02:59:13) Hello, Elon. This is Ben Limpic, I'm a musician. I was wondering, does Tesla have any future plans to make partnerships with music companies, like it has done with Tencent games or things like that, for you guys to actually kind of expand your services for artists and other types of creative people, to get involved in producing content that can be part of the Tesla ecosystem or so other people that do creative things can get involved with you guys? Elon Musk: (02:59:44) We haven't really thought about it that much, but I suppose it's probably something we should think about. We will be providing a title on the Tesla's. So we're providing more music sources that people can choose from and just generally trying to improve the entertainment experience in the cars. And I think actually as we go to a more autonomous future, the importance of entertainment and productivity will become greater and greater. I mean, to the degree that if you're just basically sitting in your car, the car is fully autonomous and driving somewhere, the car is essentially your chauffeur and then the things that become important are, okay, well let's have good entertainment and if you want to do some productivity stuff, then that actually starts to become much more important because you're no longer spending your attention driving the car. So it will be extremely important in the future. Drew Baglino: (03:00:42) Should we do some of the say.com questions? Speaker 11: (03:00:46) Yeah. Drew Baglino: (03:00:47) Okay. Should we do the second one? Elon Musk: (03:00:54) Yeah. The first one, I think we already answered. If we're able to make enough cells, which we'll try to do, we will supply other companies. It's definitely not an intentional effort to keep the cells to ourselves, if we can make enough for other companies, we will supply them. And we were trying to do the right thing for advancing the sustainable energy, whatever that is. Elon Musk: (03:01:19) Vehicle to grid, we get asked that a lot. I think one of the things that's important to note, is vehicle to grid, unless you have a power cutoff, you need to cut off your main supply to the grid, otherwise, if you lose the power in your house, you'll basically just backflow energy to the grid. So just having a reversal in the power flow, does not actually keep the lights on, you need a whole separate system to cut off power to the grid. And I think there's also the case that people really want the freedom to be able to drive and to charge at their house. And it's obviously very problematic if you get to morning and your car, instead of being charged, it discharged into the house and then you're sort of, \"Okay, now I can either drive or use the battery to power my house.\" Elon Musk: (03:02:19) I think it's actually going to be better for people's freedom of action, to have a power wall and a car separate, and then everything works that. You basically combine that with solar, either solar retrofit or solar glass roof, and local battery storage, so you basically become your own utility and then the car can be charged also with solar. I think that's the stuff that works, that said, we can certainly do vehicle to grid, I think we can basically enable that with software in Europe or something, right? Drew Baglino: (03:03:00) Yeah. Future generations of power electronics, we will be able to do this more or less everywhere, from a energy market participation perspective, but from a backing up the house and it just so happens that the way the North American connectors are, on all the cars in North America, it doesn't matter whether it's the Tesla connector or the connector that the other vehicles have, doesn't actually support powering your home. It's unfortunate, so you'd need an additional hardware to do that. But yeah, in the future, all versions of our vehicles will be able to at least do bi-directional power flow for the purposes of energy market participation. But even for that, it's important to remember that your car is in plugged in 24/7, so it's kind of an unpredictable resource for the grid. It'll have a value, but it's not the same as a stationary battery pack. Elon Musk: (03:03:49) Yeah. Honestly, a vehicle to grid sounds good, but I think actually has a much lower utility than people think. I think very few people would actually use vehicle to grid. With the original roadster, we had vehicle to grid capabilities, nobody used it. Drew Baglino: (03:04:15) How do we find the engineers to do everything we're saying? Elon Musk: (03:04:18) How do we find the engineers to do all these things? Well, I guess we recruit a lot of engineers from all parts of the world. I think Tesla has a good reputation for doing exciting engineering and that tends to attract a lot of the top engineers in the world because they know that their efforts at Tesla will really serve the greater good and we're super hardcore about engineering. Tesla is first and foremost an engineering company, it's like hardcore engineering is what we do. The sheer amount of hardcore engineering done at Tesla is insane. And if you look at say, there's various surveys done of engineering schools, where do you want to go, what's your top choices? And actually the top two choices last few years, have been Tesla and SpaceX. So sometimes it's Tesla first and sometimes SpaceX first, but those are the two top ones. Drew Baglino: (03:05:18) Yeah. I mean, if you're motivated to solve some of these problems, which are the hardest problems in the world to solve, that really fundamentally enable the future we all need, please reach out and help us work on these problems. Elon Musk: (03:05:30) Absolutely. And like you said, the battle is far from over. Less than 1% of the global automotive fleet has been converted to electric and even maybe less than 0.1% of stationary storage has been done. So stationary storage has barely begun, converting the global vehicle fleet to electric, has barely begun. So there's still a massive amount of engineering work to be done at Tesla and other companies, to accelerate this transition to sustainability. Jordan: (03:06:06) Hey, can you guys hear me? Drew Baglino: (03:06:07) Yeah. Jordan: (03:06:08) This is Jordan from Mark Asset Management. So you've talked about the importance of the factory and you've mentioned the ground up design process and a lot of the new things that you're going to be doing or started to do in Shanghai, Berlin, and Austin. Can you just maybe help us understand and quantify, how financially meaningful all of those improvements will be, and then given what you're trying to accomplish as a company, is it fair to assume that the vast majority of improvement will be given back to the customer in the form of lower prices? Elon Musk: (03:06:39) Yeah. I mean, I think certainly we will try to give back as much as possible to the customers. It's not like Tesla's profitability is crazy high, our average profitability for last four quarters, is maybe 1%. So just to be clear, it's not like we're minting money. Our evaluation makes it seem like we are, but we're not. So we do want to try to make the price as competitive as we can, without losing money. If you keep losing money, you'll just die. So this thing called profit is just like, we need to bring in more money than we spend, otherwise we're dead. Drew Baglino: (03:07:19) But affordability is key to how we scale, right? The demand goes non-linear as you reduce the price of the car. Elon Musk: (03:07:25) Yeah. I mean, it's important to sort of separate the difference between affordability and value for money or desirability of the product. So for a lot of people, they want to buy a Tesla, they simply don't have enough money. We could make the car infinitely desirable, but if somebody does not have enough money, they can't buy it. Sometimes people kind of forget this. People have to have enough money to buy the car and just making a car super desirable, but expensive, does not mean they can afford it. So it's absolutely critical that we make cars that people can actually afford. Go through some of these things, scroll down or something. Drew Baglino: (03:08:19) When do you expect Tesla vehicles to beat ICE vehicles on initial purchase price? I think a way to answer that question, is in the classes of vehicles we sell today, we're already doing that. Elon Musk: (03:08:30) Yeah. We're already pretty close. And then factoring in total cost of ownership and the fact that electric vehicles require much less servicing and are way cheaper to run, when you look at total cost of ownership. And you can always lease a car, so if you just lease a car or get a loan for a car, you've got your sort of monthly payments and then your cost for either gasoline or electricity and your cost of servicing and the fully considered cost of electric car is much less than a gasoline car of the same nominal purchase price. I mean, that said, maybe on the order of three years, when we can do lower cost, like a $25,000 car, I think that will be basically on par, maybe slightly better than a comparable gasoline car. So I think maybe it's on the order of three years-ish. Drew Baglino: (03:09:37) How have the technology advancements and increased vertical integration of battery manufacturing, influenced your ability to improve the environmental and social impact of the supply chain? And I think ... Yeah. Elon Musk: (03:09:48) We sort of have said that already. Drew Baglino: (03:09:49) Yeah. Elon Musk: (03:09:50) Do we have some ability to scroll through this? Just scroll away. Drew Baglino: (03:09:57) We covered recycling. Elon Musk: (03:09:59) Yeah. Just scroll until we've got stuff that we haven't covered. Drew Baglino: (03:10:02) We definitely covered that top one. Elon Musk: (03:10:09) Yeah, a lot of the things we've already asked really. Drew Baglino: (03:10:16) Covered that. That one. Elon Musk: (03:10:26) We literally just answered that. Drew Baglino: (03:10:27) Yeah. Oh, I saw a cathode durability question. Let's go to that one, go down, go down, go down. Good technical question. Keep going. How are you going to address the cathode durability and cost and environmental impact trifecta? Is this something you're going to leave the environment upstream and supply chain to solve? No, I think we tried to answer that directly. I mean, we really are looking at not just what happens in the cathode facility, but currently outside the cathode facility that should really be inside and removing processes that shouldn't have been there in the first place and the use of reagents that are just costly and not necessary and removing a bunch of wastewater from the process. Elon Musk: (03:11:09) Guys, is there anything you want to add to ... Maybe we can go through everyone and maybe say what you're doing and say a few words. I don't know. Speaker 10: (03:11:21) Sure. I just want to reiterate the fact that this is a massive problem. Elon Musk: (03:11:25) Massive problem. Speaker 10: (03:11:26) And it seems like Tesla's on its way and ahead, but we need everybody's help because it's everybody's planet and we're not going to get to 20 terawatt hours by ourselves. So please think about this carefully, as it affects everybody, so let's get on it. Elon Musk: (03:11:45) Yeah. And obviously, if you care about solving sustainability and doing hardcore engineering, definitely come work for Tesla. Speaker 12: (03:11:53) Yeah. We went through a couple of the manufacturing improvements and it kind of looks easy when you put together a nice slide deck, but it's super challenging. When you take materials out of the process, when you integrate processes together, you have to do a lot of things at once and that's like this immense engineering challenge. And so to appreciate that, to get through this, we need the best engineers we've got. And we've got this awesome team, I just want to shout out also to all of our team watching, you guys are awesome, you absolutely kicked ass putting this together. Drew Baglino: (03:12:36) Thank you. Thank you, Tesla team. Totally agree. Speaker 12: (03:12:44) Yeah. That's it. Rodney Westmoreland: (03:12:47) Yeah. Rodney Westmoreland, managing the construction here at Tesla. What I would like to say is, one, shout out to the team. The team has been working effortlessly, a very, very tough project here, for 24 hours a day it seems like, around the clock, to have this complete. The thing that sets us apart from a lot of other construction, we have a construction company here, the thing that sets us apart is that we're integrated in the manufacturing process. So every detail that comes from Drew's mouth, is directly implicated into the system that we're building. That way, what would typically take three or four months to create a specification, our design team is working right with the manufacturing team, to allow us to speed that process up tremendously. Drew Baglino: (03:13:36) Yeah, it's definitely a important part of the vertically integrated approach, is to be able to design the factory around the equipment, in fact, together with the equipment, so you can build the factory at lower costs and more quickly. Scott: (03:13:50) I'm Scott, I focus on cell design. I think it's hard to put into words how inspiring this is, been at it such a long time with Tesla. And I really hope others do join us- Elon Musk: (03:14:01) Since when Scott? Scott: (03:14:02) Since 2005, with many of you. Thank you. Year before Drew, who's keeping track? But I'm really stoked what the team's been able to accomplish over the last short period of time, about a year, it's been really an incredible transformation. I mean, hopefully what we've shown you, inspires you to join us or join somebody else in the effort. And I couldn't think of a greater, more intelligent, more hardworking team to be working on for this problem. Peter: (03:14:37) I Peter, I lead the manufacturing improvement team. And I guess the point that I'd like to make, is manufacturing improvements is like the accelerator. So you think about the execution that Rodney talked about, in terms of how fast we've been able to put together this factory, which is amazing and something that's been really incredible to be a part of. That's not enough, what we need to do is improve the manufacturing technology, that's the real accelerator and that's what we're really focused on. Elon talks about it all the time, that really going and improving that system is what will enable us to get to the scale and the cost that we need. Peter: (03:15:15) And then the other point that I would make is on the recruiting side, it doesn't matter if you know about batteries, if you come from any industry, you can do something fantastic in the work that we're doing. We talk to people from industries that you wouldn't imagine. Like I talked to a guy who makes golf balls and he has stuff which is really impactful for what we're doing. So if you're in any industry and you want to be impactful here, come join us, it'd be great. Tony: (03:15:45) Hi, excuse me. Hi, I'm Tony. I've been working in lithium and cathode materials for almost 23 years now and this is the most growth I've seen in a company, I've been here a little over a year and a half. We are hiring amazing people that are allowing us to leverage technology that most of the industry is struggling to achieve. So to answer the question, how are we going to do this? We are really advancing the materials manufacturing for cathodes and for lithium, beyond what has been accomplished in the previous 20 years. Drew Baglino: (03:16:26) It's exciting. Turner: (03:16:31) Yeah. My name is Turner, work closely with the team, have worked a lot with everyone here. On the cathode and upstream materials side, it's really important that everyone understand that this growth is coming. This growth is real, we are going to make all of these batteries and everyone needs to grow with us, the entire supply chain needs to grow with us. And if you have an idea that simplifies anything in the supply chain, come talk to us, come work with us and let's do it. Drew Baglino: (03:17:02) Any existing specification is wrong, any existing manufacturing method is wrong, process equipment, it's wrong, it's just a question of how wrong. Quote Elon Musk. Elon Musk: (03:17:12) Exactly. We're wrong, just the question of how wrong. Trying to be less wrong. Drew Baglino: (03:17:16) So tell us how we're wrong and how we can do it better, so that we can accelerate and improve as fast as possible. Elon Musk: (03:17:23) All right. Well, I guess thank you everyone for coming. I hope you liked the presentation. Very exciting future ahead. We're going to work our damnedest to transition the world to sustainable energy as quickly as possible, and your support and help is key to that success. So thanks again, super appreciated and look forward to the next event. Thank you. Drew Baglino: (03:17:45) Thank you.","textByLang":{"en":"Al Prescott: (41:03) Good afternoon, everyone. Welcome to Tesla's 2020 Annual Meeting of Stockholders. We're really excited that you could be here with us today. My name is Al Prescott. I'm Tesla's vice president of legal. Al Prescott: (41:15) There'll be two parts of today's meeting. First, the former part of the meeting we'll get out of the way, which we'll cover the seven items that stockholders have been asked to vote on. After the voting, I'll introduce Tesla's co-founder and CEO, Elon Musk, who will give a presentation about the company update and year in review. And then following the conclusion of the stockholder meeting, we'll start our separate Battery Day event. Al Prescott: (41:40) At this time, I'd like to thank the members of the Tesla team and our board, especially those who were able to make it out here in person today, as well as to our representative from PricewaterhouseCoopers, Tesla's independent auditor who is also here. But before we begin, I'd like to introduce you to Robyn Denholm, the chairwoman of Tesla, who would like to say a few words remotely. Robyn Denholm: (42:12) Thank you, Al. Hello everyone and welcome to the 2020 Tesla Shareholder Meeting. A special welcome to the many Tesla shareholders that have joined us today in person as well as online from across the country and around the globe. Robyn Denholm: (42:29) I wanted to start today's proceedings by thanking you, our shareholders, for your tremendous support over the last year. And especially to those of you who have been with us through our journey over the past 10 years, since the company's IPO in 2010. While we have stayed true to our mission of accelerating the world's transition to sustainable energy, in many ways, our company has evolved beyond recognition over the past decade. And that is a great thing. In fact, the pace of developments and the evolution of Tesla has further accelerated over the past 15 months since I last addressed you in June of 2019. You'll hear more about many of the specific achievements from Elon later in the agenda. Robyn Denholm: (43:16) But I would like to take this opportunity to thank all of our Tesla employees across the globe who have done a tremendous job of executing and staying focused on delivering for our customers and shareholders, as the world has gone through one of the most challenging periods in our lifetimes. As a board, we have always taken a long-term view. We have made decisions and supported decisions made by the management team that may not have seemed obvious at the time, but are delivering and will continue to deliver breakthrough results. But it's also important to remember why we do this. As a company, we are focused on addressing one of the biggest environmental challenges of our generation, how to accelerate the world's transition to sustainable energy. Robyn Denholm: (44:08) The last year in particular has seen a tremendous increase in momentum in the movement to sustainable energy from both shareholders and the general public. So in addition to developing amazing clean transportation and energy products, we are doing our part by contributing the right facts and information to this important issue. And we released an extended version of our impact report in April of 2020. In issues version, we have covered in great detail, many areas that are important to our shareholders and our customers alike, such as our environmental impact, greenhouse and other noxious gas elimination, our supply chain efforts, especially in cobalt, and our culture and people focus. We hope that by continuing to put this data out there, we will underscore to the world the importance and impact that we are having as a company. Robyn Denholm: (45:09) Lastly, continuous feedback and input from our shareholders is essential for us to do our jobs. And I would like to thank you for your support in this regard. Many of you have provided me and the team with ideas and insights that we as a board take into consideration as we evolve our governance and company practices. It's especially crucial to the board members as we pride ourselves in adaptability and the diversity of thought and experience that we collectively represent on the board. Robyn Denholm: (45:41) This brings me to my final two things today, as today is his last shareholder meeting, on behalf of the board, I would like to sincerely thank Steve Jurvetson for over a decade of service to Tesla, the board, and our shareholders. You will be missed. Finally, I would like to introduce to you our newest member of the board, Hiro Mizuno, who until recently led the largest pension fund in the world. He brings a wealth of experience to the board, but let me hand over to Hiro to say a few words. Hiro ... Hiro Mizuno: (46:17) Thank you, Robyn. Ladies and gentlemen, welcome to Tesla Annual Shareholders Meeting. It is my real pleasure to virtually meet you, Tesla shareholders, people who believe in Tesla's mission and its growth opportunities. I spent all my career in finance and asset management in Tokyo, New York, London, and the Silicon Valley. Hiro Mizuno: (46:42) Until recently, I was a chief investment officer of GPIF $1.5 trillion Japanese public pension fund. And one of my priorities as the investment chief was to promote responsible investments, which aim to make financial returns while pursuing ESG agenda, such as environment and social issues. I believe in the market where ESG is becoming mainstream. Purpose or mission driven businesses will gain long-term investors support. Hiro Mizuno: (47:19) This is why I was interested in Tesla, where our mission is to accelerate the world's transition to sustainable energy. I'm very excited to join the Tesla team on the journey and hope that [inaudible 00:47:34] Tesla deliver what investors expect by further enhancing its environmental and social impact. Once again, Tesla shareholders, thanks for your support. I'm looking forward to seeing you in person next year. Thank you. Al Prescott: (47:54) Thanks Robyn and Hiro. I will now call the meeting to order. Please refer to the meeting agenda that has been provided to you and posted also to our virtual meeting site. The time is now 1:49 PM Pacific Time. And I declare that the polls are now open. Al Prescott: (48:12) We've already received voting proxies from stockholders over the past few weeks, meaning that almost all of the votes that will be counted were already submitted before the meeting. However, if you wish to vote now or to change your prior vote, you may do so through the virtual meeting site. For those that are here in person today, ballots and ballot boxes were available to you at check-in. Al Prescott: (48:37) Tesla's board of directors has appointed Computershare Trust Company to serve as inspector of elections for the meeting. Computershare has taken and signed an oath as inspector of election and has certified that starting on August 13th, 2020, the proxy material, or a notice of internet availability of the proxy material were mailed or provided to all Tesla stockholders of record as of July 31, 2020. Al Prescott: (49:04) We have a majority of the outstanding shares represented at the meeting. So I declare that there is now a quorum present and that we may proceed with the meeting. The items on the agenda are as follows; the election of three class one directors, Elon Musk, Robyn Denholm, and Hiromichi Mizuno to each serve for or term of three years. Two, to approve Tesla's executive compensation on an advisory basis. And three, to ratify the appointment of PricewaterhouseCoopers, LLP as Tesla's independent, registered public accounting firm for the fiscal year of 2020. Tesla's board has recommended that our stockholders vote for each of the director nominees and for each of those proposals. Al Prescott: (49:59) In addition, we have also received four stockholder proposals as described in the proxy statement. I would like to remind our stockholders that Tesla's board has prepared a statement in opposition to each of these proposals, which appear in the proxy. The first stockholder proposal is an advisory vote regarding paid advertising. Our board has recommended that our stockholders vote against this stockholder proposal. This stockholder proposal comes to us from James Danforth. Al Prescott: (50:33) However, Mr. Danforth has notified us that neither he nor his representative will be presenting the proposal at the meeting today. So we will continue. The second stockholder proposal is an advisory vote regarding simple majority voting and our governing documents. Our board has recommended that our stockholders vote against this stockholder proposal. The proposal comes from James McRitchie, who is on the line to present the proposal today. Mr. McRitchie, I would like to invite you now to present. You will have three minutes. James McRitchie: (51:14) I'd like to thank the board for holding such an innovative hybrid meeting during these difficult times. Proposal number five basically asks for a majority voting standard to amend bylaw. I first introduced a proposal on this subject at the 2014 Tesla meeting. Super majority provisions generally use to entrench incumbent directors and managers. Academic research finds that reducing such devices is associated with higher returns. James McRitchie: (51:46) The board's opposition statement argues they tried to adopt a [inaudible 00:51:51] party standard last year, but shareholders rejected it. However, 99.6% of shares voted for the proposal. Only 0.4% voted against it. The problem was that a little more than 35% of shares went unvoted. The vast majority of retail shareholders often don't bother to vote. Since only 65% of shares were voted, we didn't achieve the 66.67% necessary to overturn the current super majority bylaw. James McRitchie: (52:32) It appears the proposal failed primarily for three reasons. One, the board put forth less than robust arguments in favor. Two, they added confusion with another proposal to reclassify the board, not into a single class, that's the norm, but into two classes, elected in altering years. Third, the board also failed to make a substantial effort to solicit votes in favor. Also, please consider this proposal in context with other poor corporate governance provisions at Tesla. First, shareholders can only remove directors for cause. What that basically means is the director has to be caught in criminal activity for shareholders to remove them. Second, because the board is divided into three classes, shareholders can only hold individual directors accountable every three years. And third, shareholders cannot call special meetings, nor can they act by written consent. I hope you will agree. Corporations should not be democratic-free zones. Vote for proposal number five so that 33% of shares cannot overrule the wishes of 67%. Thank you. Al Prescott: (53:54) Thank you, Mr. McRitchie. We'll now move on to our third stockholder proposal, which is an advisory vote regarding reporting on employee arbitrations. Our board has recommended that our stockholders vote against this stockholder proposal. This proposal comes from Nia Impact Capital, whose representative Kelly Hull is on the line to present the proposal today. Ms. Hull, I'd like to invite you to go ahead and present. You will have three minutes. Dr. Kristin Hull: (54:28) Hello. My name is Dr. Kristin Hull, and I'm the founder and CEO of Nia Impact Capital. I formally move [inaudible 00:20:36]. This resolution requests that Tesla board of directors overseeing the preparation of a report on the impact of the use of mandatory arbitration on Tesla's employees and on its work place culture. The report will evaluate the association of Tesla's current use of arbitration with the prevalence of both harassment and discrimination in its workplace and on employee's ability to [inaudible 00:21:01], should harassment or discrimination occur. Dr. Kristin Hull: (55:05) This proposal speaks to the widespread experience of discrimination in the workplace by Black, Latinx, and female employees, despite this discrimination being unlawful under the Civil Rights Act of 1964. Tesla has faced a number of serious allegations of racism and sexism at its Buffalo and Fremont plant. Companies that allow bias discrimination and harassment in their workplaces are at risk for unnecessary legal brand financial and human capital issues. Dr. Kristin Hull: (55:36) Support of this resolution is warranted for the following five reasons. One, research shows that companies benefit from diverse and inclusive workplaces. Two, corporate policies that allow harassment and discrimination risk investors capital. Three, the use of arbitration exposes investors to an unknown level of risk. Four, broad concerns exist with respect to fair treatment in Tesla workplace. And Tesla employees have alleged harassment and discrimination on their basically both race and gender. [inaudible 00:56:13] Tesla, a company investors love for its innovation, leadership, and [inaudible 00:56:18] is increasingly lagging behind its peers in its [inaudible 00:56:22] related to workplace diversity, equity, and inclusion. Dr. Kristin Hull: (56:26) Unlike the forward thinking and innovation in its extraordinary product lines, Tesla has not challenged proactive leadership and building a positive company culture or in addressing concerns about its workplace practices. In these material issues, Tesla lags behind its technology and automotive competitors. Dr. Kristin Hull: (56:48) The use of arbitration limits employees remedy for wrongdoing, precludes employees from stewing in court, and often keeps underlying facts, misconduct or case outcomes secret, therefore preventing employees from learning about and acting on shared concerns. Dr. Kristin Hull: (57:05) Simply stated, arbitration allows that corporate behavior like bias, harassment, and discrimination to continue to keep hidden from employees and investors. To maintain Tesla's [inaudible 00:57:17], it is essential that the board seriously assess the implications of the use of arbitration and that Tesla begins to seriously the need to ensure a fair, equitable, positive, and inclusive workplace. Thank you. Al Prescott: (57:34) Thank you, Ms. Hull. Our fourth and final proposal is an advisory vote regarding reporting on human rights. Our board has recommended that stockholders vote against this proposal. This proposal comes to us from the Sisters of Good Shepherd, New York province, whose representative, Terrence Collingsworth is on the line to present today. Mr. Collingsworth, I would like to invite you to speak now. You have three minutes for your proposal. Terry Collingsworth: (58:09) Thank you. I'm Terry Collingsworth, executive director of the International Rights Advocates. I'm here representing the Sisters of Good Shepherd New York province to present item seven on human rights disclosure, which calls upon Tesla to issue a report to describe board oversight of human rights and its human rights due diligence process, including systems to provide meaningful remedies when human rights impacts occur. Terry Collingsworth: (58:40) Tesla faces serious human rights issues and failure to establish a culture of respect for human rights will expose Tesla to new liability issues and significant reputational injury, all of which will have a material impact on the company and its shareholders. The need to set a new course for human rights compliance at Tesla is glaring. Terry Collingsworth: (59:05) Here are five examples of human rights violations occurring now in Tesla's operations: racism, sexual harassment, and disregard for human safety and dignity harm workers at the Gigafactory 2 in Buffalo, New York, every single day. And those workers urge you to remember their experiences in your vote. Terry Collingsworth: (59:28) Tesla is experienced serious labor relations issues at its production facilities and is actively discouraging union organizing. Workers are being exposed to COVID-19 and then are facing retaliation when they ask for greater protections. There are numerous worker health and safety violations as well as wage and hour issues. Terry Collingsworth: (59:50) And finally, there are serious, even deadly, human rights violations occurring in Tesla's global supply chains. On this last issue, my organization brought the pending suit against Tesla for using cobalt mined in the Democratic Republic of Congo by young children. I personally met young boys who lost limbs or were paralyzed in cobalt tunnel collapses. Tesla sources cobalt from these very mines. And its claimed to have quote, \"Zero tolerance for child labor,\" in its supplier code of conduct is simply not true. Tesla is not only tolerating child labor in its cobalt supply chain, it is tolerating the death and maiming of young child minors. Terry Collingsworth: (01:00:42) This demonstrates why the company must circle back and begin a process to report on its treatment of human rights issues as requested in this proposal. I think consumers will have zero tolerance for a company that is exposed as being indifferent to killing and maiming child minors. We are hopeful that Tesla's innovative spirit can be brought to bear on making human rights a priority at the company. Terry Collingsworth: (01:01:12) For example, if the Elon Musk cared about implementing a zero tolerance child labor policy, instead of having a useless paper policy, Tesla could employ satellites or drones at every mine it sources from to actually monitor child labor. I encourage all Tesla shareholders to vote for item seven, human rights disclosure. Thank you for your attention. Al Prescott: (01:01:40) Thank you Mr. Collingsworth. At this time, I'd like to thank our stockholders for all of their active participation in today's meeting and for those who just presented on the line. I'd also like to read some of the comments that have been submitted by you over the course of the meeting. The first comment comes from Michael [Overbaugh 00:01:02:01]. \"I take great pride in the fact that we haven't had to stoop to the level of what advertising represents to get where we are today. I'd hate to give into that kind of temptation now, when we're so close to becoming a household name that's based solely on our merit alone. But if assets do end up having to be set aside for marketing, I'd like to suggest that rather than shoving ads down the customer's throat, we established some sort of hardcore nationwide campaign and event with the goal of getting as many people as possible behind the wheel of a Tesla for an introduction drive. It's well-known how far just doing that alone goes to converting people into fans.\" Al Prescott: (01:02:46) \"A line I recently ran across says, 'You can talk all about the specs as much as you want, but when it comes to buying a car, what ultimately puts butts in seats is the feeling that the vehicle gives you.' By demonstrating that Tesla clearly has both the specs and the feeling, what more needs saying?\" Al Prescott: (01:03:08) Our second comment comes from the United Steel workers on behalf of the Clean Air Now Coalition of Western New York by Sabrina Lu. And it reads as follows, \"Proposal six and seven up for vote this year are the results of widespread concern about mistreatment of Tesla workers at US factories and across the supply chain. It is clear that Tesla is not interested in addressing the harm they have caused to their workers as their board is advising shareholders to vote against the proposal. We're urging all shareholders to vote in favor of proposal six and seven. And on behalf of our workers at the United Steelworkers here in Western New York and for Tesla employees across the country and across the global supply chain, while this doesn't repair the harm, that's already been caused to countless employees, nor repair harm to children and communities forced into slave labor in the DRC, they represent steps towards a more just workplace at Tesla.\" Al Prescott: (01:04:19) This concludes all of the comments. Thank you all for your participation in the comments. We'll now have a final opportunity for any of you to submit proxies in order for them to be counted. So I'll pause and wait for a moment for you to do that. Al Prescott: (01:04:51) Okay. I declare that the polls are now closed. So based on the proxies that we have previously received, I'd like to announce on a preliminary basis that our stockholders have approved the recommendations of Tesla's board on all agenda items, other than the stockholder proposal for an advisory vote regarding simple majority voting in our governing documents. After the final tabulation is completed, we'll formally announce the results of the voting by filing a form 8-K with the SEC within four business days of today. This now concludes the official business of Tesla's 2020 annual stockholders meeting, which is now adjourned. Al Prescott: (01:05:37) Next, we will have a company update and a year in review presented by Elon. And then we will start our Battery Day Event. During the course of those following sessions, we may discuss our business outlook and make forward looking statements. Such statements or predictions based on our current expectations. Actual events or results could materially differ due to a number of risks and uncertainties, including those disclosed in our most recent 10-Q filed with the SEC. These forward looking statements represent our views. As of today. They shouldn't be relied on after today and we disclaim any obligation to update them after today as well. Al Prescott: (01:06:23) We will now continue with the company update and year in review. And it's my pleasure to introduce Tesla co-founder and CEO, Mr. Elon Musk. Elon Musk: (01:06:45) Everyone. Well, I mean, this is definitely a new approach. We've got the Tesla drive in movie theater, basically. It's good to see everyone. It's a little hard to read the room with everyone being in cars, but it's the only way we can do it. So hopefully it's cool. And hopefully you can hear me. Can you guys hear me? Elon Musk: (01:07:08) Okay. All right. Great. Elon Musk: (01:07:12) Well, thanks for coming. I think it's been an incredible year and I'd like to just thank you for your support through tough times, good times. It's been great. Really appreciate everyone who's put their heart and money into Tesla and I think it's worked out pretty well. This has been a good year. And I think there's many good years to come. So I'll go through the shareholder presentation fairly quickly because the real main event here is Battery Day. And really, I'm just going through a recap of what's happened over the past a year or so. Elon Musk: (01:07:58) I think starting from in terms of our ability to create a ... Elon Musk: (01:08:03) In terms of our ability to create a factory, huge kudos to the Tesla Shanghai team for being able to go from literally a dirt pile to volume production in 15 months. It's like, damn. Yeah. And I think something that's really quite noteworthy here is Tesla's the only foreign manufacturer to have a hundred percent owned factory in China. So this is often not well understood or not appreciated, but to have the only hundred percent owned foreign factory in China is a really big deal, and it's paying huge dividends here. So we really wouldn't have the results that we have had this year without the great efforts of the Tesla China team, so I'm super appreciative of that, and we'll see the Shanghai factory continue to scale quite a bit from where it is right now. I think we really could expect that to be, over time, a factory that produces over a million vehicles a year. Elon Musk: (01:09:16) Yeah, it's cool. So let's see. So we also reached in the past year of volume production of the Model Y, and this was the smoothest launch that we've ever had, so I think we're definitely getting better at a new vehicle launches and building factories and scaling production. As you've heard me say before, the hardest thing is scaling production, especially of a new technology. It's insanely difficult. Making a prototype is relatively easy, and if I think, like, what is the real achievement of Tesla in sort of car company terms, it's like it wasn't making sort of exciting prototypes. It was that Tesla was really the first company in about a century in the U.S., the first U.S. company in the U.S. to reach volume production and be sustainably profitable. The crazy thing is this has really not happened in a hundred years. That's the actual super hard part, and we now have four vehicles in volume production, S3XY. Also, the toughest joke I think maybe ever. It was a very difficult joke to make. Elon Musk: (01:10:38) So we also introduced the lowest cost solar in the U.S. It's only a dollar 49 a watt, and we really just simplified the whole value chain, so reduced sales and advertising, got rid of a bunch of unnecessary costs, and really are just relying upon the fact that it's just the lowest cost, most efficient solar in the U.S., providing both a retrofit and the solar glass roof, which I think is a really great product. A hard product to make work, but it will be a major pipeline in the future. Elon Musk: (01:11:13) And we also got four consecutive quarters of gap profitability, which was very difficult. Yeah. And certainly a testament to the hard work of people at Tesla. I mean, to do this in extremely difficult times against a wide range of adverse circumstances was insanely hard, but we got it done, and I think the future is looking I think, very promising from a sort of an annual profitability standpoint. So in order to sort of do well financially, you really need economies of scale, and you need ideally the best technology, and I think we've had the best technology for a while, but now we are also achieving economies of scale, and we're also rapidly improving autonomy, which is a massive value add to each car. So, I think the value of Tesla is going to be like total, just on the vehicle side, total vehicles produced times the value of autonomy. That's a way to think about the future value of Tesla. Elon Musk: (01:12:35) We also have consistent free cashflow generation. This is really important for growth, and a key element here is tightening up the time from when a car is ordered to when it is built and delivered. So for a company that is growing rapidly, it's extremely important to tighten the supply chain and to have, from when parts arrive, put it into a car very quickly and deliver the car very quickly to the customer. And if you can do that inside soft of your payables timeline, then the faster you grow, the more cash you have. Or conversely, if you're unable to do it within your payables timeline, the faster you grow, the less money you will have, which is obviously bad for capital intensive situation. So just tightening up and having the parts move very quickly to the factory, put it in a car, get it to a customer makes a massive difference to cashflow generation. Elon Musk: (01:13:34) I mean, that's why it's extremely important to have a factory in each continent, because if you don't at least have a factory in the continent, it isn't impossible to achieve this. So having a factory in China, that's able to serve China, and then soon many other countries in the region will be key to us tightening that total sort of chain of cashflow, and essentially the faster we grow, the more cash. This is really important. That's also why it's important to have Giga Berlin complete, because then we'll have a factory in China, a factory in the U.S. and soon a second factory in the U.S. in Austin, and a factory in Europe. Elon Musk: (01:14:18) I mean, even if for Giga Texas in Austin, even if we had exactly the same cost as in California, it would still be advantageous to do it there because it's roughly two-thirds of the way across the U.S., so in terms of delivering cars to the central U.S. and to the East Coast, it's just faster, it costs less, and it fundamentally improves our economics. So I think this is also maybe something that's not fully appreciated of just how important it is to have a factory at least on the continent or reasonably close to where the end customers is, so you can tighten that whole chain. Elon Musk: (01:14:56) Industry performance. While the rest of industry is, has gone down, Tesla has gone up, I think this speaks to ... Thanks. And so I'd like to thank all the customers for taking a chance on Tesla and buying our product and really hope you're enjoying it. This is really, our sales, as [inaudible 01:15:21] was saying, it really grew by word of mouth, so this is really, I think it's very pure in the sense that it's growing on the basis of existing owners recommending it to others to new customers. This is, really, I think, a good way to grow. Elon Musk: (01:15:40) So, and then in 2019 we had 50% growth, and I think we'll do really pretty well in 2020. Probably somewhere between 30-40% growth, despite a lot of very difficult circumstances. I mean, there's so many. Pandemic, the wildfires. It's a whole bunch of difficult production issues, but thanks to the hard work of the Tesla team and a lot of innovative approaches to overcoming issues, we're able to still see significant growth in one of the most difficult. In fact, I'd say probably the most difficult year of Tesla's existence. Elon Musk: (01:16:25) We also published our extended impact report. At Tesla, we try very hard to do the right thing. If what I think does not happen, it's just because we maybe made a mistake or weren't aware of it, but we always try to do the right thing to the best of our ability, and then we published the extended impact report to show just a self-examination of, okay, what are we doing, right? What are we doing wrong? What can we do better in the future? We're definitely trying to accomplish the most good, and so if we occasionally make a mistake, we work quickly to fix it and do the right thing. So it's worth looking at the average life cycle of emissions in the U.S. and just how much better a Tesla is or electric car than any kind of gasoline car, and what we'll talk about in the Battery Day is also just how much the grids around the world, and actually especially in the U.S., are greening. It's actually much faster than I think people realize, the U.S. is moving towards sustainable energy. And so as we move more and more to sustainable energy, then effectively you end up building the solar factories and the car factories themselves with solar or with sustainable energy. Over time, you will even mine with sustainable energy, and eventually it will get to an effective emissions of zero, so that's where things will end up. Yeah. Elon Musk: (01:17:59) So we also have safety at the core of our design. The Tesla cars are the safest cars ever designed. We have the lowest probability of injury of any cars ever tested by the U.S. government, And that's just passive safety. When you add active safety into that, it's even better, so it's really ... If safety is important to you, which obviously it is, the safest car you could drive is a Tesla. So I think some people aren't aware of this, but it's really safety is paramount. It is actually the number one design objective when we build a Tesla is safety. Elon Musk: (01:18:41) Our factories are also becoming safer, and if you look at the sort of accidents per vehicle, total vehicle made it's dramatically better than in the past, and it's already better than industry average, and we're confident we can get it to the best in the auto industry. Autopilot functionality continues to improve, and you can see it in the safety report that we publish every quarter. It's just getting better and better. The U.S. average for collisions is at roughly 2.1 per million miles, and with autopilot engaged, it's 0.3. I mean, this is a profound difference, really massive, and this will get even better. So we're confident that over time we can get the probability of an accident, especially the probability of injury, to 10 times better than the industry average, like an order of magnitude better. So that's just a lot of lives saved and a lot of injuries avoided, so that's a huge priority for us. Elon Musk: (01:19:50) Yeah, the autopilot front, I think it's hard for people to judge the progress of autopilot. I'm driving ... As a matter of course, I've always done this. I drive the bleeding edge alpha build of autopilot, and so I sort of have insight into what is going on. Previously about a couple of years ago, we were kind of stuck in a local maximum, so we're improving, but the improvements kind of started tailing off and just not getting where they needed to be. I call this sort of getting trapped in a local maximum, and so we had to do a fundamental rewrite of the entire autopilot software stack and all of the labeling software as well. Elon Musk: (01:20:40) So we are now labeling in 3D video, so this is hugely different from the previously where we were labeling essentially a bunch of single images from the eight cameras, and they would be labeled at different times by different people, and some of the labels, you literally can't tell what it is you're labeling. So it basically made it sort of in some cases impossible to label, and the labels had a lot of errors. Now with our new labeling tools, we label it in video, so we actually label entire video segments in the system, so you get basically a surround video thing to label with the surround video and with time. So it's now taking all cameras simultaneously and looking at how the image has changed over time and labeling that, and then the sophistication of the neural nets in the car and the overall logic in the car has improved dramatically. Elon Musk: (01:21:44) I think we'll hopefully release a private beta of autopilot, of the full self-driving version of autopilot in, I think, a month or so, and then people will really understand just the magnitude of the change. It's profound. So, yeah. Anyway, so you'll see it. It's just like a hell of a step change, but because we had to rewrite everything, labeling software, just the entire code base, it took us quite a while. The sort of new ... I call it like 4D in the sense that it's three dimensions plus time. It's just taken us a while to rewrite everything, and so you'll see what it's like. It's amazing. Yeah. It's just clearly going to work. Elon Musk: (01:22:42) At Tesla, the core competencies, we've got engineering, obviously, but also manufacturing. I think manufacturing is underappreciated in general, and the difficulty of designing the machine that makes the machine is vastly harder than the machine itself. So the designing, like making a Model 3 or Model Y or Cybertruck truck prototype is really quite trivial compared to designing the factory that makes it, especially if it's new technology, and you want to use new manufacturing methods. It's just at least 10 to 100 times harder to do the factory than the prototype, and that's why you see a lot of companies out there or startups they'll bring out a prototype, but they just can't get it over the hump for who manufacturing, because manufacturing of new technology especially is the hardest thing by fa. Basically, the prototype is at best 10% of the difficulty and probably closer to 1%. Elon Musk: (01:23:50) And then software. Tesla is both a hardware and a software company, so a huge percentage of our engineers are actually software engineers, and you can think of our car as kind of like a laptop on wheels, so software is incredibly important. Actually, not just in the car, but also in the factory. So the factory software is extremely important. Just software in general. I mean, these are fundamental. These are the three critical areas that are needed to make for an awesome company. So, yeah. Elon Musk: (01:24:29) So we have ... Now we'll soon have three new factories incremental on ... Well, we have one already. On three different continents. Shanghai, we're expanding the Shanghai with the second phase. Berlin is making rapid progress, and Texas is making even faster progress. So, yeah. With each factory, what we're trying to do is also improve the manufacturing technology, so in some cases like the Model Y made in Berlin might look the same, but it actually is made in a much more efficient way. Yeah, we'll talk about that later in the battery presentation. Elon Musk: (01:25:15) Yeah, we launched Megapack. It's three megawatt hours all in one energy storage solution, so it's been great overall. Yeah. All right. And I think that's basically it, right? All right, thank you. All right. Well, thanks, everyone, for coming, and we'll be back in a little bit to go through the battery stuff, and there's a little bit more. In addition to the battery stuff, we've got a few extras as well. So I think you'll really like what we have to say on batteries. Elon Musk: (01:25:53) The battery stuff we're going to talk about is truly revolutionary and essential to Tesla's goal. The fundamental good of Tesla, it's like, if you look back in history and say, \"What good did Tesla do?\" The good will by how many years did we accelerate sustainable energy? That's the true metric of success. It matters if sustainable energy happens faster or slower, and so that's really how I think about Tesla and how we should assess our progress. By how many years did we accelerate sustainable energy? And what we're going to talk about with batteries and a few other things will really explain how we're going to make a step change improvement in the acceleration of sustainable energy. Thank you. [inaudible 00:18:44]. Speaker 1: (01:26:50) Hi, folks. That was great. We're going to take a short break before we begin the Battery Day event, so stay tuned. If you're local and here in the audience today, you can feel free to get out of the cars and stretch your legs, but try to stay near the cars, because we're going to begin properly in a little bit. See you soon. (silence). Elon Musk: (01:27:06) [inaudible 01:40:28]. Drew Baglino: (01:40:32) Hello, everyone. Elon Musk: (01:40:37) Great. Should you start? Drew Baglino: (01:40:38) Sure. Thanks, Elon. Hi. I'm Drew Baglino, SVP of Powertrain and Energy Engineering at Tesla, and I'm incredibly excited to talk about what we've been doing with batteries here at Tesla. Elon Musk: (01:40:48) Great. So let's see. You've got the clicker? Drew Baglino: (01:40:53) I've got the clicker, yeah. Elon Musk: (01:40:54) Okay. Let's ... Yeah. I'll take it at first, perhaps. Drew Baglino: (01:40:57) Sure. Elon Musk: (01:40:58) So obviously the issues we're facing are very serious with climate change, and we're experiencing these issues on a day-to-day basis. It's incredibly important that we accelerate the advent of sustainable energy. Time really matters. This presentation is about accelerating the time to sustainable energy. Elon Musk: (01:41:23) So the past five years were the hottest on record. We have what looks like a wall for CO2 PPM. It's obviously ... This time is not like the past. It's really important that we take action. Running this climate experiment is insane, so ... Drew Baglino: (01:41:46) Especially when it's just a transitory one, anyway. Elon Musk: (01:41:49) Yes. Drew Baglino: (01:41:50) We're going to run out of these fossil fuels. Let's just move to the future and not run this experiment any longer. Yeah. Elon Musk: (01:41:55) Talk a bit louder. Drew Baglino: (01:41:56) You got it. Elon Musk: (01:41:57) Okay. So anyway, there is a lot of good ... Elon Musk: (01:42:03) There is a lot of good news though. A lot of people may not be aware that that wind and solar comprise 75% of new electricity capacity in the US this year. So this is really major. So the grid is going sustainable very quickly. Now, it's also worth noting that the length of time that power plants lasts is on the order of 25 years. So even if a hundred percent of energy generation was sustainable, it will still take 25 years to convert the grid. And it's also worth noting that in the past 10 years, power production from coal has dropped in half. So it went from 46% of electricity in 2010 to 23% in 2020. So this is a massive improvement. So good things are happening on a lot of levels. We just need to go faster. Elon Musk: (01:43:06) So Tesla's contribution, we've delivered over a million electric vehicles, 26 billion electric miles driven, and many gigawatt hours of stationary batteries, 17 terawatt hours of solar generated. So I think solar is sometimes underweighted at Tesla, but it is a massive part of our future. The three parts of a sustainable energy future are sustainable energy generation, storage, and electric vehicles. So we intend to play a significant role in all three. So to accelerate the transition to sustainable energy, we must produce more EVs that need to be affordable and a lot more energy storage, while building factories faster and with far less investment. So goal number one is a terawatt hour scale battery production. So tera is the new giga. And a terawatt is a thousand times more than a gigawatt. So we used to talk in terms of gigawatts, in the future, we'll be talking in terms of terawatt hours. So this is what's needed in order to transition the world to sustainability. Drew Baglino: (01:44:24) Yeah, and you can see it's... We're talking about a hundred X growth in batteries for electric vehicles to achieve this mission. And we are going to get there. It's just a matter of how fast. And our intention is to accelerate it. Elon Musk: (01:44:38) Yeah, you basically need on the order of roughly 10 terawatt hours a year of battery production to transition the global fleet of vehicles to electric. Drew Baglino: (01:44:48) And the average vehicle lasts 15 years. So we're talking about 150 terawatt hours give or take to transition the whole electric, all vehicles of all types, to electric. Elon Musk: (01:45:00) Yeah. So it's a lot of batteries, basically. And so- Drew Baglino: (01:45:07) Yeah. And then on the grid side, we have a similar mountain to climb, 1600 times growth from today's grid batteries to go a hundred percent renewable on the grid and to take all of the existing heating fossil fuel uses in homes and businesses, a hundred percent electric. Elon Musk: (01:45:24) Yeah. And this number I think might grow even more. As the world economy matures, and as countries with high populations industrialize, we could see this number be even more. But let's say it's like roughly 20 to 25 terawatt hours per year sustained for 15 to 25 years to transition the world to renewable. This is a lot. Drew Baglino: (01:45:53) Yeah. Elon Musk: (01:45:55) So today's batteries cannot scale fast enough. They're just too small. For Giga Nevada, 150 gigawatt hours per year is what we probably expect to make out of there. But this is really pretty small in the grand scheme of things. That's only 0.15 terawatt hours. And it costs too much. Drew Baglino: (01:46:16) We would need 135 fully built out in Nevada Giga factories to achieve 20 terawatt hours a year. It's not scalable enough of a solution. We need a dramatic rethink of the cell manufacturing system to scale as fast as we can and should. Elon Musk: (01:46:32) Yeah, and I think we should view this as more than just a question of money. Money is sort of an ethereal thing, but it's really the amount of effort. You have a certain amount of effort in terms of people and machines, and depending on how efficient that effort is, for a given amount of effort, you want the most amount of batteries. So it's not just the question of well, if we have $2 trillion, tomorrow you could make this. It's not that easy. You actually need to organize a massive number of people, build a lot of machines, build the machines that make the machines. And so it's incredibly important to have that effort yield the most number of batteries. Elon Musk: (01:47:16) So, and then goal two, obviously we need to make more affordable cars. I think one of the things that troubles me the most is that we don't yet have a truly affordable car, and that is something that we will make in the future. But in order to do that, we've got to get the cost of batteries down. We've got to make, and we've got to be better at manufacturing. And we need to do something about this curve. The curve of the cost per kilowatt hour of batteries is not improving fast enough. So we've given this a lot of thought over many years to say, okay, how can we radically improve the cost per kilowatt hour curve? It's been somewhat flattening out actually in recent years. Drew Baglino: (01:48:02) Yeah. I mean, early growth was promising, but you can see we're kind of plateauing. So that's what's motivating us to rethink how cells are produced and designed. Elon Musk: (01:48:10) Yeah, exactly. So yeah. And EV market share is growing, but EVs still aren't accessible to all. And you can see, as you Drew were saying, it's like starting to flatten out a little bit because the rate of improvement of the affordability of cars is just not fast enough. So that's why we've got Battery Day. Drew Baglino: (01:48:33) Yeah. To make the best cars in the world, we designed vehicles in factories from the ground up. Next. And now we do this for batteries as well. Elon Musk: (01:48:45) Yeah. It's weird, the slides don't show up quite right here. What shows up on the screen is not quite what shows up there. Drew Baglino: (01:48:55) Oh, okay. Elon Musk: (01:48:56) It's different. Drew Baglino: (01:48:57) Yeah. I think it's because that's... Yeah. Elon Musk: (01:48:59) That one's current, supposed to be current. Whatever. Drew Baglino: (01:49:02) So let's get started. We have a plan to have the cost per kilowatt hour. And it's not a plan that rests on a single innovation, some research project that will never see the light of day. It's a plan that has taken creative engineering and industrialization across every facet of what makes a cell into a battery pack, from raw material to the finished thing. And we're going to go through that plan with you today, step-by-step, and build up how we get to these goals and how we accelerate this transition and make our vehicles and our grid batteries more affordable. Elon Musk: (01:49:45) Yeah. I mean, we basically thought through every element of the battery, or almost every element. There are a few more elements that we won't get to today, but we will get to in the future. Drew Baglino: (01:49:53) Yes. So first before we get too far into it, let's talk about what is in a battery cell. We've got the cap and the can, negative and positive terminals of the cell. When you open that cell, you've got a tab connected to those terminals, what we call the jelly roll, which is the wound electrodes on the inside. You can actually see what this looks like as you unwind it. This is over a meter long in a typical 2170 cell. So it's quite a long winding process. And you can see the tab still there. And then to explain what's actually going on here, we've identified, we've got anode, cathode, separator, positive and negative terminal. Drew Baglino: (01:50:37) Watch what happens as we, there we go, discharge the cell. Got lithium moving from anode to cathode. And then the reverse when we charge the cell, lithium moving from cathode to anode across the separator. This is the basic of what makes all lithium-ion batteries, no matter what the form factor is. And when we look at what's happened today, at least in our products, we've moved from the 18650 form factor to the 2170 form factor through great collaboration with our partners, Panasonic, new partners like LG and CATL and probably others in the future. Elon Musk: (01:51:20) Actually, slight note on why is the one called 18650, although not on the slide, versus the 2170, is that the first two digits refer to the diameter, and the second two digits refer to the length. So that helps explain what's up with these weird numbers. But nobody could explain to me why there was an extra zero. So I, so I said, \"Okay, well, we're deleting the zero that nobody can explain in future form factors.\" So that's why it's technically, it's like the 18650 bizarrely, but going forward it's the 2170, because we just got rid of the extra zero because it's pointless. Drew Baglino: (01:51:56) And this was a evolutionary step going from 1865 to 2170, bringing 50% more energy into the cell. But when we look to the ideal cell design, if we were to do it ourselves, we need to go beyond just what we're looking at us in front of us and study the full spectrum of options. So as you can see, we kind of swept the key figures of merit, how much we can reduce the cost and how much vehicle range increases as we change the outer diameter of the cell. We found a sweet spot somewhere around 46 millimeters. But it's not just about a bigger form factor. Anybody could make a bigger form factor. Elon Musk: (01:52:37) Any fool, any fool could make a bigger form factor. Are we not any fool? Drew Baglino: (01:52:42) Yeah, exactly. There are problems as you make cells larger. In fact, supercharging and thermals in general become really challenging as you make bigger cells. And this was the challenge that our team set our sights on to overcome. And we did, we came up with this tabless architecture that maybe you've heard about, that basically removes the thermal problem from the equation and allows us to go to the absolute lowest cost form factor and the simplest manufacturing process. And this is what we mean when we talk about tabless. It's kind of a beautiful thing. Elon Musk: (01:53:22) Yeah. That's what these t-shirts mean, but it's very esoteric. It was like, nobody could figure it out. Drew Baglino: (01:53:26) Yeah, we basically took the existing foils, laser pattered them, and enabled dozens of connections into the active material through this shingled spiral you can see with simpler manufacturing, fewer parts, 50 millimeter versus 250 millimeter electrical path length, which is how we get all the thermal benefits. Elon Musk: (01:53:46) Yeah. This is important to appreciate. Basically the distance that that electron has to travel, it's just much less. So you actually have a shorter path length in a large tabless cell than you have in the smaller cell with tabs. This is a big deal. So even though the cell is bigger, it actually has more power. The power to weight ratio is actually better than the smaller cell with tabs. This is, again, this is quite hard to do it. Nobody's done it before and it really took a tremendous amount of effort within Tesla engineering to figure out how do we make a frigging tabless cell and have it actually work and then connect that to the top cap. There's a whole bunch of things that we're keeping a little secret sauce here that we're not telling everything, but- Drew Baglino: (01:54:40) Sometimes what's elegant and simple is still hard. And it took us a lot of trials, but we're happy where we ended up. Elon Musk: (01:54:46) Yeah. I mean, everything is simple in recollection, after you... it's hard until it's discovered and then it's simple. So anyway, there's a lot of really cool things going on that enable tabless. And it was really due to a really great engineering team. Drew and the rest of the team had done amazing work in achieving this tabless construction. I think it may sort of sound a bit silly to some people, but for people that really know cells, this is a massive breakthrough. Drew Baglino: (01:55:19) For cylindricals to be able to get rid of the tabs dramatically simplifies winding and coding. And has an awesome thermal and performance benefit. Elon Musk: (01:55:28) Yeah. Just to elaborate on that a bit, it's like when the cell is going through the system, it has to keep stopping where all the tabs are. So you can't do a continuous motion production if you have tabs. You have to keep stopping and then there's a rate at which you can start and stop and accelerate again and it really slows down the rate of production. And then sometimes you get the tabs wrong and you also lose a little bit of active area. It's really a huge pain in the ass to have tabs from a production standpoint. Drew Baglino: (01:56:03) Yes. And so when we put it all together and go to our new 80 millimeter length, 4680 we call this a new cell design, we get five times the energy with six times the power and enable 16% range increase, just form factor alone. Elon Musk: (01:56:23) Yeah. So these... Yeah. It's pretty great. And just to clarify, when we see these plus 16% or whatever the percentage rate increase is, these are the amounts due just to that particular innovation. So we'll list a whole bunch of innovations and then when you add them up, you get a total improvement in energy density and cost. But these numbers are what refer to just this thing. Drew Baglino: (01:56:56) Yeah. And I want to stress, this is not just a concept or a rendering. We're starting to ramp up manufacturing of these cells at our pilot 10 gigawatt hour production facility, just around the corner. Elon Musk: (01:57:08) Yeah. So. Yeah. It's a video of some of what's going on in the plant. Now. I mean, to be clear, it will take about a year to reach the 10 gigawatt hour capacity. So this is important to appreciate. When you build a factory, there's a certain capacity that you design to, and then it takes some period of time to actually achieve that capacity. So I would say it's probably about a year before we get to the 10 gigawatt hour annualized rate with the pilot plant. And this is just a pilot plant. The actual production plants will be more on the order of maybe 200 gigawatt hours, maybe more over time. Drew Baglino: (01:58:00) And... Thank you. But let's stack up everything we just saw at the cell level. So just the cell form factor change enables a 14% dollar per kilowatt hour reduction, just that cell form factor change. And now that you've been teased on this factory, we're going to go on and walk step-by-step through that factory and discuss a series of innovations there. When thinking about the ideal cell factory, we have inspirations behind us in the paper and bottling industry, where from humble beginnings, over a century of innovation has enabled mass scale, continuous motion, unbelievably low manufacturing costs. And when we think about the lithium-ion industry, which is really only in its third decade of high volume production, it has so far to go to achieve similar scale and simplicity. And that was the inspiration that we set out to the team as we thought about how to marry cell design and manufacturing in the best possible factory. Drew Baglino: (01:59:05) And let's talk a little bit about what's in a cell factory. First, there's an electrode process where the active materials are coated into films onto foils. Then those coated foils are wound in the winding process we just talked about where if you do have tabs, you have to start and stop a lot. Then the jelly roll is assembled into the can, sealed, filled with electrolyte, and then sent to formation where the cell is charged for the first time and where the sort of the electrochemistry is set and the quality of the cell is verified. And we set out at every step of this process to try to take that inspiration we just showed and think about how we make those processes fundamentally better and more scalable. And one of the most important processes is where it all begins, the wet process of the electrode coding. And just to give you all a sense of scale, I'm going to walk through what's in that wet process. Drew Baglino: (02:00:09) You've got mixing where the powders are mixed with either a water or a solvent, solvents for the cathode. That mix then goes into a large coat and dry oven where the slurry is coated onto the foil, huge ovens, tens of meters long, dried, and that solvent then has to be recovered. You can see the solvent recovery system. And then finally the coated foil is compressed to the final density. And when you're looking at this, you're like, wow, that's a lot of equipment for one step, especially when you consider that little spec next to the coating oven is a person. This is serious iron involved in making batteries. Wouldn't it be great if we could skip that solvent step, which is one of those dig a ditch, and then fill it kind of things where you put the solvent in and then take it out and recycle it, and just go straight to a dry mix to coat? And that's what the dry process really is about. And in the most basic form, you can see it here on a benchtop, literally powder into film, as simple as that. Elon Musk: (02:01:25) I mean, it's hard actually, just to be clear. If this was easy, everyone would do it. It's not like dry coating electrode is actually easy. It's actually very hard to do what appears to be a simple thing. And it's worth noting, we did acquire Maxwell a little over a year ago, I guess, and certainly a good company and everything, but the dry coating they had was like, it's like sort of, I would call proof of concept. Since the acquisition. We've actually ramped the machine that does dry coating four times. So revision full post acquisition of the machine, and there's still a lot of work to do. So I would not say this is completely in the bag. It's still a lot of work to do. And as you grow, as you scale, go from benchtop to lab to pilot to volume production, there are actually major issues that you encounter at every level. It's not like you make something work on your bench and bingo, now you can make a bazillion of it. Drew Baglino: (02:02:26) Absolutely. Elon Musk: (02:02:26) It's insanely difficult to scale up. Yeah. Drew Baglino: (02:02:31) Yeah, but if you do scale it up, what you saw before becomes this. So you can see the motivation. A 10 times reduction in footprint, a 10 times reduction in energy and a massive reduction in investment. But as Elon was saying, simple is hard. Elon Musk: (02:02:48) Yeah. I mean, to be clear, I would like to not say that right now, it just totally working. It's close to working, but it's not, even now at the pilot plant level, it is close to working. It's fair to say probably it does work, but with not a good, not a high yield. Drew Baglino: (02:03:07) Yeah. We're still ironing out the kinks, but we've made tens of thousands of cells, thousands of kilometers of electrode. I mean, we are on the fourth generation of the equipment so we've learned a lot along the way. I mean, it is super demanding because every atom has its place if you want to deliver the energy density and the cycle life and the supercharging. But we're confident that we will get there, but it will be a lot of work along that. Elon Musk: (02:03:29) There's a clear path to success, but a ton of work between here and there. But this is a really profound improvement. Again, for people that know battery manufacturing, this is gigantic. We'll probably be on machine revision six or seven by the time we do large scale production. The rate at which the machines are being improved is extremely rapid. Literally every three or four months, there's a new rev. Drew Baglino: (02:03:55) Yeah. And beyond the electrode, we continue to innovate on every other process steps. So let's talk a little bit about assembly, which is next. The key to a high-performing assembly line is accomplishing processes while in motion, continuous motion. And thinking of the line as a highway, max velocity down the highway, no start and stop, no city driving. Elon Musk: (02:04:24) Exactly, no stop lights and traffic lights sort of thing. You want the highway. Drew Baglino: (02:04:28) You want the highway. And together with our internal design team that makes this equipment and designs this equipment, we coupled thinking about how to make the best cell with thinking about how to make the best equipment so that we could accomplish the fastest parts per minute rates on all of these tools. And through all of that development, we were able to get to the point where we can implement assembly lines, one line, 20 gigawatt hours, seven times increase in output per line. And when you're thinking about scalability and pure effort, having one line be seven X the capability is just effort multiplying. Elon Musk: (02:05:10) Yeah. So you can sort of think about the sort of the fundamental physics of a factory or something. I think it's actually quite a lot like the rocket equation where you've got basically the rocket equation you've got your exhaust velocity and then the log of [end 02:05:24] masses. So it's basically saying how fast are things going and what percentage of the factory volume is doing useful work? And conveyance does not count as useful work. Drew Baglino: (02:05:34) Only the value added steps. Elon Musk: (02:05:37) Yeah. If you break the factory down into cubic meter sections and say... or smaller. Could be like one liter sections, and say, \"Is a majority of this volume of doing useful work?\" You'd be astounded at how bad most factories are. They'd be like maybe two or 3%, including our factory in Fremont. So I think it's possible to get to at least 10 times that of volumetric efficiency. So more like 30%ish, maybe more, and be 10x better, which means the factory can be 10 times smaller. And then the other thing is how fast are things going through the factory? It's like speed and density. A factory that's moving at say twice the speed of another factory is equivalent to two factories basically. And the company that will be successful is the company that with one factory can accomplish what other companies take two or three or four factories to do. So this is what we're trying to do here is say, okay, how do we, with one factory achieve what maybe five or even 10 factories would normally be required to achieve? Drew Baglino: (02:06:43) And the vertical integration with the machine design teams at Grohmann and Hibar and others allows us to really accomplish that because we don't have any of these edge conditions between one piece of equipment and another, we can design the entire machine to be one machine and remove all of these unnecessary steps. Elon Musk: (02:07:03) Yeah. I mean, basically Tesla is aiming to be the best at manufacturing of any company on Earth. This is the thing that's actually most important in the long run I think, just from a company standpoint and from basically achieving sustainability as fast as possible. But I think also for long-term competitiveness, eventually every car company will have long range electric cars. Eventually every company will have autonomy, I think, but not every company will be a great at manufacturing. Tesla will be absolutely head and shoulders above anyone else in manufacturing, that is our goal. Drew Baglino: (02:07:44) Manufacturing is hard and hard problems are fun to solve. Okay. Now let's talk about formation. In a typical cell factory, formation represents 25% of the investment. And what is formation? Is it's charging and discharging cells and verifying the quality of the cell. It turns out we've charged and discharged billions and billions of cells in our vehicles so we know a thing or two about that. The typical formation set up is you charge and discharge each cell individually. In our car, we charge thousands of cells at once. And we took our principal and our power electronics, leveraging Powerwall vehicle battery management systems and others to dramatically improve the formation equipment cost-effectiveness and density. 86% reduction in formation investment, 75% reduction in footprint. You want to take this one? Elon Musk: (02:08:43) Sure. So essentially what this translates to based on what we know today is about a 75% reduction in the investment per kilowatt hour. Or gigawatt hour. It's just basically four times better than the current state of the art to the best of our knowledge. And I think there's probably room to improve even beyond that. Drew Baglino: (02:09:04) Definitely. Elon Musk: (02:09:05) Definitely. Yeah. So we're able to, from a volume standpoint, actually get what, in a smaller form factor than Giga Nevada, we're able to get many times the cell output. So you can see basically we can get a terawatt hour in less space than it took to make a gigawatt hour, 150 gigawatt hours. So this is pretty profound. I would actually not have thought this was possible several years ago, that we could actually get to a terawatt scale in less space than what we currently envisioned for doing 150 gigawatt hours. Drew Baglino: (02:09:48) Yes. Simpler accelerates terawatt scale. And that's what we need to do to accelerate our mission. And as Elon said, we're going to try to even improve on this as we push towards our goals, which are... Elon Musk: (02:10:02) Yeah. So this is just talking about Tesla internal cell production. As I tweeted out earlier, we will continue to use our cell suppliers, Panasonic and LG and CATL. And so this is a hundred gigawatt hours supplemental to what we buy from suppliers. And yeah, essentially, this does reduce our weighted average cost of a sale, but it allows us to make a lot more cars and a lot more stationary storage. And then long-term, we're expecting to make on the order of a 3000 gigawatt hours or three terawatt hours per year. I think we've got a good chance of achieving this actually before 2030, but I'm highly confident that we could do it by 2030. Drew Baglino: (02:10:58) When you look at the size of that factory on the previous page, it really shows how enabling all of these advancements are in achieving a three terawatt hour goal by 2030. And not only is all of that manufacturing innovation fantastic for enabling scale, it's also an additional 18% reduction in dollar per kilowatt hour at the battery pack level. Elon Musk: (02:11:18) But wait, there's more. Drew Baglino: (02:11:19) But wait, there's more. So we have a manufacturing system, we've got a cell design. What are the active materials we're going to put in that cell design? Let's talk about the anode first. Let's talk about silicon. Why is silicon awesome? It's awesome because it's the most abundant element in the Earth's crust after oxygen, which means it's everywhere. It's sand. Elon Musk: (02:11:44) Sand is silicon dioxide. Drew Baglino: (02:11:47) And it happens to store nine times more lithium than graphite, which is the typical anode material in lithium-ion batteries today. So why isn't everybody using it? The main reason is because the challenge with silicon is that it expands four X when fully charged with lithium. And basically all of that expansion stress on the particle, the particles start cracking, they start electrically isolating, you lose capacity. The energy retention of the battery starts to fade. And it also gumps up with a passivation layer that has to keep reforming as the particles expand. Elon Musk: (02:12:19) Yeah. Basically with silicon, the cookie crumbles and gets gooey. That's basically what happens. Drew Baglino: (02:12:24) Good analogy. And current approaches to solve this, which exist, I mean, we have silicon in the cars that you're all in right now, involved highly engineered, expensive materials in the scheme of things. Now they're still great and they enable some of the benefits of silicon. They just don't enable all of it and they're not scalable enough. And you can see some of the things that maybe you've heard of, SIO, silicon with carbon, or silicon nanowires. That's kind of the space right now. What we're proposing is a step change in capability and a step change in cost. And what that really is is to just go to the raw metallurgical silicon itself, don't engineer the base metal, just start with that and design for it to expand in how you think of the particle in the electrode design and how you coat it. Elon Musk: (02:13:14) Yeah. And I'm not sure if you saw this. Basically a dollar per kilowatt hours basically. If you use simple silicon, it's dramatically less than even the silicon that is currently used in the batteries that are made today, and you can use a lot more of it. Drew Baglino: (02:13:29) The anode would cost, yeah, with this silicon, the anode costs a dollar and 20 cents a kilowatt hour. Elon Musk: (02:13:38) Yeah. Drew Baglino: (02:13:41) And how does it work? Start with raw metallurgical silicon, stabilize the surface with an elastic ion conducting polymer coating that is applied through a very scalable approach. No chemical vapor deposition, no highly engineered high capacity, high cap ex solutions, and then integrate it into the electrode through a robust network formed out of a highly elastic binder. And in the end, by leveraging this silicon to its potential, we can increase the range of our vehicles by an additional 20%. Just this improvement. Elon Musk: (02:14:13) Yeah. It gets cheaper and longer range. Okay. Drew Baglino: (02:14:20) Yeah. And when we take that anode cost reduction, we're looking at another 5% dollar per kilowatt hour reduction at the battery pack level. And there's more. Let's talk about cathodes. What is a battery cathode? Cathodes are like bookshelves where the metal, the nickel, the cobalt, the manganese or aluminum is like the shelf, and the lithium is the book. And really what sets apart these different metals is how many books of lithium they can fit on the shelves and how sturdy the shelves are. Cobalt is a- Elon Musk: (02:14:53) Sorry, I was going to say it's tough to exactly figure out what the right analogy is to explain a cathode and anode. But a bookshelf is probably a pretty good one in the sense that you need a stable structure to contain the ions. So you want a structure that does not crumble or get gooey, or basically that that holds its shape in both the cathode and the anode. As you're moving these ions back and forth, it needs to retain its structure. So if it doesn't retain a structure, then you lose cycle life and your battery capacity drops very quickly. Drew Baglino: (02:15:30) Absolutely. Yeah. I totally agree. And I think people are always talking about like, oh, what's the catheter going to be? Is it [NCA 02:15:38] or whatever? The thing to consider is just fundamentally what the nickel, the metals are capable of. And that's what we have on the chart here. Dollar per kilowatt hour cathode of just the metal using just LME, London Metal Exchange prices, versus the energy density of just the cathode. And you can see nickel is the cheapest and the highest energy density. And that's why increasing nickel is a goal of ours and really everybody's in the energy- Drew Baglino: (02:16:03) Why increasing nickel is a goal of ours, and really everybody's in the battery industry. But one of the reasons why cobalt is even used at all is because it is a very stable bookshelf. And the challenge with going to pure nickel is stabilizing that bookshelf with only nickel. And that's what we've been working on with our high nickel cathode development, which has zero cobalt in it, leveraging novel coatings and dopants We can get a 15% reduction in cathode dollar per kilowatt hour. Elon Musk: (02:16:31) Yeah. Big deal. Drew Baglino: (02:16:38) But it's not just about nickel. You want a? Elon Musk: (02:16:41) Yeah. Sure. So in order to scale, we really need to make sure that we're not constrained by total nickel availability. I actually spoke with the CEOs of the biggest mining companies in the world and said, \"Please make more nickel, this is very important.\" And so I think they are going to make more nickel. I think we need to have a kind of a three-tiered approach to batteries. Elon Musk: (02:17:07) So starting with iron, that's kind of like a medium range, and then nickel manganese as sort of a medium plus intermediate and then high nickel for long range applications like Cyber Truck and the semi. Something like a semi-truck, it's extremely important to have high energy density in order to get long range. And just to give sort of iron up a bit more time, if you look at [inaudible 02:17:37] per kilogram at the cathode level of iron, it looks like nickel's twice as good, but when you fully consider it at the pack level, everything else taken into account, nickel is about maybe 50 or 60% better than iron. Elon Musk: (02:17:52) So iron is a little better than it would seem, when you look at it at the pack level fully considered. It's not as good as nickel, nickel's like 50 to 60% better, but it's actually pretty good. Good for stationary storage and for medium range applications where energy density is not paramount. And then, like I said, for intermediate, it's kind of a nickel manganese, and it's a relatively straightforward to do a cathode that's two-thirds nickel one third manganese, which would then allow us to make 50% more cell volume with the same amount of nickel. Drew Baglino: (02:18:32) And with very little energy trade-off. Just enough to have, you still want to use 100% nickel for something like a semi-truck, but really not much of a sacrifice at all. Elon Musk: (02:18:41) Yeah. Drew Baglino: (02:18:44) And beyond the metals, because a lot of people spend time talking about the metals. Actually the cathode process itself is a big target. 35% of the cathode dollar per kilowatt hour is just in transferring it into its final form. And so we see that as a big target. And we decided to take that on. Drew Baglino: (02:19:03) Here's a view of the traditional cathode process. Effectively, if you start at the left and you have the metal from the mine, the first thing that happens is the metal from the mine is changed into an intermediate thing called a metal sulfate, because that just happened to be what chemists wanted a long time ago. And then when you're making the cathode you have to take this intermediate thing called the metal sulfate fate, add chemicals, add a whole bunch of water, a whole bunch of stuff happens in the middle, and at the end you get that little bit of cathode and a whole bunch of wastewater and byproducts. Elon Musk: (02:19:34) It's insanely complicated. It's a small world journey of, \"I am a nickel atom, what happens to me?\" And it is crazy. You're going around the world three times, there's the moral equivalent of digging the ditch, filling the ditch and digging the ditch again, it's total madness basically. And these things just grew up, they're just kind of like legacy things, it's like how it was done before and then they connected the dots but really didn't think of the whole thing from a first principle standpoint saying, \"How do we get from the nickel ore in the ground to the finished nickel product for a battery?\" So we've looked at the entire value chain and said, \"How can we make this as simple as possible?\" Drew Baglino: (02:20:18) And that's what we're proposing here with our process. As you can see, a whole lot less is going on here. We get rid of the intermediate, metal, water, final product cathode, recirculate the water, no wastewater at all. And when you summarize all of that it's a 66% reduction in CapEx investment, a 76 reduction in process costs and zero waste water. Much more scalable solution. Drew Baglino: (02:20:48) And then when you think about the fact that now we're actually just directly consuming the raw metal nickel powder, it dramatically simplifies the metal refining part of the whole process. So we can eliminate billions in battery grade nickel intermediate production. It's not needed at all. And we can also use that same process we showed on the previous page to directly consume the metal powder coming out of recycled electric vehicle and grid storage batteries. So this process enables both simpler mining and simpler recycling. Drew Baglino: (02:21:21) And now that we have this process, obviously we're going to go and start building our own cathode facility in North America and leveraging all of the North American resources that exist for nickel and lithium. And just doing that, just localizing our cathode supply chain and production, we can reduce miles traveled by all the materials that end up in the cathode by 80%, which is huge for cost. Elon Musk: (02:21:45) Yeah. To be clear, cathode production would be part of our the Tesla cell production plant. So it would just be basically raw materials coming from the mine and from raw materials in the mine out comes a battery. Drew Baglino: (02:21:59) And on that note, the way the lithium ends up in the cell is through the cathode. So then we should obviously on-site lithium conversion as well, which is what we will do using a new process that we're going to pioneer. That's a sulfate-free process again, skip the intermediate, 33% reduction in lithium cost, a hundred percent electric facility co-located with the cathode plant. Elon Musk: (02:22:25) So it's important to note that there is a massive amount of lithium on earth. So lithium is not like oil. There's a massive amount of it, pretty much everywhere. In fact is there's enough lithium in the United States to convert the entire United States fleet to electric, all the cars in the United States. Like 300 million or something like that. Every vehicle in the United States can be converted to electric using only lithium that is available in the United States. Drew Baglino: (02:22:55) Discovered today. Elon Musk: (02:22:57) Yeah, what we already know is exist. Drew Baglino: (02:22:58) People really haven't even been looking. Elon Musk: (02:22:59) Yeah, people haven't been trying because it's just widely available. But it is important to say, \"Okay, what is the smartest way to take ores and extract the lithium and do so in an environmentally friendly way?\" And we actually discovered... Again, looking at it a first principles physics standpoint, instead of just the way it's always been done, is we found that we can actually use table salt, sodium chloride, to basically extract the lithium from the ores. Nobody's done this before, to the best of my knowledge, nobody's done this. And all the elements are reusable, it's a very sustainable way of obtaining lithium. And we actually got rights to a lithium clay deposit in Nevada. Drew Baglino: (02:23:51) Over 10,000 acres. Elon Musk: (02:23:52) Over 10,000 acres. And then the nature of the mining is actually also very environmentally sensitive in that we sort of take a chunk of dirt out of the ground, remove the lithium, and then put the chunk of dirt back where it was. So it will look pretty much the same as before, it will not look like terrible. And yeah, it'll be nice. Drew Baglino: (02:24:13) Simply mix clay with salt, put it in water, salt comes out with the lithium, done. Elon Musk: (02:24:18) Yeah. It's pretty crazy. Drew Baglino: (02:24:19) Yeah. So we're really excited about this and there really is enough lithium in Nevada alone to electrify the entire US fleet. Elon Musk: (02:24:27) Yeah, that's true. Actually, just what's in Nevada. Basically, there's so much damn lithium on Earth it's crazy. It's one of the most common elements on the planet. Drew Baglino: (02:24:39) And eventually, as we said at the beginning, when we get to this steady state 20 terawatt hours per year of production, we will transfer the entire non-renewable fleet of both power plants, home heating and industry heating and vehicles to electric. And at that point, we have an awesome resource in those batteries to recycle, to make new batteries. So we don't need to do any more mining at that point. And you can see why. The difference in the value of the material coming back from the vehicle versus the ground, you'd always go to the vehicle. And we recycle a hundred percent of our vehicle batteries today. And actually, we are starting our pilot full-scale recycling production at Gigafactory Reno next quarter to continue to develop this process as our recycling returns grow. Elon Musk: (02:25:30) To date, it's been done by third parties, but we think we can recycle the batteries more effectively, especially since our batteries, we're making the same battery as the thing we're recycling. Whereas third party recyclers have to consider batteries of all kinds. Drew Baglino: (02:25:46) Yeah. And just to think about what this actually means, the recycling resource is always 10 or greater years delayed because batteries last a really long time, but eventually it is the way that all resources will be made available. And that's why we're investing in this recycling facility in Nevada. Elon Musk: (02:26:04) Yeah. Long-term, new batteries will come from old batteries once the fleet reaches steady state. Drew Baglino: (02:26:11) Right. Okay. So we just talked about scaling cathode and recycling, all of the benefits that you just saw are added to this benefit of a 12% reduction in dollars per kilowatt hour at the battery pack level, almost at our half of the cost goal, but there's one more section. Take it away, Elon. Elon Musk: (02:26:31) So there's an architecture that we've been wanting to do at Tesla for a long time, and we've finally figured it out. And I think it's the way that all electric cars in the future will ultimately be made, it's the right way to do things. Elon Musk: (02:26:52) So it starts with having a single piece casting for the front body and the rear body. And in order to do this, we commissioned the largest casting machine that has ever been made. And it's currently working just over the road at our Fremont plant. It's pretty sweet. Currently making the entire rear section of the car as a single piece, high pressure die-cast aluminum. And in order to do this, we actually had to develop our own alloy because we wanted a high strength casting alloy that did not require coatings or heat treatment. This is a big deal for castings. Especially with a large casting. If you heat treat it afterwards it tends to deform, it kind of does this like potato chip thing. So it's very hard to keep a large casting to have its shape. Elon Musk: (02:27:47) So in order to achieve this, there was no alloy that existed that could do this, so we developed our own alloy, a special allow of aluminum, that has high strength without heat treat and is very castable. So that's a great achievement of our materials team. In fact, in general, we've got a lot of advanced materials coming for Tesla, new alloys and materials that have never existed before. Elon Musk: (02:28:10) So, you're basically making the front and rear of the car is a single piece and that then interfaces to what we call it, the structural battery. Where the battery for the first time will have dual use. The battery will both have the use as an energy device and as structure. This is absolutely the way things are done. In the early days of aircraft they would carry the fuel tanks as cargo. So the fuel tanks actually were quite difficult to carry. They're basically worse than cargo, you had to add to kind of bolt them down. It was very difficult. And then somebody said, \"Hey, what if we just make the fuel tank in wing shape?\" So all modern airplanes, your wing is just a fuel tank in wing shape. This is absolutely the way to do it. And then the fuel tanks serves this dual structure, and it's no longer cargo. It's fundamental to the structure of the aircraft. This was a major breakthrough. We're doing the same for cars. Elon Musk: (02:29:26) So this is really quite profound. Effectively the non cell portion of the battery has negative mass. So we saved more mass than the rest of the vehicle than the non cell portion of the battery. So it's like, \"How do you really minimize the mass of a battery? Make it negative. Make the non cell portion of battery pack negative.\" So it also allows us to pack the cells more densely because we do not have intermediate structure in the battery pack. So instead of having these supports and stabilizers and stringers and structural elements in the battery, we now have a lot more space in the battery because the pack itself is structural. Elon Musk: (02:30:10) What we do essentially, instead of having just a filler that is a flame retardant, which is currently what is in the 3NY battery packs, we have a filler that is a structural adhesive, as well as flame-retardant. So it effectively glues the cells to the top and bottom sheet. And this allows you to do shear transfer between upper and lower sheet. Just like if you have a formula one craft or a racing boat, and you have carbon fiber face sheets and aluminum honeycomb between them, this gives you incredible stiffness and it's really the way that any super fast thing works is you create basically a honeycomb sandwich with two face sheets. Elon Musk: (02:30:58) This is actually even better than what aircraft do. Because aircraft do not do this. They can't do this because fuel is liquid. So in our case the batteries are solid. So we can actually use the steel shell case of the battery to transfer shear from the upper and lower face sheet, which makes for an incredibly stiff structure, even stiffer than a regular car. In fact, if this was a convertible that had no upper structure, that convertible will be stiffer than a regular car. So it's just really major. Elon Musk: (02:31:38) So it improves the mass efficiency of the battery. And then those castings are also quite important because you want to transfer load into the structural battery pack in a very smooth, continuous way. So you don't put arbitrary point loads into the battery. So you want to sort of feather the load out from the front and rear into the structural battery. It also allows us to move the cells closer to the center of the car, because we don't have the... In the top one we've got all the supports and stuff, so the volumetric efficiency of the structural pack is as much better than a non-structural pack. And we're going to actually bring the cells closer to the center and because they're closer to the center it reduces the probability of a side impact potentially contacting the cells because in any kind of side impact has to go further in order to reach the cells. Elon Musk: (02:32:36) It also proves what's called the polar moment of inertia. Which is if you think of when there's a ice skater arms out or arms in. Arms in, you rotate faster. So if you can bring things closer to the center, you reduce the polar moment of inertia and that means the car maneuvers better. It just feels better. You won't know why, but it just feels more agile. So it's really cool. This is really major. Elon Musk: (02:33:03) Like it says, so 10% mass reduction in the body of the car, 14% range increase, 370 fewer parts. I really think that long-term in any cars that do not take this architecture will not be competitive, Drew Baglino: (02:33:22) And it's not just at the product level, a better product. But in the factory, it's a massive simplification. You saw the part removal, it's casting machines, it's the structural battery pack. So we're looking at over 50% reduction in investment per gigawatt hour, 35% reduction in floor space. And we'll continue to improve that as we make the vehicle factory of the future. Elon Musk: (02:33:45) Yeah. So major improvements on all fronts from the cell all the way to the vehicle. Drew Baglino: (02:33:52) And in addition to the improvements we just said on enabling additional range and improving the structural performance of the vehicle, it is worth another 7% dollar per kilowatt hour reduction at the battery pack level, bring our total reductions now to 56% dollars per kilowatt. Drew Baglino: (02:34:17) All right. So stacking it up. We're not just talking about cost or range. We've got to look at all the facets. So range increase, we're unlocking up to 54% increase in range for our vehicles and energy density for our energy products. 56% reduction in dollars per kilowatt hour at the battery pack level, and a 69% reduction in investment per gigawatt hour, which is the true enabler when we talk back about how do we achieve this scale problem here. Elon Musk: (02:34:47) Yeah. So I think it's pretty nice that investment per gigawatt hour reduction is 69%. I mean, who would have thought? Drew Baglino: (02:34:57) Yeah, just happened to come out that way. Elon Musk: (02:35:03) I mean, 0.420%, of course. So what this enables us to do is achieve a new trajectory in the reduction of cell cost. And now to be clear, it will take us probably a year to 18 months to start realizing these advantages and to fully realize the advantages probably it's about three years or thereabouts. So if we could do this instantly we would, but it just really bodes well for the future and means that the long-term scaling of Tesla and the sustainable energy products that we make will be massively increased. So, what tends to happen as companies get bigger is things tend to slow down, actually they're going to speed up. Drew Baglino: (02:36:00) And they have to speed up if we're going to accelerate the transition to sustainable energy. Elon Musk: (02:36:04) Yeah. Long-term we want to try to replace at least 1% of the total vehicle fleet on Earth, which is about 2 billion vehicles. So long-term, we want to try and make about 20 million vehicles a year. Drew Baglino: (02:36:25) But I think it's important to point out that when we talked about three terawatt hours by 2030, the problem is a 20 terawatt hour problem. So everybody needs to be accelerating their efforts to accomplish these objectives. It doesn't matter where you are in the value chain. There is a ton to do, you need to rethink from first principles how you do it, so that you can scale to meet all of our objectives. Elon Musk: (02:36:47) Yep. Drew Baglino: (02:36:49) And, Elon. Elon Musk: (02:36:50) Sure. Drew Baglino: (02:36:53) What does this mean... Elon Musk: (02:36:55) What does this mean for our future products? So we're confident that long-term we can design and manufacturer a compelling $25,000 electric vehicle. This has always been our dream from the beginning of the company. I even wrote a blog piece about it because our first car was an expensive sports car, then it was a slightly less expensive sedan, and then finally sort of a, I don't know, mass market premium, the Model 3 and Model Y. But it was always our goal to try to make an affordable electric car. And I think probably, like I said, about three years from now, we're confident we can make a very compelling $25,000 electric vehicle that's also fully autonomous. Drew Baglino: (02:37:48) And when you think about the $25,000 price point, you have to consider how much less expensive it is to own an electric vehicle. So actually it becomes even more affordable at that $25,000 price point. Elon Musk: (02:38:02) Yeah. So we have and extreme performance and range. And we should probably talk about, more or less, Plaid. What about that? So, yeah. Anyway, we took the latest Plaid out to Laguna Seco on Sunday, it got a minute 30, and we think probably there's another three seconds or more to take off that time. So we're confident the Model Plaid will achieve the best track time of any production vehicle ever, of any kind, two-door or otherwise. Elon Musk: (02:39:15) And you can order it now. And it's available basically in the next year. And now we'll move to Q&A. Drew Baglino: (02:39:26) Absolutely. Elon Musk: (02:39:27) So we'll invite a few people on stage. Drew Baglino: (02:39:31) Come on up team. Elon Musk: (02:39:32) This is just a small portion of the team, but I thought it'd be great to show you some more of the team and when we do Q&A we can give various people different questions to answer. Drew Baglino: (02:39:49) Sounds great. Actually, I don't know how we're getting the questions. Elon Musk: (02:39:54) Actually, I don't know either. You can maybe get out of the car for two seconds and yell it at us. How are we getting the questions? Speaker 2: (02:40:07) [inaudible 02:40:08]. Drew Baglino: (02:40:09) Oh, there are mics. Wait for the mic. Elon Musk: (02:40:11) Oh, there are mics. Okay, great, great. Drew Baglino: (02:40:14) All right. Elon Musk: (02:40:16) Okay. We'll definitely needs to give people mics cause otherwise there's no way. Sorry? All right. We're going to pass some mics out. Oh, we don't have a name for the $25,000 car yet. Drew Baglino: (02:40:45) That's a great question, though. Speaker 3: (02:40:45) Elon, you talked about in Berlin that you were going to [inaudible 02:40:45] manufacturing [inaudible 02:40:44]. Elon Musk: (02:40:45) Yes, we will be manufacturing cells in Berlin. Yep. Drew Baglino: (02:40:52) Thermal management system? Speaker 4: (02:40:53) [inaudible 02:40:56]. Drew Baglino: (02:40:55) For homes. Elon Musk: (02:40:56) Oh, you mean like the home HVA? Yeah. That's a pet project that I'd love to get going on. I don't know, maybe we'll start working on that next year. Because I just think, man, you could really make a way better home HVAC system that's really quiet and super efficient, super energy efficient, and also has a way better filter for particles. And it works very reliably, and we've already developed that for the car. So the heat pump in the Model Y is really pretty spectacular. It's tiny, it's efficient, it has to last for 15 years, it's got to work in all kinds of conditions from the coldest winter to the hottest summer. So we've actually already done a massive amount of the work necessary for a really kick-ass home HVAC. Elon Musk: (02:41:53) And they could also stack them. So if you want to say, depending upon the size of your house or whatever, how much you need, you can just basically stack them and just have a very compelling, super efficient home HVAC. And then you could also communicate with the car and it'll know when you're coming home. So it's like, \"Oh, I don't need to keep the house cold all day, I'll just cool it down because I knew you were coming home.\" So the pack can communicate with the car and just really dial it into when you actually need cooling and heating. It'll be great. Drew Baglino: (02:42:25) Fun product. Who's next? Eli: (02:42:30) Hello? Hey guys, Eli here from Tesla Owners Club, my Tesla adventure. Just quick question, so I'm a huge fan of car camping in my Tesla with my dream case, my all time favorite activity, is it going to be possible to get climate control to the back of the cyber truck? Because that would be the ultimate camping machine if we can get all night climate control. Elon Musk: (02:42:51) We'll try to do that. Yeah. I agree, that would be really cool. Yeah. Drew Baglino: (02:42:59) All right. Who's next? Speaker 5: (02:43:00) Hello, longtime fan, Elon, great guy. Just a question, how does the ICE industry look like in the future? Elon Musk: (02:43:12) Well, I don't think there will be at ICE industry longterm. Well, I guess there might be like a few things that it's a like curious thing. There's still like some steam engines made somewhere, but they're just basically sort of quirky collector's items. I mean, that will be the future of the internal combustion engine car. Ryan McCaffrey: (02:43:36) Hi, Elon, to your left here in the white Model y. Ryan McCaffrey from the Ride the Lightning Tesla podcast. Curious about cyber truck, it was interesting to see where you had it in on the battery technology front. I'm sort of curious what you see for it in the production front. Trucks are so popular in America, do you see its volume equaling the 3 or the Y in the future? And also, were Tesla's able to legally be sold in Texas as part of the Giga Texas deal? Elon Musk: (02:44:09) Well, it's hard to say what the volume exactly would be for the cyber truck. The orders are gigantic. We have like, I don't know, well over half a million orders, I think maybe six or 600,000. It's a lot, basically, we stopped counting. So I think there's probably room for, I don't know, at least like a unit volume of like 250 to 300,000 a year, maybe more. Now, we are designing the cyber truck to meet the American spec. Because if you try to design a car to meet the super set of all global requirements you can't make the cyber truck, it's impossible. So it really is designed for the American market, but this is the biggest market. Our North American market is the biggest market for pickup trucks by far or large pickup trucks. Elon Musk: (02:44:59) And then I think we'll probably make an international version of cyber truck that'll be kind of smaller, kind of like a tight Wolverine package. It'll still be cooler, but it'll be smaller because you just can't make a giant truck like that for most markets. So, yeah, but it's going to be great. Elon Musk: (02:45:17) And I don't know. I think probably we'll be able to sell directly in Texas. We do pretty well right now, but it is a bit weird not being able to actually conclude a transaction in Texas, but it's got to be like a click on a server based in California. But weirdly we can do leasing in Texas, but not selling. Hopefully that'll get cleared up in the future. Speaker 6: (02:45:42) Elon, great job with everything that you're doing. It's Ross Gerber from Gerber Kawasaki. Your team's amazing. What I'm most curious about, these innovations are incredible but on my drive up here fully on autopilot for 400 miles, the entire state is brown and this is ultimately about climate. Has there been some analysis done if all these things are achieved, what will its direct impact be on climate? Elon Musk: (02:46:10) I think it will have a very significant impact because it will stop the CO2 PPM from growing as it is every year. I should say, I try to view the whole climate thing as a science question as much as possible. Science, you always question your hypothesis, is it true? Is not true? Or assign a probability to a given hypothesis. And I should say that my original interest in electric vehicles predates the climate issue. When I was in high school, I thought, \"Man, if we don't figure out electric cars, the whole economy's going to collapse when we run out of oil.\" So we better figure out electric cars and sustainable energy or civilization's going to crumble. Elon Musk: (02:46:57) And then it was only later that the significance of the climate risk became apparent. And we were also able, using tracking and other types of technology to access a lot more fossil fuels than previously thought, which is helpful for lowering the cost of gasoline, but it's pretty bad for the total tonnage of CO2 that you could put in the atmosphere. It's now greatly beyond what people previously thought. As we were just going through this presentation, it is a absolutely monumental task to accelerate the advent of sustainable energy. The entire global economy is still more than 99% dependent on, or call it roughly 99% dependent on, fossil fuels. So although electric cars get a lot of press right now, as a percentage of the total global fleet it's practically nothing. I would say yes, less than 1% of the global fleet is electric right now. Because of two billion cars and trucks and whatnot in use. So there's a massive amount of work ahead. Just insane, like hard to comprehend how much work is ahead to get the new vehicle production to be sustainable, to massively increase the amount of stationary storage, which is critical because renewable energy is intermittent, wind, and solar is intermittent, sometimes the wind doesn't blow and this obviously sun doesn't shine at night, so you got to have batteries, a massive, massive number of batteries. Drew Baglino: (02:48:44) Yeah, it's hard to measure in direct impact, but it's an experiment that we shouldn't be performing. And the sooner we can end the experiment the sooner we can kind of move on in a fully sustainable way that is actually lower cost. I think the thing that people haven't fully internalized is once we do get to the 25K car, the ownership cost of that car is incredibly lower than the prior car. And then on the solar side and wind, with the cost of solar wind coming down and with batteries coming down with them, the actual cost of energy on the grid is going down. So we're sort of moving towards a sustainable lower cost future. So there's not like a sacrifice. Elon Musk: (02:49:21) That's true. It is a false dichotomy to say that it's either prosperity or sustainability. This is often used by oil and gas to say like, \"Oh, well, do you want people to lose their jobs? Do you want to lower people's standards of living? Do you want to make all these economic sacrifices really in order to have sustainability?\" And the reality, as Drew was saying, is that sustainable energy is going to be lower cost, not higher cost than fossil fuels. Speaker 7: (02:49:52) Elon, quick question for you, right here in front. First, thanks for having everyone. I was telling a friend, the one company to go work for that's going to have the biggest structural... Speaker 8: (02:50:03) And the one company to go work for, that's going to have the biggest structural impact over the next 10 years at scale, it's probably Tesla. So kudos to everyone at Tesla for what they've done to this point and going forward. The two questions for you, as you've looked at the auto in the storage markets, I know you've talked about it at kind of 50/50 longterm, but it seems like a lot of the battery cost curve achievements that you presented today, really make some of these storage opportunities much more feasible over the next five years. And so I guess the first part of the question is, does your calculus upon learning and improving these things, change on that 50/50 mix, or is there a role where storage becomes bigger? And then the second part of the question, with all these huge grand visions, who's going to be with Tesla from a corporate perspective, accomplishing these things? Obviously, Tesla can't do it alone, but when you look at some of the traditional auto industry or power, et cetera, I don't see a lot of other Tesla's. Elon Musk: (02:50:59) Well actually, there's a lot of companies in China that I think are doing great work with electric vehicles and also with stationary storage, although we don't see that much in the US yet, but I think probably we will in the future. I don't know, obviously we're doing everything we can to encourage other companies to move to sustainable transport and also make stationary storage batteries. We made our patents freely available, we really try to tell these companies, \"Hey, you really need to do this, or you won't exist in the future,\" but they don't believe it. So we've talked until we're blue in the face. What are we supposed to do? But we really are hopeful that other companies will also do what we're doing and that will make a sustainable future come sooner. Drew Baglino: (02:51:53) From a fundamental market size perspective, we did the first ground up work to show the size of the market in terawatt hours and they are roughly 50/50. 10 terawatt hours for transportation, 10 terawatt hours for the grid. And part of that is because the grid batteries, because when you're making a power plant, you're making a large investment, our 25-year assets are greater. If the grid batteries were 10-year kind of things, the grid market would be bigger, but because it's a longer duration asset, they're roughly the same size. Speaker 9: (02:52:31) Thinking long-term, is there any other segments that this new battery will be able to disrupt or electrify, beyond just the initial Model 2 or cheaper sedan? Like a Boring Company loop, plane- Elon Musk: (02:52:44) Where are you? Are you there? Speaker 9: (02:52:45) What's up? Right here. Elon Musk: (02:52:46) Okay, great. It's like ventriloquism here, we just get the sound out of the speaker and can't tell where the heck it's coming from. Speaker 9: (02:52:55) Yeah. Any hints or is the model too such a big deal because it decreases the cost of transportation, that that is really the disruption, or should we get hyped that this new cost curve opens up different vehicle categories, like a high passenger density bus, Boring loop, boat, plane? Elon Musk: (02:53:12) Well, I mean, there are batteries in limited production right now, that do exceed 400 watt hours per kilogram, which I think is about the number you need for a decent range, medium range aircraft. And I think our batteries will, over time, start to approach the 400 watt hours per kilogram range as well. So yeah, I mean, I think over time, we'll see all modes of transport, with the ironic exception of rockets, transition to sustainability or to electric basically. On the rocket front, what we're planning to do is, about 80% of Starship is liquid oxygen and we're actually already running a power line to be able to use wind power to create the liquid oxygen. So we're making some decent progress on sustainability on the rocket front, but there's just no way to have an electric rocket. And it's important for the future of life and consciousness, that we become a multi planet species, so got to keep doing that. Josh Phillips: (02:54:21) Hi Elon, Josh Phillips here, retail investor. I have a question in regards to the lithium and nickel industries and the likely price spikes and shortages of high grade materials the EV industry is likely to see if they don't act fast to address future supply. Tesla have clearly made the right moves that are necessary, but there's a real worry that the potential supply issues and price spikes will create a drag on the rest of the EV industry and therefore a drag on global EV adoption. What advice would you give to the EV and mining industries to quickly solve this looming hurdle? Because for a sustainable energy future, the spice must flow. Thank you. Elon Musk: (02:55:07) Yeah, indeed. The spice must flow. The new spice. I don't know. I'm not sure. I guess we can try to basically overdo it in cell production and perhaps supply cells to others, but we do see the fundamental constraint, as total cell production. That's why we're putting so much effort into making cells and kind of trying to reinvent every aspect of cell production, from mining the ore, to a complete battery pack, because it's the fundamental constraint. We're not getting into the cell business just for the hell of it, it's because it's the fundamental constraint, it's the thing that is the limiting factor for rapid growth. But we could certainly try to overdo it on cell production and perhaps sell cells to others, although we are going at absolute top speed, so it's not like we're holding it back. Elon Musk: (02:56:15) I think just making really efficient cars that have lower drag coefficient, low rolling resistance, efficient powertrains, I mean, that's kind of what we've done in order to make iron phosphate still have a good range. So the iron phosphate's a lower energy density solution, but while there are some limitations on the total amount of nickel produced every year, there's really no limit on the iron. There's so much iron it's ridiculous. So you can really scale up iron phosphate at a raw materials basis, more than you can nickel. Drew Baglino: (02:57:00) And just to point out, when we were walking through this presentation, we intentionally separated all the different aspects. The benefits of structural battery, apply to an iron based cathode in the same way they apply to a nickel based cathode. So you get longer range, iron base vehicles. And also the silicon benefit can apply to the iron based vehicles as well. So we can do a lot to extend the range of an iron based vehicle, which is why it's a key part of the roadmap going forward. And then I invited Turner up here to talk about what the mining industry can do. Turner: (02:57:31) Yeah. Diversification on the cathode side, is obviously massive and EVs are all about efficiency. And so for the EV industry, for the vehicle industry, we need to see powertrain efficiency really increase, all other companies, matching Tesla powertrain efficiency, so that everyone can have that diversified cathode approach, where LFP is used in medium range, and even really make a 300 mile vehicle with LFP. And really the goal that we were trying to present here, was a model for vertical integration, strategic vertical integration, that a lot of different people can do. What we need to see is vertical integration that shortens the process path, from mine to cathode. And what we're doing here is novel and we're trying to push the industry in that direction. So we're presenting a model here that anyone can can follow. Elon Musk: (02:58:27) Yeah. In fact, if there's anything that you guys want to comment on, feel free to step forward and say something. Speaker 10: (02:58:34) I think the key is to be smart about your chemistry choices, your materials choices. Elon Musk: (02:58:38) Talk louder. Speaker 10: (02:58:38) Yeah. If you're smart about your materials choices, the spice will continue to flow. You don't need to use the same kind everywhere. It's about strategically planning it out and for miners, I think we are incentivizing them quite a bit, to ramp up their production. Drew Baglino: (02:58:57) Yeah. And actually we had good calls, they're all motivated. I think, they've been sort of sitting back being like, \"Are you going to grow like crazy?\" And we're like, \"Yeah, we're going to grow like crazy.\" And then I think this indicates we're going to grow like crazy and that's what the miners want to hear and then they'll go make the investments. Ben Limpic: (02:59:13) Hello, Elon. This is Ben Limpic, I'm a musician. I was wondering, does Tesla have any future plans to make partnerships with music companies, like it has done with Tencent games or things like that, for you guys to actually kind of expand your services for artists and other types of creative people, to get involved in producing content that can be part of the Tesla ecosystem or so other people that do creative things can get involved with you guys? Elon Musk: (02:59:44) We haven't really thought about it that much, but I suppose it's probably something we should think about. We will be providing a title on the Tesla's. So we're providing more music sources that people can choose from and just generally trying to improve the entertainment experience in the cars. And I think actually as we go to a more autonomous future, the importance of entertainment and productivity will become greater and greater. I mean, to the degree that if you're just basically sitting in your car, the car is fully autonomous and driving somewhere, the car is essentially your chauffeur and then the things that become important are, okay, well let's have good entertainment and if you want to do some productivity stuff, then that actually starts to become much more important because you're no longer spending your attention driving the car. So it will be extremely important in the future. Drew Baglino: (03:00:42) Should we do some of the say.com questions? Speaker 11: (03:00:46) Yeah. Drew Baglino: (03:00:47) Okay. Should we do the second one? Elon Musk: (03:00:54) Yeah. The first one, I think we already answered. If we're able to make enough cells, which we'll try to do, we will supply other companies. It's definitely not an intentional effort to keep the cells to ourselves, if we can make enough for other companies, we will supply them. And we were trying to do the right thing for advancing the sustainable energy, whatever that is. Elon Musk: (03:01:19) Vehicle to grid, we get asked that a lot. I think one of the things that's important to note, is vehicle to grid, unless you have a power cutoff, you need to cut off your main supply to the grid, otherwise, if you lose the power in your house, you'll basically just backflow energy to the grid. So just having a reversal in the power flow, does not actually keep the lights on, you need a whole separate system to cut off power to the grid. And I think there's also the case that people really want the freedom to be able to drive and to charge at their house. And it's obviously very problematic if you get to morning and your car, instead of being charged, it discharged into the house and then you're sort of, \"Okay, now I can either drive or use the battery to power my house.\" Elon Musk: (03:02:19) I think it's actually going to be better for people's freedom of action, to have a power wall and a car separate, and then everything works that. You basically combine that with solar, either solar retrofit or solar glass roof, and local battery storage, so you basically become your own utility and then the car can be charged also with solar. I think that's the stuff that works, that said, we can certainly do vehicle to grid, I think we can basically enable that with software in Europe or something, right? Drew Baglino: (03:03:00) Yeah. Future generations of power electronics, we will be able to do this more or less everywhere, from a energy market participation perspective, but from a backing up the house and it just so happens that the way the North American connectors are, on all the cars in North America, it doesn't matter whether it's the Tesla connector or the connector that the other vehicles have, doesn't actually support powering your home. It's unfortunate, so you'd need an additional hardware to do that. But yeah, in the future, all versions of our vehicles will be able to at least do bi-directional power flow for the purposes of energy market participation. But even for that, it's important to remember that your car is in plugged in 24/7, so it's kind of an unpredictable resource for the grid. It'll have a value, but it's not the same as a stationary battery pack. Elon Musk: (03:03:49) Yeah. Honestly, a vehicle to grid sounds good, but I think actually has a much lower utility than people think. I think very few people would actually use vehicle to grid. With the original roadster, we had vehicle to grid capabilities, nobody used it. Drew Baglino: (03:04:15) How do we find the engineers to do everything we're saying? Elon Musk: (03:04:18) How do we find the engineers to do all these things? Well, I guess we recruit a lot of engineers from all parts of the world. I think Tesla has a good reputation for doing exciting engineering and that tends to attract a lot of the top engineers in the world because they know that their efforts at Tesla will really serve the greater good and we're super hardcore about engineering. Tesla is first and foremost an engineering company, it's like hardcore engineering is what we do. The sheer amount of hardcore engineering done at Tesla is insane. And if you look at say, there's various surveys done of engineering schools, where do you want to go, what's your top choices? And actually the top two choices last few years, have been Tesla and SpaceX. So sometimes it's Tesla first and sometimes SpaceX first, but those are the two top ones. Drew Baglino: (03:05:18) Yeah. I mean, if you're motivated to solve some of these problems, which are the hardest problems in the world to solve, that really fundamentally enable the future we all need, please reach out and help us work on these problems. Elon Musk: (03:05:30) Absolutely. And like you said, the battle is far from over. Less than 1% of the global automotive fleet has been converted to electric and even maybe less than 0.1% of stationary storage has been done. So stationary storage has barely begun, converting the global vehicle fleet to electric, has barely begun. So there's still a massive amount of engineering work to be done at Tesla and other companies, to accelerate this transition to sustainability. Jordan: (03:06:06) Hey, can you guys hear me? Drew Baglino: (03:06:07) Yeah. Jordan: (03:06:08) This is Jordan from Mark Asset Management. So you've talked about the importance of the factory and you've mentioned the ground up design process and a lot of the new things that you're going to be doing or started to do in Shanghai, Berlin, and Austin. Can you just maybe help us understand and quantify, how financially meaningful all of those improvements will be, and then given what you're trying to accomplish as a company, is it fair to assume that the vast majority of improvement will be given back to the customer in the form of lower prices? Elon Musk: (03:06:39) Yeah. I mean, I think certainly we will try to give back as much as possible to the customers. It's not like Tesla's profitability is crazy high, our average profitability for last four quarters, is maybe 1%. So just to be clear, it's not like we're minting money. Our evaluation makes it seem like we are, but we're not. So we do want to try to make the price as competitive as we can, without losing money. If you keep losing money, you'll just die. So this thing called profit is just like, we need to bring in more money than we spend, otherwise we're dead. Drew Baglino: (03:07:19) But affordability is key to how we scale, right? The demand goes non-linear as you reduce the price of the car. Elon Musk: (03:07:25) Yeah. I mean, it's important to sort of separate the difference between affordability and value for money or desirability of the product. So for a lot of people, they want to buy a Tesla, they simply don't have enough money. We could make the car infinitely desirable, but if somebody does not have enough money, they can't buy it. Sometimes people kind of forget this. People have to have enough money to buy the car and just making a car super desirable, but expensive, does not mean they can afford it. So it's absolutely critical that we make cars that people can actually afford. Go through some of these things, scroll down or something. Drew Baglino: (03:08:19) When do you expect Tesla vehicles to beat ICE vehicles on initial purchase price? I think a way to answer that question, is in the classes of vehicles we sell today, we're already doing that. Elon Musk: (03:08:30) Yeah. We're already pretty close. And then factoring in total cost of ownership and the fact that electric vehicles require much less servicing and are way cheaper to run, when you look at total cost of ownership. And you can always lease a car, so if you just lease a car or get a loan for a car, you've got your sort of monthly payments and then your cost for either gasoline or electricity and your cost of servicing and the fully considered cost of electric car is much less than a gasoline car of the same nominal purchase price. I mean, that said, maybe on the order of three years, when we can do lower cost, like a $25,000 car, I think that will be basically on par, maybe slightly better than a comparable gasoline car. So I think maybe it's on the order of three years-ish. Drew Baglino: (03:09:37) How have the technology advancements and increased vertical integration of battery manufacturing, influenced your ability to improve the environmental and social impact of the supply chain? And I think ... Yeah. Elon Musk: (03:09:48) We sort of have said that already. Drew Baglino: (03:09:49) Yeah. Elon Musk: (03:09:50) Do we have some ability to scroll through this? Just scroll away. Drew Baglino: (03:09:57) We covered recycling. Elon Musk: (03:09:59) Yeah. Just scroll until we've got stuff that we haven't covered. Drew Baglino: (03:10:02) We definitely covered that top one. Elon Musk: (03:10:09) Yeah, a lot of the things we've already asked really. Drew Baglino: (03:10:16) Covered that. That one. Elon Musk: (03:10:26) We literally just answered that. Drew Baglino: (03:10:27) Yeah. Oh, I saw a cathode durability question. Let's go to that one, go down, go down, go down. Good technical question. Keep going. How are you going to address the cathode durability and cost and environmental impact trifecta? Is this something you're going to leave the environment upstream and supply chain to solve? No, I think we tried to answer that directly. I mean, we really are looking at not just what happens in the cathode facility, but currently outside the cathode facility that should really be inside and removing processes that shouldn't have been there in the first place and the use of reagents that are just costly and not necessary and removing a bunch of wastewater from the process. Elon Musk: (03:11:09) Guys, is there anything you want to add to ... Maybe we can go through everyone and maybe say what you're doing and say a few words. I don't know. Speaker 10: (03:11:21) Sure. I just want to reiterate the fact that this is a massive problem. Elon Musk: (03:11:25) Massive problem. Speaker 10: (03:11:26) And it seems like Tesla's on its way and ahead, but we need everybody's help because it's everybody's planet and we're not going to get to 20 terawatt hours by ourselves. So please think about this carefully, as it affects everybody, so let's get on it. Elon Musk: (03:11:45) Yeah. And obviously, if you care about solving sustainability and doing hardcore engineering, definitely come work for Tesla. Speaker 12: (03:11:53) Yeah. We went through a couple of the manufacturing improvements and it kind of looks easy when you put together a nice slide deck, but it's super challenging. When you take materials out of the process, when you integrate processes together, you have to do a lot of things at once and that's like this immense engineering challenge. And so to appreciate that, to get through this, we need the best engineers we've got. And we've got this awesome team, I just want to shout out also to all of our team watching, you guys are awesome, you absolutely kicked ass putting this together. Drew Baglino: (03:12:36) Thank you. Thank you, Tesla team. Totally agree. Speaker 12: (03:12:44) Yeah. That's it. Rodney Westmoreland: (03:12:47) Yeah. Rodney Westmoreland, managing the construction here at Tesla. What I would like to say is, one, shout out to the team. The team has been working effortlessly, a very, very tough project here, for 24 hours a day it seems like, around the clock, to have this complete. The thing that sets us apart from a lot of other construction, we have a construction company here, the thing that sets us apart is that we're integrated in the manufacturing process. So every detail that comes from Drew's mouth, is directly implicated into the system that we're building. That way, what would typically take three or four months to create a specification, our design team is working right with the manufacturing team, to allow us to speed that process up tremendously. Drew Baglino: (03:13:36) Yeah, it's definitely a important part of the vertically integrated approach, is to be able to design the factory around the equipment, in fact, together with the equipment, so you can build the factory at lower costs and more quickly. Scott: (03:13:50) I'm Scott, I focus on cell design. I think it's hard to put into words how inspiring this is, been at it such a long time with Tesla. And I really hope others do join us- Elon Musk: (03:14:01) Since when Scott? Scott: (03:14:02) Since 2005, with many of you. Thank you. Year before Drew, who's keeping track? But I'm really stoked what the team's been able to accomplish over the last short period of time, about a year, it's been really an incredible transformation. I mean, hopefully what we've shown you, inspires you to join us or join somebody else in the effort. And I couldn't think of a greater, more intelligent, more hardworking team to be working on for this problem. Peter: (03:14:37) I Peter, I lead the manufacturing improvement team. And I guess the point that I'd like to make, is manufacturing improvements is like the accelerator. So you think about the execution that Rodney talked about, in terms of how fast we've been able to put together this factory, which is amazing and something that's been really incredible to be a part of. That's not enough, what we need to do is improve the manufacturing technology, that's the real accelerator and that's what we're really focused on. Elon talks about it all the time, that really going and improving that system is what will enable us to get to the scale and the cost that we need. Peter: (03:15:15) And then the other point that I would make is on the recruiting side, it doesn't matter if you know about batteries, if you come from any industry, you can do something fantastic in the work that we're doing. We talk to people from industries that you wouldn't imagine. Like I talked to a guy who makes golf balls and he has stuff which is really impactful for what we're doing. So if you're in any industry and you want to be impactful here, come join us, it'd be great. Tony: (03:15:45) Hi, excuse me. Hi, I'm Tony. I've been working in lithium and cathode materials for almost 23 years now and this is the most growth I've seen in a company, I've been here a little over a year and a half. We are hiring amazing people that are allowing us to leverage technology that most of the industry is struggling to achieve. So to answer the question, how are we going to do this? We are really advancing the materials manufacturing for cathodes and for lithium, beyond what has been accomplished in the previous 20 years. Drew Baglino: (03:16:26) It's exciting. Turner: (03:16:31) Yeah. My name is Turner, work closely with the team, have worked a lot with everyone here. On the cathode and upstream materials side, it's really important that everyone understand that this growth is coming. This growth is real, we are going to make all of these batteries and everyone needs to grow with us, the entire supply chain needs to grow with us. And if you have an idea that simplifies anything in the supply chain, come talk to us, come work with us and let's do it. Drew Baglino: (03:17:02) Any existing specification is wrong, any existing manufacturing method is wrong, process equipment, it's wrong, it's just a question of how wrong. Quote Elon Musk. Elon Musk: (03:17:12) Exactly. We're wrong, just the question of how wrong. Trying to be less wrong. Drew Baglino: (03:17:16) So tell us how we're wrong and how we can do it better, so that we can accelerate and improve as fast as possible. Elon Musk: (03:17:23) All right. Well, I guess thank you everyone for coming. I hope you liked the presentation. Very exciting future ahead. We're going to work our damnedest to transition the world to sustainable energy as quickly as possible, and your support and help is key to that success. So thanks again, super appreciated and look forward to the next event. Thank you. Drew Baglino: (03:17:45) Thank you."},"languages":["en"],"lang":"en","transcriptSource":"https://www.rev.com/transcripts/tesla-2020-battery-day-transcript-september-22"},{"id":"spacex-starship-update-2019","type":"video","url":"https://www.youtube.com/watch?v=3N7L8Xhkzqo","title":"SpaceX Starship Update","titles":{"en":"SpaceX Starship Update","de":"SpaceX Starship Update","fr":"SpaceX Starship Update"},"date":"2019-09-28","summary":"In front of the assembled Starship Mk1 at Boca Chica, Elon Musk reviews SpaceX's history and presents the stainless-steel, fully reusable Starship.","text":"Elon Musk: (00:21) This is, I think, the most inspiring thing that I’ve ever seen. And I just like to thank the SpaceX team and the suppliers, and the people of Boca Chica and Brownsville. Thank you for your support. And, just like, wow, what an incredible job by such a great team to build this incredible vehicle. First of all I want to start with that. I mean, I’m just so proud to work with such a great team. And it’s really windy here, by the way. If you’re watching this online, it is really windy.\n\nSo, the point of this presentation and this event is really… there are two elements to it. One is to inspire the public, get people excited about our future in space, and get people fired up about the future. There are so many things to worry about, so many things to be concerned about. There are many troubles in the world, of course, and these are important, and we need to solve them. But we also need things that make us excited to be alive, that make us glad to wake up in the morning and be fired up about the future and think, yeah, the future is going to be great.\n\nThis space exploration is one of those things; and becoming a spacefaring civilization being out there among the stars. This is one of the things that I know it makes me glad to be alive; I think it makes many people glad to be alive. It’s one of the best things. We are faced with a choice: Which future do you want? Do you want the future where we become a spacefaring civilization and are in many worlds and are out there among the stars or one where we are forever confined to Earth? And I say it is the first and I hope you agree with me. (02:30)\n\nThe critical breakthrough that’s needed for us to become a spacefaring civilization is to make space travel like air travel. So, with air travel, when you fly a plane, you fly that plane many times. I mean, the risk of stating the obvious, it really almost any motor transport, whether it’s a plane or a car, a horse, a bicycle is reusable. You use that motor transport many times. If you had to get a new plane every time you flew somewhere and even get to have two planes for the return journey, very few people could afford to fly. Or if you could use a car only once, very few people could afford to drive a car. So, the critical breakthrough that’s necessary is a rapidly reusable orbital rocket. This is basically the Holy Grail of space. The fundamental thing that’s required.\n\nAnd it is a very hard thing to do. It’s only barely possible with the physics of Earth. I mean, if Earth’s gravity was a little heavier, it would be impossible. And if Earth’s gravity was a little lighter, it would be quite easy. So, we’re really right on the cusp of what is physically possible. So, in order to create a rapidly reusable rocket and fully reusable orbital rocket, you have to have engines that have incredibly high, specific impulse (Isp), that essentially are extremely efficient.\n\nYou need to have a structure that is also incredibly mass efficient. And then that all needs to come back to the launch pad and be able to be refilled with propellant and flown again very quickly, just like an aircraft. It’s just because of the physics of Earth being quite a deep gravity well and having quite a thick atmosphere; this is a very tough but not impossible thing. But it is the most fundamental thing. With SpaceX, we started out 17 years ago, and the first rocket we designed was the Falcon 1, which was that guy right there. (05:00)\n\nWhen we started off, we were very naive. And in fact, the reason I should say… it’s September 28th; this is the 11th anniversary of the first time SpaceX reached orbit. Eleven years ago today, SpaceX made orbit for the first time. It was actually our fourth launch. And if that launch had not succeeded, that would have been the end of SpaceX. I’d run out of money, there were no more investors, and that would have been it. So, if that fourth launch had not succeeded, that would have been curtains.\n\nBut fortunately, fate smiled on us that day, and we made it to orbit. I have great respect for anyone who makes it to orbit. That is a hard thing. We were very naive, obviously very naive on many levels back then, because we did actually try to recover the first stage. The first stage had a parachute on it, and we thought, okay, we’ll just pop the parachute when it comes back into the atmosphere. Then it’ll land somewhere in the ocean, and we’ll go fish it out of the ocean with a boat. This does not work. I actually remember getting mad at the parachute supplier, like, your parachute doesn’t work. Nah, it wasn’t their fault.\n\nWhen the rocket comes in from space, that first stage is coming in like, you know, Mach 10 to 12, and it hits the atmosphere like it’s a concrete wall and ‘boom’. So, you actually have to orient the rocket carefully. You have to have aerodynamic surfaces, and you have to do an entry burn to slow it down. Then you’ve got to guide it through the atmosphere and then do a propulsive landing. This took us many, many attempts.\n\nWe actually did a video, a blooper reel of all the times we failed, which was a lot. (07:30) I think it might have taken us like 14 attempts or something before we finally successfully landed the rocket. If we’ve gone to the next slide, you can take a look at… – This is Grasshopper, that’s actually Falcon 9. It’s hard to tell the scale, but that’s a Falcon 9 size booster with one engine and big legs with giant shock absorbers; we didn’t know what the heck we were doing.\n\nElon Musk (voice-over): Now amazingly, Grasshopper had zero blessures; Grasshopper is still alive.\n\nElon Musk: They have Falcon1; what you saw there was a Falcon 9 size vehicle. What’s really kind of hard to grasp at a visceral level is that this giant ship will do the same thing that Grasshopper did. This thing is going to take off, fly to 65,000 feet, about 20 kilometers, and come back and land in about one or two months. So that giant thing – it’s really going to be pretty epic to see that thing take off and come back. And then hopefully, yeah, …(applause) – Yeah, it’s wild.\n\nThis is a quite radical… I’ll talk about it later in the presentation; it’s, this is quite a new approach to controlling a rocket – much more akin to a skydiver than a plane. But I’ll talk about that later. (10:00) So, going from Falcon 1 to Falcon 9 to Falcon Heavy, which we launched… actually, the first launch of Falcon Heavy was only February of last year. So, it’s only been about a year and a half since the first Falcon Heavy launch when we did two side-by-side booster landings. And I always liked this video. It was done by my friend Jonah.\n\nYeah. I never thought that would happen, actually. (12:30) I’m glad that it did. Some people were like, wait, why do we have the Roadster with the astronaut, you know, Starman. Actually, this came from a discussion with my friend Jonah. I was at his kitchen, and I was like, you know, normally when they do a rocket launch, there’s a launch of a rock of concrete, but that doesn’t sound very inspiring. So, what do you think the most sort of fun thing is that we could launch? And he was like, well, why don’t you launch your Tesla? And I was like, that’s a great idea.\n\nAnd another friend of mine, she said: “Why don’t you put a tiny Tesla on the dashboard?” So we put a tiny Tesla on the dashboard with a tiny Starman in the tiny Tesla. This is just to confuse the aliens in the future. They’ll be like, what the heck is this? You know, just want something that captures the imagination, gets people excited about space.\n\nSo, let’s see Starship. You can really see it right there, obviously. There’s a picture, more a rendering. It’s about 50 meters, sort of 165 feet or so.\n\nActually, I notice we have an error in our ship dry mass here; my apologies. I wish it was 85 tons. This ship dry mass has to be approximately 120 tons. The initial Mach 1 prototype is closer to 200 tons, and in series production, I think it’ll probably be about 120 tons. If we get really lucky, it might get down to 110; 99 would be super epic. So, in terms of its usefulness, it’ll be able to do about 150 tons with full reusability to orbit and back.\n\nThis is a very big number for full reusability. The very initial versions, we’re confident will do over a 100 tons, but I think there’s a clear path to 150 tons. The cost of a fully reusable system is basically (15:00) the cost of the propellant, which is mostly oxygen. This is three and a half tons of oxygen for every one ton of fuel. So, one of the advantages of this architecture over the Falcon architecture is that we actually use more oxygen per unit of fuel rather than less. Merlin or the Falcon architecture is about two and a half tons of oxygen for every one ton of fuel. This is three and a half tons of oxygen for every one ton of fuel. So when it ascends, it’s really mostly liquid oxygen because when you get to vacuum, there’s no air, basically.\n\nSo, the next slide. Earlier I was talking about how Starship enters and how it’s controlled.\n\nIt’s quite different from anything else. It’s really falling, and so we’re doing a controlled fall. With a rocket you’re actually trying to break as opposed to… you’re trying to create drag instead of lift, it’s really the opposite of an aircraft. You want the most amount of drag that you can produce. And you want some lift, especially when you’re in the upper atmosphere, mostly so that you don’t…  you can control the maximum heating rate. You want enough lift to keep yourself high in the low-density portion of the atmosphere, so you can burn off velocity. Basically, it goes like, if this is the Earth, it goes at about a 60 degree.\n\nMy hand is the rocket – it’s going at about 60 degrees. So when in orbit, you’re actually going at around 25 times the speed of sound horizontal to the ground. This is a very important concept that is counterintuitive to our normal daily life. Being in orbit, being in zero G is not about altitude. It’s about velocity. How fast are you going – horizontally? When something’s in orbit, it’s zooming around the Earth so fast that the outward acceleration, outward radial acceleration, is equal to the inward acceleration of gravity. And then you have zero gravity. This is why you actually have zero gravity.\n\nPeople often think the Space Station is stationary, (17:30) but it’s actually going around the world at 25 times the speed of sound or about 17,000 miles an hour. It always looks stationary in the pictures. And since there’s no air, you don’t have to have an aerodynamic structure. So it can be a totally crazy structure that doesn’t look like it should be able to go 25 times the speed of sound, but it does. And you can only feel the acceleration. You can’t feel velocity. People sometimes wonder, what does it feel like to go 25 times the speed of sound? Actually, it feels like nothing. Only accelerating to there feels like something.\n\nSo, the Starship is coming in – this platform is the Earth – it’s coming in at hypersonic velocity like this, sort of around a 60-degree angle. So, it comes like this and then starts falling and then just falls like a skydiver, and it’s just controlling itself – and then it turns and lands like that. That’s an incredibly elaborate explanation. There you can get a sense for it. This is much better.\n\nThere you go. See, same thing. It’ll look totally nuts to see that thing land. Yeah, that’ll be crazy.\n\nSo let’s talk about the Raptor engine. The ship will have a total of six engines.\n\nThree of the sea-level variety of Raptor, and those are actually on the rocket right now. So, we have the three sea-level… in fact, that’s a picture of just inside – that’s what it looks like. So, we’ve got the three sea-level Raptor engines, and they gimbal, which means that the whole engine moves. So, the way the rocket steers is by moving the entire engine. Whereas an aircraft engine is static, and you move by moving the control surfaces, like the ailerons and rudder and elevator and flaps… – The rocket – when the engines are powered, you moved the entire engine to steer it. The Starship will have three sea-level engines that move up to about 15 degrees angle and three vacuum engines that are optimized for efficiency in orbit that will not move. They will be just fixed it in place.\n\nAnd that allows us to have the biggest bell nozzle (20:00) for the vacuum Raptor engines. Aspirationally, the target is a 380 second Isp for the vacuum engine. This is a very… – In sort of space geek terms, this is like really a great number. And even for the steel alloy engines to get over a 350 second Isp is also really great. So actually… – sorry, I’m looking at the slide here, and you’re not. So, that’s what I meant by ‘it looks like that on the inside’… – go back one slide. That’s the inside of the Starship right now.\n\nThat’s what it looks like in the base. All right. Then heat shield.\n\nI’ve gone through various iterations of heat shield. There’s a lot of ways to skin the cat here.\n\nUltimately, we decided to have heat shield hexagonal tiles, ceramic tiles, basically like a tiny glass vermicelli at a microstructure level. Very light, but very crack resistant – essentially glass tiles. And there, because Starship is a steel construction… – At first, it feels like, “Oh, it’s steel. Does that mean it’s heavy?” No, actually, it’s the lightest construction. I think the best design decision on this whole thing is 301 stainless steel. Because at cryogenic temperatures, a 301 stainless actually has about the same effective strength as an advanced composite or aluminum-lithium. Unlike most steals, which get brittle at low temperatures, 301 stainless gets much stronger.\n\nAnd if it’s in the extra hard condition, meaning it’s cold-rolled to extra hard condition, it also gets way stronger. Actually, its strength-to-weight ratio at cryogenic temperatures is equivalent, or even perhaps slightly better than advanced composites or aluminum-lithium. This is not well appreciated. (22:30) Because if you just look at the materials manual and say like, what is the strength of stainless steel? It looks much weaker than it is. If you say, what is the strength at cryogenic temperature? Oh, much stronger; at very low temperature, almost twice as strong. That’s when it becomes better than carbon fiber or aluminum-lithium.\n\nAnd this is another benefit: It also has a high melting temperature. So, for a reusable ship, you’re coming in like a meteor. You want something that does not melt at a low temperature. You want something that melts at a high temperature. And this is where steel is extremely good as well. Steel has a melting temperature around sort of 1500 degrees centigrade whereas aluminum, you know, maybe 300 or 400 degrees, and same thing for carbon fiber. And that’s really pushing it.\n\nHaving that much higher melting temperature means that you don’t need any shielding on the leeward side of the ship when it comes in for entry. And the shielding you need on the windward side – the hot side – is massively reduced because the thickness of the tile is actually for a reusable system – It’s dependent on what back shell temperature, like how hot does the back of the tile that interfaces with the airframe get. And because the steel can take a much higher temperature, your heat shield, even on the windward side, is much lighter. But the net effect is that a 301 stainless steel rocket is actually the lightest possible reusable architecture.\n\nThen, to come to cost. The carbon fiber we were using was $130 a ton. The steel is $2,500 a ton. Oh, sorry, $130,000 a ton versus $2,500 a ton. That makes much more sense. It’s $130,000 a ton for the carbon fiber and $2,500 a ton for the steel. So, the steel is about 2% of the cost of the carbon fiber. So, this is a good thing we changed from carbon fiber to steel, by far. (25:00) And it’s very easy to weld stainless steel, the evidence being that we welded it outdoors without a factory. Great skills by the team, but with carbon fiber, this is impossible; with aluminum-lithium, also impossible. But steel is easy to weld,  and it is resilient to the elements.\n\nAnd also, actually, (… 25:36), like on Mars, you can cut that up, you can weld it, you can modify it. No problem. Yeah. That’s a good point. You’re out there on the Moon or Mars; you want something that you can modify, that you can cut up and use for other things. That’s like for sure a great thing. So anyway, steel – obviously I’m in love with steel; I had to say it, you know. So, let’s see, going on to the booster.\n\nSo, the booster is designed to take up to 37 Raptor engines. I’m not sure if we’ll go that high, but you can really have 31. I think the minimum number you’d want is maybe around 24. But the booster is designed to be able to take multiple engines out, so you can actually add or subtract engines as you’d like. You basically just need a lot of force pushing up. Over time, I think you probably want around a 7,500 ton force rocket which is about twice the thrust of a Saturn V, a little more than twice the thrust, and on a roughly 5,000 ton gross liftoff mass for roughly one and a half thrust-to-weight.\n\nFor a reusable rocket, you actually want a high thrust-to-weight rather than with an expendable rocket, where you want a low thrust-to-weight, because any thrust-to-weight below one is not useful; like, if you have a less thrust than your weight, you don’t move. So, you actually want a high thrust-to-weight for a reusable rocket. This is a very important (27:30) design optimization change. So that’s why I think more engines are probably good and getting up to around 7,500 tons over time and a one and a half thrust-to-weight ratio, or more.\n\nWe think we’re probably going to adjust the grid fins designed to be kind of like a diamond shape. It looks cooler. It works better too. And then the rear fins are actually just legs. They’re not needed for stabilization or guidance. They’re essentially there for legs.\n\nAll right. So let’s go into some of the development testing. This is a Raptor firing.\n\nAll right. And then, obviously, we had a Raptor fire on the Starhopper. Yeah.\n\nIt’s kind of hard to see it to appreciate scale, but it’s the same diameter as the Starship. And obviously, it’s just right over there. So, it’s kind of hard to tell if it’s the size of a trashcan or, you know, how big it is, but it’s about… the body diameter is about 9 meters or 30 feet and not including the legs span. (30:00) Yeah. So, this gives you a sense of size.\n\nSo the little pixel there… little pixels are a human. And then there’s the Hopper next to it, the Millennium Falcon for comparison, then Starship, which is what you see before you. And then what it will look like with the full stack, which is almost two and a half times as tall as this vehicle. This simulation will give you a sense of the scale of things.\n\nElon Musk (voice-over): It slightly reminds me of a scene from Spaceballs.\n\nElon Musk (voice-over): This is the orbital refilling. Orbital refilling is extremely important for getting to Mars and getting to the Moon to establish a city on Moon or Mars. This is a vital step.\n\nA rapidly reusable orbital launcher rocket is… a rapidly reusable rocket is required for – alliteration – for getting a breakthrough in cost of access to space; that you don’t throw the rockets away every flight. But another key step is refilling on orbit so that Starship can get to orbit with, let’s say, 150 tons of payload for the Moon or Mars or beyond. And then it can get tanked up to fill up its propellant tanks and so that it can depart from low Earth orbit (35:00) with 1,200 tons of propellant. This is a very big thing so that your delta velocity is enough to transport about 150 tons to the surface of the Moon or Mars with full reusability and orbital refilling.\n\nThe orbital refilling is actually a simplified version of what SpaceX does in docking with the Space Station. So it’s actually harder to dock with the Space Station than it is to do orbital refilling, but in practicing docking with the Space Station, SpaceX has also learned how to rendezvous and dock in orbit in a complex environment. So this is one of the other critical pieces of the puzzle needed to establish a base on the Moon and Mars, a city, ultimately. And yeah, so those are the critical ingredients. So, we think it will be very exciting to have a base on the Moon.\n\nEven if it’s just a science base that… – for example, we have a base at Antarctica. Many countries have bases in Antarctica for science research, and this would be an incredible area for research. So whether or not people want to live on the Moon, there’s definitely a lot of science to be done. And I think it’s close as well. So, that would be quite exciting to do. And then, of course, we can go to other places in the solar system like Saturn. But the critical thing that we need to focus on, I think, is the fastest path to a self-sustaining city on Mars. This is the fundamental thing.\n\nAs far as we know, we are the only consciousness or the only life that’s out there. There might be other life, but we’ve seen no signs of it. And people often ask me, what do you know about the aliens and that, and I’m like, man, I tell you, I’m pretty sure I’d know if there were aliens. I have not seen any sign of aliens. And what if the military is hiding aliens in area 51 or something, you know; that’s a popular meme. (37:30) Well, let me tell you: the biggest, the fastest way to increase defense funding would be to bring up like, “Hey, we found an alien.” (…37:40) like, “Ah, there’s more money for defense, definitely”. Guaranteed, there (…37:46) would be like on display in two seconds. The reality is, as far as we know, this is the only place, at least in this part of the galaxy or in the Milky Way, where there is consciousness. And it’s taken a long time for us to get to this point.\n\nYou know, according to the geological records, Earth has been around for about four and a half billion years, although it was mostly molten magma for about half a billion years. So, but still, several billion years with at least bacterial life and multicellular life for several hundred million years. But here’s the interesting part. The sun is gradually getting hotter and bigger, and over time even in the absence of global warming – man-made stuff – the sun will expand, and it will overheat the Earth. My guess is probably… – On human timescales, this is a long time, but there are only several hundred million years left. That’s all, that’s all we got. Okay. Several hundred million years. But sort of from an evolutionary standpoint, basically, if it took an extra 10% longer for conscious life to evolve on Earth, it wouldn’t evolve at all because it would have been incinerated by the sun.\n\nWhat I’m saying is that it appears that consciousness is a very rare and precious thing, and we should take whatever steps we can to preserve the light of consciousness. The window has been opened only now – after four and a half billion years, is that window open. That’s a long time to wait, and it might not stay open for long. I’m pretty optimistic by nature, but there’s some chance, there’s some chance that window will not be open for long. I think we should become a multi-planet civilization while that window is open. And if we do, I think the probable outcome for Earth is even better because then Mars could help Earth one day. And so I think we should really do our very best to become a multi-planet species and to extend consciousness beyond Earth. And we should do it now. Thank you. (40:13)","textByLang":{"en":"Elon Musk: (00:21) This is, I think, the most inspiring thing that I’ve ever seen. And I just like to thank the SpaceX team and the suppliers, and the people of Boca Chica and Brownsville. Thank you for your support. And, just like, wow, what an incredible job by such a great team to build this incredible vehicle. First of all I want to start with that. I mean, I’m just so proud to work with such a great team. And it’s really windy here, by the way. If you’re watching this online, it is really windy.\n\nSo, the point of this presentation and this event is really… there are two elements to it. One is to inspire the public, get people excited about our future in space, and get people fired up about the future. There are so many things to worry about, so many things to be concerned about. There are many troubles in the world, of course, and these are important, and we need to solve them. But we also need things that make us excited to be alive, that make us glad to wake up in the morning and be fired up about the future and think, yeah, the future is going to be great.\n\nThis space exploration is one of those things; and becoming a spacefaring civilization being out there among the stars. This is one of the things that I know it makes me glad to be alive; I think it makes many people glad to be alive. It’s one of the best things. We are faced with a choice: Which future do you want? Do you want the future where we become a spacefaring civilization and are in many worlds and are out there among the stars or one where we are forever confined to Earth? And I say it is the first and I hope you agree with me. (02:30)\n\nThe critical breakthrough that’s needed for us to become a spacefaring civilization is to make space travel like air travel. So, with air travel, when you fly a plane, you fly that plane many times. I mean, the risk of stating the obvious, it really almost any motor transport, whether it’s a plane or a car, a horse, a bicycle is reusable. You use that motor transport many times. If you had to get a new plane every time you flew somewhere and even get to have two planes for the return journey, very few people could afford to fly. Or if you could use a car only once, very few people could afford to drive a car. So, the critical breakthrough that’s necessary is a rapidly reusable orbital rocket. This is basically the Holy Grail of space. The fundamental thing that’s required.\n\nAnd it is a very hard thing to do. It’s only barely possible with the physics of Earth. I mean, if Earth’s gravity was a little heavier, it would be impossible. And if Earth’s gravity was a little lighter, it would be quite easy. So, we’re really right on the cusp of what is physically possible. So, in order to create a rapidly reusable rocket and fully reusable orbital rocket, you have to have engines that have incredibly high, specific impulse (Isp), that essentially are extremely efficient.\n\nYou need to have a structure that is also incredibly mass efficient. And then that all needs to come back to the launch pad and be able to be refilled with propellant and flown again very quickly, just like an aircraft. It’s just because of the physics of Earth being quite a deep gravity well and having quite a thick atmosphere; this is a very tough but not impossible thing. But it is the most fundamental thing. With SpaceX, we started out 17 years ago, and the first rocket we designed was the Falcon 1, which was that guy right there. (05:00)\n\nWhen we started off, we were very naive. And in fact, the reason I should say… it’s September 28th; this is the 11th anniversary of the first time SpaceX reached orbit. Eleven years ago today, SpaceX made orbit for the first time. It was actually our fourth launch. And if that launch had not succeeded, that would have been the end of SpaceX. I’d run out of money, there were no more investors, and that would have been it. So, if that fourth launch had not succeeded, that would have been curtains.\n\nBut fortunately, fate smiled on us that day, and we made it to orbit. I have great respect for anyone who makes it to orbit. That is a hard thing. We were very naive, obviously very naive on many levels back then, because we did actually try to recover the first stage. The first stage had a parachute on it, and we thought, okay, we’ll just pop the parachute when it comes back into the atmosphere. Then it’ll land somewhere in the ocean, and we’ll go fish it out of the ocean with a boat. This does not work. I actually remember getting mad at the parachute supplier, like, your parachute doesn’t work. Nah, it wasn’t their fault.\n\nWhen the rocket comes in from space, that first stage is coming in like, you know, Mach 10 to 12, and it hits the atmosphere like it’s a concrete wall and ‘boom’. So, you actually have to orient the rocket carefully. You have to have aerodynamic surfaces, and you have to do an entry burn to slow it down. Then you’ve got to guide it through the atmosphere and then do a propulsive landing. This took us many, many attempts.\n\nWe actually did a video, a blooper reel of all the times we failed, which was a lot. (07:30) I think it might have taken us like 14 attempts or something before we finally successfully landed the rocket. If we’ve gone to the next slide, you can take a look at… – This is Grasshopper, that’s actually Falcon 9. It’s hard to tell the scale, but that’s a Falcon 9 size booster with one engine and big legs with giant shock absorbers; we didn’t know what the heck we were doing.\n\nElon Musk (voice-over): Now amazingly, Grasshopper had zero blessures; Grasshopper is still alive.\n\nElon Musk: They have Falcon1; what you saw there was a Falcon 9 size vehicle. What’s really kind of hard to grasp at a visceral level is that this giant ship will do the same thing that Grasshopper did. This thing is going to take off, fly to 65,000 feet, about 20 kilometers, and come back and land in about one or two months. So that giant thing – it’s really going to be pretty epic to see that thing take off and come back. And then hopefully, yeah, …(applause) – Yeah, it’s wild.\n\nThis is a quite radical… I’ll talk about it later in the presentation; it’s, this is quite a new approach to controlling a rocket – much more akin to a skydiver than a plane. But I’ll talk about that later. (10:00) So, going from Falcon 1 to Falcon 9 to Falcon Heavy, which we launched… actually, the first launch of Falcon Heavy was only February of last year. So, it’s only been about a year and a half since the first Falcon Heavy launch when we did two side-by-side booster landings. And I always liked this video. It was done by my friend Jonah.\n\nYeah. I never thought that would happen, actually. (12:30) I’m glad that it did. Some people were like, wait, why do we have the Roadster with the astronaut, you know, Starman. Actually, this came from a discussion with my friend Jonah. I was at his kitchen, and I was like, you know, normally when they do a rocket launch, there’s a launch of a rock of concrete, but that doesn’t sound very inspiring. So, what do you think the most sort of fun thing is that we could launch? And he was like, well, why don’t you launch your Tesla? And I was like, that’s a great idea.\n\nAnd another friend of mine, she said: “Why don’t you put a tiny Tesla on the dashboard?” So we put a tiny Tesla on the dashboard with a tiny Starman in the tiny Tesla. This is just to confuse the aliens in the future. They’ll be like, what the heck is this? You know, just want something that captures the imagination, gets people excited about space.\n\nSo, let’s see Starship. You can really see it right there, obviously. There’s a picture, more a rendering. It’s about 50 meters, sort of 165 feet or so.\n\nActually, I notice we have an error in our ship dry mass here; my apologies. I wish it was 85 tons. This ship dry mass has to be approximately 120 tons. The initial Mach 1 prototype is closer to 200 tons, and in series production, I think it’ll probably be about 120 tons. If we get really lucky, it might get down to 110; 99 would be super epic. So, in terms of its usefulness, it’ll be able to do about 150 tons with full reusability to orbit and back.\n\nThis is a very big number for full reusability. The very initial versions, we’re confident will do over a 100 tons, but I think there’s a clear path to 150 tons. The cost of a fully reusable system is basically (15:00) the cost of the propellant, which is mostly oxygen. This is three and a half tons of oxygen for every one ton of fuel. So, one of the advantages of this architecture over the Falcon architecture is that we actually use more oxygen per unit of fuel rather than less. Merlin or the Falcon architecture is about two and a half tons of oxygen for every one ton of fuel. This is three and a half tons of oxygen for every one ton of fuel. So when it ascends, it’s really mostly liquid oxygen because when you get to vacuum, there’s no air, basically.\n\nSo, the next slide. Earlier I was talking about how Starship enters and how it’s controlled.\n\nIt’s quite different from anything else. It’s really falling, and so we’re doing a controlled fall. With a rocket you’re actually trying to break as opposed to… you’re trying to create drag instead of lift, it’s really the opposite of an aircraft. You want the most amount of drag that you can produce. And you want some lift, especially when you’re in the upper atmosphere, mostly so that you don’t…  you can control the maximum heating rate. You want enough lift to keep yourself high in the low-density portion of the atmosphere, so you can burn off velocity. Basically, it goes like, if this is the Earth, it goes at about a 60 degree.\n\nMy hand is the rocket – it’s going at about 60 degrees. So when in orbit, you’re actually going at around 25 times the speed of sound horizontal to the ground. This is a very important concept that is counterintuitive to our normal daily life. Being in orbit, being in zero G is not about altitude. It’s about velocity. How fast are you going – horizontally? When something’s in orbit, it’s zooming around the Earth so fast that the outward acceleration, outward radial acceleration, is equal to the inward acceleration of gravity. And then you have zero gravity. This is why you actually have zero gravity.\n\nPeople often think the Space Station is stationary, (17:30) but it’s actually going around the world at 25 times the speed of sound or about 17,000 miles an hour. It always looks stationary in the pictures. And since there’s no air, you don’t have to have an aerodynamic structure. So it can be a totally crazy structure that doesn’t look like it should be able to go 25 times the speed of sound, but it does. And you can only feel the acceleration. You can’t feel velocity. People sometimes wonder, what does it feel like to go 25 times the speed of sound? Actually, it feels like nothing. Only accelerating to there feels like something.\n\nSo, the Starship is coming in – this platform is the Earth – it’s coming in at hypersonic velocity like this, sort of around a 60-degree angle. So, it comes like this and then starts falling and then just falls like a skydiver, and it’s just controlling itself – and then it turns and lands like that. That’s an incredibly elaborate explanation. There you can get a sense for it. This is much better.\n\nThere you go. See, same thing. It’ll look totally nuts to see that thing land. Yeah, that’ll be crazy.\n\nSo let’s talk about the Raptor engine. The ship will have a total of six engines.\n\nThree of the sea-level variety of Raptor, and those are actually on the rocket right now. So, we have the three sea-level… in fact, that’s a picture of just inside – that’s what it looks like. So, we’ve got the three sea-level Raptor engines, and they gimbal, which means that the whole engine moves. So, the way the rocket steers is by moving the entire engine. Whereas an aircraft engine is static, and you move by moving the control surfaces, like the ailerons and rudder and elevator and flaps… – The rocket – when the engines are powered, you moved the entire engine to steer it. The Starship will have three sea-level engines that move up to about 15 degrees angle and three vacuum engines that are optimized for efficiency in orbit that will not move. They will be just fixed it in place.\n\nAnd that allows us to have the biggest bell nozzle (20:00) for the vacuum Raptor engines. Aspirationally, the target is a 380 second Isp for the vacuum engine. This is a very… – In sort of space geek terms, this is like really a great number. And even for the steel alloy engines to get over a 350 second Isp is also really great. So actually… – sorry, I’m looking at the slide here, and you’re not. So, that’s what I meant by ‘it looks like that on the inside’… – go back one slide. That’s the inside of the Starship right now.\n\nThat’s what it looks like in the base. All right. Then heat shield.\n\nI’ve gone through various iterations of heat shield. There’s a lot of ways to skin the cat here.\n\nUltimately, we decided to have heat shield hexagonal tiles, ceramic tiles, basically like a tiny glass vermicelli at a microstructure level. Very light, but very crack resistant – essentially glass tiles. And there, because Starship is a steel construction… – At first, it feels like, “Oh, it’s steel. Does that mean it’s heavy?” No, actually, it’s the lightest construction. I think the best design decision on this whole thing is 301 stainless steel. Because at cryogenic temperatures, a 301 stainless actually has about the same effective strength as an advanced composite or aluminum-lithium. Unlike most steals, which get brittle at low temperatures, 301 stainless gets much stronger.\n\nAnd if it’s in the extra hard condition, meaning it’s cold-rolled to extra hard condition, it also gets way stronger. Actually, its strength-to-weight ratio at cryogenic temperatures is equivalent, or even perhaps slightly better than advanced composites or aluminum-lithium. This is not well appreciated. (22:30) Because if you just look at the materials manual and say like, what is the strength of stainless steel? It looks much weaker than it is. If you say, what is the strength at cryogenic temperature? Oh, much stronger; at very low temperature, almost twice as strong. That’s when it becomes better than carbon fiber or aluminum-lithium.\n\nAnd this is another benefit: It also has a high melting temperature. So, for a reusable ship, you’re coming in like a meteor. You want something that does not melt at a low temperature. You want something that melts at a high temperature. And this is where steel is extremely good as well. Steel has a melting temperature around sort of 1500 degrees centigrade whereas aluminum, you know, maybe 300 or 400 degrees, and same thing for carbon fiber. And that’s really pushing it.\n\nHaving that much higher melting temperature means that you don’t need any shielding on the leeward side of the ship when it comes in for entry. And the shielding you need on the windward side – the hot side – is massively reduced because the thickness of the tile is actually for a reusable system – It’s dependent on what back shell temperature, like how hot does the back of the tile that interfaces with the airframe get. And because the steel can take a much higher temperature, your heat shield, even on the windward side, is much lighter. But the net effect is that a 301 stainless steel rocket is actually the lightest possible reusable architecture.\n\nThen, to come to cost. The carbon fiber we were using was $130 a ton. The steel is $2,500 a ton. Oh, sorry, $130,000 a ton versus $2,500 a ton. That makes much more sense. It’s $130,000 a ton for the carbon fiber and $2,500 a ton for the steel. So, the steel is about 2% of the cost of the carbon fiber. So, this is a good thing we changed from carbon fiber to steel, by far. (25:00) And it’s very easy to weld stainless steel, the evidence being that we welded it outdoors without a factory. Great skills by the team, but with carbon fiber, this is impossible; with aluminum-lithium, also impossible. But steel is easy to weld,  and it is resilient to the elements.\n\nAnd also, actually, (… 25:36), like on Mars, you can cut that up, you can weld it, you can modify it. No problem. Yeah. That’s a good point. You’re out there on the Moon or Mars; you want something that you can modify, that you can cut up and use for other things. That’s like for sure a great thing. So anyway, steel – obviously I’m in love with steel; I had to say it, you know. So, let’s see, going on to the booster.\n\nSo, the booster is designed to take up to 37 Raptor engines. I’m not sure if we’ll go that high, but you can really have 31. I think the minimum number you’d want is maybe around 24. But the booster is designed to be able to take multiple engines out, so you can actually add or subtract engines as you’d like. You basically just need a lot of force pushing up. Over time, I think you probably want around a 7,500 ton force rocket which is about twice the thrust of a Saturn V, a little more than twice the thrust, and on a roughly 5,000 ton gross liftoff mass for roughly one and a half thrust-to-weight.\n\nFor a reusable rocket, you actually want a high thrust-to-weight rather than with an expendable rocket, where you want a low thrust-to-weight, because any thrust-to-weight below one is not useful; like, if you have a less thrust than your weight, you don’t move. So, you actually want a high thrust-to-weight for a reusable rocket. This is a very important (27:30) design optimization change. So that’s why I think more engines are probably good and getting up to around 7,500 tons over time and a one and a half thrust-to-weight ratio, or more.\n\nWe think we’re probably going to adjust the grid fins designed to be kind of like a diamond shape. It looks cooler. It works better too. And then the rear fins are actually just legs. They’re not needed for stabilization or guidance. They’re essentially there for legs.\n\nAll right. So let’s go into some of the development testing. This is a Raptor firing.\n\nAll right. And then, obviously, we had a Raptor fire on the Starhopper. Yeah.\n\nIt’s kind of hard to see it to appreciate scale, but it’s the same diameter as the Starship. And obviously, it’s just right over there. So, it’s kind of hard to tell if it’s the size of a trashcan or, you know, how big it is, but it’s about… the body diameter is about 9 meters or 30 feet and not including the legs span. (30:00) Yeah. So, this gives you a sense of size.\n\nSo the little pixel there… little pixels are a human. And then there’s the Hopper next to it, the Millennium Falcon for comparison, then Starship, which is what you see before you. And then what it will look like with the full stack, which is almost two and a half times as tall as this vehicle. This simulation will give you a sense of the scale of things.\n\nElon Musk (voice-over): It slightly reminds me of a scene from Spaceballs.\n\nElon Musk (voice-over): This is the orbital refilling. Orbital refilling is extremely important for getting to Mars and getting to the Moon to establish a city on Moon or Mars. This is a vital step.\n\nA rapidly reusable orbital launcher rocket is… a rapidly reusable rocket is required for – alliteration – for getting a breakthrough in cost of access to space; that you don’t throw the rockets away every flight. But another key step is refilling on orbit so that Starship can get to orbit with, let’s say, 150 tons of payload for the Moon or Mars or beyond. And then it can get tanked up to fill up its propellant tanks and so that it can depart from low Earth orbit (35:00) with 1,200 tons of propellant. This is a very big thing so that your delta velocity is enough to transport about 150 tons to the surface of the Moon or Mars with full reusability and orbital refilling.\n\nThe orbital refilling is actually a simplified version of what SpaceX does in docking with the Space Station. So it’s actually harder to dock with the Space Station than it is to do orbital refilling, but in practicing docking with the Space Station, SpaceX has also learned how to rendezvous and dock in orbit in a complex environment. So this is one of the other critical pieces of the puzzle needed to establish a base on the Moon and Mars, a city, ultimately. And yeah, so those are the critical ingredients. So, we think it will be very exciting to have a base on the Moon.\n\nEven if it’s just a science base that… – for example, we have a base at Antarctica. Many countries have bases in Antarctica for science research, and this would be an incredible area for research. So whether or not people want to live on the Moon, there’s definitely a lot of science to be done. And I think it’s close as well. So, that would be quite exciting to do. And then, of course, we can go to other places in the solar system like Saturn. But the critical thing that we need to focus on, I think, is the fastest path to a self-sustaining city on Mars. This is the fundamental thing.\n\nAs far as we know, we are the only consciousness or the only life that’s out there. There might be other life, but we’ve seen no signs of it. And people often ask me, what do you know about the aliens and that, and I’m like, man, I tell you, I’m pretty sure I’d know if there were aliens. I have not seen any sign of aliens. And what if the military is hiding aliens in area 51 or something, you know; that’s a popular meme. (37:30) Well, let me tell you: the biggest, the fastest way to increase defense funding would be to bring up like, “Hey, we found an alien.” (…37:40) like, “Ah, there’s more money for defense, definitely”. Guaranteed, there (…37:46) would be like on display in two seconds. The reality is, as far as we know, this is the only place, at least in this part of the galaxy or in the Milky Way, where there is consciousness. And it’s taken a long time for us to get to this point.\n\nYou know, according to the geological records, Earth has been around for about four and a half billion years, although it was mostly molten magma for about half a billion years. So, but still, several billion years with at least bacterial life and multicellular life for several hundred million years. But here’s the interesting part. The sun is gradually getting hotter and bigger, and over time even in the absence of global warming – man-made stuff – the sun will expand, and it will overheat the Earth. My guess is probably… – On human timescales, this is a long time, but there are only several hundred million years left. That’s all, that’s all we got. Okay. Several hundred million years. But sort of from an evolutionary standpoint, basically, if it took an extra 10% longer for conscious life to evolve on Earth, it wouldn’t evolve at all because it would have been incinerated by the sun.\n\nWhat I’m saying is that it appears that consciousness is a very rare and precious thing, and we should take whatever steps we can to preserve the light of consciousness. The window has been opened only now – after four and a half billion years, is that window open. That’s a long time to wait, and it might not stay open for long. I’m pretty optimistic by nature, but there’s some chance, there’s some chance that window will not be open for long. I think we should become a multi-planet civilization while that window is open. And if we do, I think the probable outcome for Earth is even better because then Mars could help Earth one day. And so I think we should really do our very best to become a multi-planet species and to extend consciousness beyond Earth. And we should do it now. Thank you. (40:13)"},"languages":["en"],"lang":"en","transcriptSource":"https://elonmuskinterviews.wordpress.com/2021/03/27/starship-presentation-2019-english/"}]}