Tesla AI Day 2022
Questions on ElonsFAQ asked in this interview (1)
Transcript
14:33Elon MuskAll right, welcome everybody. Give everyone a moment to get back in the audience and all right, great. Welcome to Tesla AI Day 2022.
14:57Elon MuskWe'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 you know, last year it was just a person in a robot suit.
15:16Elon MuskBut we've come a long way and it's, I think, you know, compared to that, it's going to be very impressive. And we're going to 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.
15:37Elon MuskAnd 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 I think actually Tesla's a good entity 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.
16:03Elon MuskAnd I think that's actually a good thing. So if I go crazy, you can fire me. This is important. Maybe I've gone crazy. I don't know. So, yeah, so we're going to talk a lot about our progress in AI autopilot as well as progress, progress in with Dojo.
16:23Elon MuskAnd then we're going to bring the team out and do a long Q and A. So you can ask tough questions,
16:31Sri (Tesla)whatever
16:31Elon Muskyou'd like, existential questions, technical questions, but we want to have as much time for Q and A as possible. So let's see with that. You guys want.
16:45Milan KovacHey guys, I'm Milan. I work on Autopilot and the Tesla.
16:48Lizzie MiskovetzAnd I'm Lizzy, mechanical engineer on the project as well.
16:53Elon MuskOkay, so should we bring up the bot?
16:58Lizzie MiskovetzBefore we do that, we have one, one little bonus tip for the day. This is actually the first time we try this robot without any backup support. Cranes, mechanical mechanisms, no cables, nothing.
17:12John EmmonsYeah, want to do it with you
17:13Milan Kovacguys tonight, but it's the first time.
17:15Pete BannonLet's see.
17:16Lizzie MiskovetzYou ready?
17:17Kate Park (Tesla)Let's go.
17:42Kate Park (Tesla)Go. Sam,
18:10Eric (Tesla)I think the bug got some boobs here.
18:24Tesla PresenterSo this is essentially the same self
18:27Milan Kovacdriving computer that runs in your Tesla cars.
18:29Tesla PresenterBy the way,
18:34Eric (Tesla)This is the.
18:35Elon MuskIt's literally the first time the robot has operated without a tether was on stage tonight.
19:14Elon MuskSo 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 show you some videos now of the robot doing a bunch of other things.
19:27Tesla PresenterYeah.
19:28Elon MuskWhich are less risky. Yeah.
19:31Milan KovacWe should close the screen, guys.
19:34Tesla PresenterYeah,
19:40Tesla PresenterYeah.
19:41Milan KovacWe wanted to show a little bit more what we've done over the past
19:43Eric (Tesla)few months with the pod and just
19:45Milan Kovacwalking around and dancing on stage. Just humble beginnings. But you can see the autopilot neural networks running as is, just retrained for the bot directly on that, on that new platform.
19:59Milan KovacThat's my watering can.
20:00Elon MuskYeah. 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 it should pick up.
20:11Elon MuskPicking it up. Yeah.
20:18Milan KovacWe use the same process as we did for autopilot to collect data in train neural networks that we then deploy on the robot. That's an example that illustrates the upper
20:27Elon Muskbody a little bit more.
20:31Milan KovacSomething that we'll like try to nail down in a few months. Over the next few months, I would say to perfection.
20:38Lizzie MiskovetzThis is really an actual station in the Fremont factory as well that it's working at.
20:43Phil (Tesla)Yep.
20:43Elon MuskSo.
20:54Lizzie MiskovetzAnd that's not the only thing we have to show today, right?
20:56Elon MuskYeah, absolutely. So what you saw was what we call Bumble C, 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.
21:13Elon MuskThe team's done an incredible job and we actually have an Optimus bot with fully Tesla designed and built actuators, battery pack control system, everything. It wasn't quite ready to walk, but I think it will walk in a few weeks.
21:31Elon MuskBut we wanted to show you the robot something that's actually fairly close to what will go into production and show you all the things it can do. So let's bring it out.
21:44Elon MuskDo it,
22:06Kate Park (Tesla)Sam.
22:33Elon MuskSo here you're seeing Optimus with
22:39Tesla Presenterthe
22:39Elon Muskdegrees of freedom that we expect to have in optimus production unit 1, which is the ability to move all the fingers independently, to have the thumb have two degrees of freedom.
22:51Elon MuskSo it has opposable thumbs and both left and right hand. So it's able to operate tools and, and do useful things. Our goal is to make a useful humanoid robot as quickly as possible.
23:04Elon MuskAnd we've also designed it using the same discipline that we use in designing the car, which is to say to design it for manufacturing such that it's possible to make the robot in high volume at low cost with high reliability.
23:21Elon MuskSo that's incredibly important. I mean you full scene, very impressive humanoid robot Demonstrations, and that's great, but what are they missing? They're missing a brain. They don't have the intelligence to navigate the world by themselves.
23:37Elon MuskAnd they're also very expensive and made in low volume, whereas Optimus is designed to be an extremely capable robot, but made in very high volume, probably ultimately millions of units.
23:52Elon MuskAnd it is expected to cost much less than a car. I'll just bring it directly to the right here. I would say probably less than $20,000 would be my guess.
24:09Elon MuskThe potential for optimists is, I think, appreciated by very few people.
24:15Felix (Tesla)Hey,
24:19Elon Muskas usual, Tesla demos are coming in hot. So.
24:23Tesla PresenterOkay, that's good, that's good.
24:26Elon MuskYeah. The team's put up. The team has put in an incredible amount of work. Working days, you know, seven days a week, burning the 3:00am oil to get to the demonstration today.
24:41Elon MuskSuper proud of what they've done. They've really done a great job. I'd just like to give a hand to the whole Optimus team.
24:57Elon MuskSo, you know, now there's still a lot of work to be done to refine Optimus and improve it. Obviously this is just Optimus 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, to join Tesla and help make it a reality and bring it to fruition at scale such that it can help millions of people.
25:28Elon MuskAnd the potential, like I said, really boggles the mind because you have to say, like, what is an economy? An economy is sort of productive entities. Times the productivity capita times output productivity per capita.
25:45Elon MuskAt the point at which there is not a limitation on capita, it's not clear what an economy even means at that point. An economy becomes quasi infinite. So what, you know, taken to fruition in the hopefully benign scenario, the, this means a future of abundance, a future where there is no poverty, where people, you can have whatever you want in terms of products and services.
26:22Elon MuskIt 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 safe. 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, is very important and should not be overlooked.
26:50Elon MuskI 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.
27:03Elon MuskLike, it's very important that I can't just do what I want. You know, sometimes people think, but it's not true.
27:10Phil (Tesla)So,
27:13Elon Muskyou know, it's very important that the corporate entity that has, that makes this happen is something that the public can properly influence. And so I think the Tesla structure is ideal for that.
27:32Elon MuskLike I said, self driving cars will certainly have a tremendous impact on the world. I think they will improve, improve the productivity of transport by at least a half order of magnitude, perhaps an order of magnitude, perhaps more.
27:52Elon MuskOptimus I think has maybe a two order of magnitude potential improvement in economic output. It's not clear what the them it actually even is. So but we need to do this in the right way.
28:15Elon MuskWe need to do it carefully and safely and ensure that the outcome is one that is beneficial to civilization and one that humanity wants. This is also extremely important, obviously.
28:31Elon MuskSo, and I hope you will consider joining Tesla to achieve those goals. At Tesla, we really care about doing the right thing here or aspire to do the right thing and really not pave the road to hell with good intentions.
28:53Elon MuskAnd I think the road to hell is mostly paved with bad intentions. But every now and again there's a good intention in there. So we want to do it, do the right thing.
29:00Elon MuskSo, you know, consider joining us and helping make it happen. With that, let's, let's move on to the next phase.
29:08Lizzie MiskovetzRight on. Thank you, Elon.
29:15Lizzie MiskovetzAll 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.
29:25Lizzie MiskovetzWe 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.
29:36Lizzie MiskovetzWorking 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 foundation of sort of the vehicle design process.
29:52Lizzie MiskovetzYou 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.
30:00Lizzie MiskovetzYou'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.
30:11Lizzie MiskovetzWe have this incredible range of motion, typically really amazing strength. 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.
30:31Lizzie MiskovetzBut 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.
30:44Lizzie MiskovetzAnd 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.
30:56Lizzie MiskovetzThat's great. 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.
31:17Lizzie MiskovetzSo let's talk about this latest generation in some detail, shall we? So on the screen here, you'll see in orange our actuators, which we'll get to in a little bit, and in blue, our electrical system.
31:29Lizzie MiskovetzSo 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.
31:37Kate Park (Tesla)For this design.
31:39Lizzie MiskovetzAgain, 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.
31:56Lizzie MiskovetzHow are we going to do that? Well, we're going to reduce our part count and our power consumption of every element possible, possible. We're going to do things like reduce the sensing and the wiring at our extremities.
32:07Lizzie MiskovetzYou 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.
32:21Lizzie MiskovetzSo 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.
32:30Sri (Tesla)Work.
32:31Lizzie MiskovetzWhat'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.
32:48Lizzie MiskovetzWe'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.
33:05Lizzie MiskovetzAnd 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.
33:16Lizzie MiskovetzAlso 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.
33:30Lizzie MiskovetzBut because it's different in requirements and in form factor, we're going to change a few things first. So we still are going to, it's going to do everything that a human brain does, processing vision data, making split second decisions based on multiple sensory inputs, and also communications.
33:48Lizzie MiskovetzSo 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.
34:02Lizzie MiskovetzSo now that we have our sort of core, we're going to need some limbs on this guy. And we'd love to show you a little bit about our actuators and our fully functional hands as well.
34:12Lizzie MiskovetzBut 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.
34:24Tesla PresenterThank you, lizzie.
34:29Tesla PresenterTesla 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 Model 3 superimposed on top of the actual physical crash.
34:41Tesla PresenterIt's actually incredible how accurate it is. Just to give you an idea of the complexity of this model, it includes every nut, bolt and washer, every spot weld, and it has 35 million degrees of freedom.
34:53Tesla PresenterQuite 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?
35:09Tesla PresenterWell, we can make a model. And since we have crash software, we use 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.
35:22Tesla PresenterWe don't want it to, for example, break its gearbox and its arms. That's equivalent to 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 been given.
35:37Tesla PresenterWe can 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.
35:51Tesla PresenterThese 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.
36:03Tesla PresenterThat 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.
36:22Tesla PresenterWho is it 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.
36:34Tesla PresenterWe haven't got the luxury of making from carbon fiber and titanium. We want to make them on plastic. Things are not quite so stiff. So we can't have these high targets, I call them dumb targets.
36:46Tesla PresenterWe've got to make them work at lower targets. So is that going to work? Well, if you think about it, sorry about this, but we're just bags of soggy jelly and bones thrown in.
36:57Tesla PresenterWe're not high frequency. If I stand on my leg, I don't vibrate at 10 hertz. People operate at a low frequency. So we know the robot actually can. It just makes controls harder.
37:08Tesla PresenterSo we take the information from this, the modal data and the stiffness, and feed that into the control system that allows it to walk. Just changing tack slightly. Looking at the knee, we can take some inspiration from biology and we can look to see what the mechanical advantage of the knee is.
37:27Tesla PresenterIt turns out it actually represents 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.
37:40Tesla PresenterSo you'd expect a non linear function. And in fact the biology is nonlinear. This matches it quite accurately. So that's the representation. The four bar link is obviously not physically a four bar link.
37:53Tesla PresenterAs 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 this graph. This is showing picking things up, walking, squatting, the tasks I said we did on the stress.
38:09Tesla PresenterAnd that's the torque seen at the knee against the knee bend on the horizontal axis. This is showing the requirement for the knee to do all these tasks. I've then put a curve through it, surfing over the top of the peaks.
38:23Tesla PresenterAnd that's saying this is what's required to make the robot do these tasks.
38:32Tesla PresenterSo if we look at the four bar lead, 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 lowered the force.
38:44Tesla PresenterThat'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? Well, the blue curve is actually if we didn't have a four bar link, we just had an arm sticking out my leg here with an actuator on it, a simple two bar link.
39:00Tesla PresenterThat's the best you could do with a simple two bar link. And it shows that that would create much more force in the actuator which would not be efficient.
39:09Tesla PresenterSo what does that look like in practice? Well, as you'll see, it's very tightly packaged in the knee. You'll see it go transparent on a second. You'll see the four bar link there.
39:19Tesla PresenterIt's operating on the actuator. This is determining the force and the displacements on the actuator. I now pass you over to Constantino to tell you a lot more detail about how these actuators are made and designs optimized.
39:31Tesla PresenterThank you.
39:34Elon MuskThank you, Michael.
39:41Tesla PresenterSo 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.
39:56Tesla PresenterThe 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 0-60 mph time or drive the city drive size, while the robot that has 28 actuators, it's not obvious what are the tasks at actuator level.
40:32Tesla PresenterSo 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.
40:56Tesla PresenterTo 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 laid over an efficiency map of an actuator.
41:14Tesla PresenterAnd 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.
41:33Tesla PresenterThen 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.
41:44Tesla PresenterSo 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.
41:58Tesla PresenterWe parse our cloud again for every joint spec. And the red axis, this time denotes the bespoke actuator designs for every join. The problem here is that we have too many unique actuator designs.
42:12Tesla PresenterAnd 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.
42:23Tesla PresenterTherefore, 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.
42:37Tesla PresenterSo the resulting portfolio is six actuators and they show in a color map in the middle figure. 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.
42:59Tesla PresenterThe 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.
43:12Tesla PresenterAnd the gear train is a strain wave gear. There are three integrated sensors here and a bespoke permanent magnet machine, the linear actuator.
43:31John EmmonsI'm sorry.
43:33Tesla PresenterThe 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.
43:58Tesla PresenterAnd I will let you enjoy the video.
44:07Tesla PresenterSo our actuator is able to lift.
44:19Tesla PresenterA half ton nine foot concert grand piano. And.
44:31Tesla PresenterThis 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, or quadriceps muscles can do the same thing.
44:43Tesla PresenterIt's just that the knee is an up gearing linkage system that converts the force into velocity at the end effector of our heels for purposes of giving to the human body agility.
44:56Tesla PresenterSo 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.
45:06Tesla PresenterThank you very much. Thanks Josephinos. So we just saw how powerful a
45:14Felix (Tesla)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. It has tens of thousands of tactile sensors, and it has the ability to grasp and manipulate almost every object in our daily lives.
45:36Felix (Tesla)For our robotic hand design, we were inspired by biology. We have five fingers and an opposable thumb. Our fingers are driven by metallic tendons that are both flexible and strong.
45:48Felix (Tesla)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?
46:00Felix (Tesla)Well, the main reason is that our factories and 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.
46:17Felix (Tesla)The converse there 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.
46:27Felix (Tesla)Some basic stats about our hand is that it has 6 actuators and 11 degrees of freedom. It has an in hand controller which drives the fingers and receives sensor feedback.
46:37Felix (Tesla)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.
46:48Felix (Tesla)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 to the objects that's being grasped.
46:58Felix (Tesla)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.
47:07Felix (Tesla)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 real bot to life.
47:18John EmmonsThanks, Michael.
47:24Milan KovacAll right, so all those good 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 on Autopilot over the past few years.
47:36Milan KovacMost of those components ported quite easily over to the bots environment. If you think about it, we're just moving from a robot on wheels to a robot on legs.
47:45Milan KovacSo some of the components are pretty similar and some other require more heavy lifting. So for example, our computer vision neural networks reported directly from Autopilot to the bots situation.
47:59Milan KovacIt'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.
48:07Milan KovacThe 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.
48:25Milan KovacFor 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?
48:45Milan KovacSo 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.
48:58Milan KovacAnd 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.
49:11Milan KovacAnd this is literally the Autopilot simulator to which we've integrated the robot's locomotion code. And this is a video of the motion control code running in the OPILOT simulator simulator showing the evolution of the robot's work over time.
49:26Milan KovacAnd so, as you can see, we started quite slowly in April and start accelerating as we unlock more joints and deploy more advanced techniques like arms balancing over the past few months.
49:37Milan KovacAnd so locomotion is specifically one component that's very different as we're moving from the car to the bots environment. And so I think it warrants a little bit more depth.
49:45Milan KovacAnd I'd like my colleagues to start talking about this now.
49:56Elon MuskThank you, Milan.
49:58Felix (Tesla)Hi, everyone, I'm Felix, I'm a robotics engineer on the project and I'm going
50:02Tesla Presenterto talk about walking.
50:04Felix (Tesla)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 an engineering perspective.
50:15Felix (Tesla)For 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.
50:28Felix (Tesla)Also having an energy efficient gait. 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.
50:45Felix (Tesla)So now humans do all of this naturally. But as engineers or roboticists, we have to think about these problems and therefore I'm going to show you how we address them in our locomotion planning and control strategy.
50:57Felix (Tesla)So we start with locomotion planning and our representation of the bot. That means a model of the robot's 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.
51:16Felix (Tesla)This means feasible trajectories with respect to the assumptions of our model. The planner currently works in three stages. It starts planning footsteps and ends with the entire motion photo system.
51:29Felix (Tesla)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.
51:46Eric (Tesla)Just as the humans.
51:48Felix (Tesla)Just as humans do. And this gives us larger stride 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.
52:04Felix (Tesla)As we all know, plans are good, but we also have to realize them in reality. Let's see how we can do this.
52:20Elon MuskThank you, Felix.
52:21Tesla PresenterHello 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.
52:33Tesla PresenterLet's see what happens. It takes a couple steps and falls down. Well, that's a little disappointing, but we
52:43Phil (Tesla)are missing a few key pieces here
52:45Tesla Presenterwhich will make it work. 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.
52:59Tesla PresenterIt also expresses its intention through trajectories and wrenches, wrenches of forces and torques that it wants to exert on the world. To locomote reality is way more complex than any FEMOR model.
53:16Tesla PresenterAlso, 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?
53:30Tesla PresenterWell, 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 biped locomotion. So what can we do about it?
53:45Tesla PresenterWell, we measure reality. We use sensors and our understanding of the world to do state estimation and state estimation. 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's walking in the office environment.
54:07Tesla PresenterNow 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.
54:31Tesla PresenterThe robot certainly doesn't appreciate being poked, but it does 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
54:45Phil (Tesla)and arms to be useful.
54:48Tesla PresenterLet's talk about manipulation.
55:00Eric (Tesla)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.
55:17Eric (Tesla)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.
55:33John EmmonsSo let's say we have a human
55:34Eric (Tesla)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 locations of the hands, the elbows, the torso.
55:46Eric (Tesla)We can map that to the robot using inverse kinematics. And if we collect a lot of
55:51John Emmonsthese, now we have a library that we can work with.
55:55Eric (Tesla)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.
56:21Eric (Tesla)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 what's next for the Optimus Tesla bot.
56:36Tesla PresenterThanks.
56:38Elon MuskThanks, Eric.
56:45Milan KovacRight, so hopefully by now you guys got a good idea of what we've been up to over the past few months. We started being something that's usable, but it's far from being useful.
56:54Milan KovacThere's still a long and exciting road ahead of us. I think the first thing within the next few weeks is to get Optimus at least at par with Bumble C, the other bug prototype you saw earlier and probably beyond.
57:07Milan KovacWe 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 iron out all the elements needed to deploy this product in the real world.
57:20Milan KovacI was mentioning earlier, you know, indoor navigation, graceful fault management, or even servicing all components needed to scale this product up. But 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 and make this product a reality and change the entire economy.
57:42Milan KovacSo I would like to thank the entire Optimus 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.
57:51Milan KovacThank you very much.
58:07Tesla PresenterHey, everyone.
58:10Ashok ElluswamyHi, I'm Ashok. I lead the Autopilot team alongside Milan. God, it's going to be so hard to top that optimist 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.
58:30Ashok ElluswamyWe have been working on the software to add higher and higher levels of autonomy this time around. Last year, we had roughly 2000 cars driving our FSD beta software.
58:43Ashok ElluswamySince then, we have significantly improved the software's robustness and capability that we have now shipped it to 160,000 customers as of today.
58:54Tesla PresenterThank you.
59:00Ashok ElluswamyThis did not come for free. It came from the sweat and blood of the engineering team. Over the last one year, for example, we trained 75,000 neural network models just last one year.
59:11Ashok ElluswamyThat's roughly a model every eight minutes that's coming out of the team. And then we evaluate them on our large clusters, and then we ship 281 of those models that actually improve the performance of the car.
59:24Ashok ElluswamyAnd this space of innovation is happening throughout the stack. The planning Software, the infrastructure, the tools, even hiring, everything is progressing to the next level. The FSG beta software is quite capable of driving the car.
59:42Ashok ElluswamyIt 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.
59:56Ashok ElluswamyAll of this comes from the 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 to form the world model around the car.
1:00:08Ashok ElluswamyAnd 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.
1:00:22Ashok ElluswamyThis 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, what have you, it predicts.
1:00:39Ashok ElluswamyIf 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.
1:00:59Ashok ElluswamySo it's actually a really difficult problem for typical computer vision techniques to predict the set of planes and their connectivities. So we reached all the way into language technologies and then pulled the state of the art from other domains and not just computer vision.
1:01:12Ashok ElluswamyTo make this task possible for vehicles, we need their full kinematic 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 kinematic state.
1:01:31Ashok ElluswamyThe positions, velocities, acceleration, jerk, all of that directly comes out of the networks with minimal post processing. That's really fascinating to me because how is this even possible? What world do we live in that this magic is possible?
1:01:44Ashok ElluswamyThat these networks predicts four 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 required tons of data.
1:01:57Ashok ElluswamySo we had to build sophisticated auto labeling systems that churn through raw sensor data, 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.
1:02:15Ashok ElluswamyOn 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.
1:02:36Ashok ElluswamyAnd once we know the failures, we mine 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.
1:02:48Ashok ElluswamyAnd we do this for every task that runs in the car.
1:02:51Tesla PresenterYeah.
1:02:52Milan KovacAnd to train these new massive neural networks this year we extended our training infrastructure by roughly 40 to 50%. So that sits us at about 14,000 GPUs today across multiple training clusters in the United States.
1:03:07Milan KovacWe also worked on our AI compiler which now supports new operations needed by those neural networks and map them to 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.
1:03:26Milan KovacEssentially two independent computers interconnected within the same full self driving computer. And to make this possible, we had to keep a tight control on the end to end latency of this new system.
1:03:37Milan KovacSo we deployed more advanced scheduling code across the 4fsd platform.
1:03:43Ashok ElluswamyAll 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. 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.
1:04:07Ashok ElluswamyToday 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.
1:04:22Ashok ElluswamyWith that I welcome Padil to start talking about the planning section.
1:04:33Paril JainHi all, I'm Paril Jain. 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.
1:04:49Paril JainRight as we 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.
1:05:08Paril JainSo 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.
1:05:27Paril JainSo 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 and lateral profile.
1:05:36Paril JainNow obviously we are being a jerk to the pedestrian and we would spook the pedestrian and his cute pit. We could have moved forward slowly, shot for a gap between the pedestrian and the vehicle from the right.
1:05:48Paril JainAgain, 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.
1:06:09Paril JainNow, 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 ride when you assert in front of it?
1:06:25Paril JainRelying 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.
1:06:42Paril JainNow, how much ever you optimize, there's gonna be a limit to how fast you can run this optimization problem. It will be close to order of 10 milliseconds, even after a lot of incremental approximations.
1:06:54Paril JainNow 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.
1:07:08Paril JainThe planner needs to make a decision every 50 milliseconds. So how do we solve this in real time? We rely on a framework what we call as interaction search, which is basically a parallelized research over a bunch of maneuver trajectories.
1:07:23Paril JainThe state space here corresponds to the kinematic state of ego, the kinematic state of other agents, the nominal future multimodal predictions and all the static entities in the scene.
1:07:36Paril JainThe 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.
1:07:50Paril JainLet'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 model moving objects. These get represented as sparse attractions as well as latent features.
1:08:05Paril JainWe 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 seed trajectories using a combination of classical optimization approaches as well as our network planner, again trained on data from the customer fleet.
1:08:28Paril JainNow, once we get a bunch of these fleet 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.
1:08:43Paril JainObviously 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.
1:08:54Paril JainBuilding 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 tricks here lie in evaluation of each of this node of the tree search inside each node.
1:09:12Paril JainInitially, 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.
1:09:25Paril JainNow, even though this is fairly good number, when you want to evaluate more than 100% tractions, this does not scale. So we ended up building lightweight queryable networks that you can run in the loop of the planner.
1:09:38Paril JainThese 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 runtime down to close to 100 microseconds per action.
1:09:58Paril JainNow, 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.
1:10:07Paril JainSo you need to do 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.
1:10:16Paril JainWhat is the jerk and access required for a given maneuver? The customer fleet data plays an important role. Here again we run two sets of again lightweight variable networks, both really augmenting each other.
1:10:29Paril JainOne of them trained from interventions from the FSC beta fleet, which gives a score on how likely is a given maneuver to result in interventions over the next few seconds.
1:10:38Paril JainAnd 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.
1:10:59Paril JainThe 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.
1:11:14Paril JainNow a lot of what I described was with respect to the agents we could observe in the scene. But the same framework extends to objects behind occlusions. We use the video feed from eight cameras to generate the 3D occupancy of the world.
1:11:30Paril JainThe blue mask here corresponds to the visibility region we call basically gets blocked at 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.
1:11:45Paril JainNow if you model the spawn regions and the state transitions of these 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.
1:11:59Paril JainNow I'll pass it on to Phil to describe more on how we generate these occupancy networks. Thank you.
1:12:11Phil (Tesla)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 world using in 3D around our cars.
1:12:23Phil (Tesla)And it is currently not shown in our customer facing visualization. And what you will see here is the raw network output from our internal dev tool. The occupancy network takes video streams of all our eight cameras as input produces a single unified volumetric occupancy in vector space directly for every 3D location around our car.
1:12:48Phil (Tesla)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.
1:13:16Phil (Tesla)Occupancy flow is also predicted for motion. Since the model is a generalized network, it does not tell static and dynamic object explicitly. It is able to produce and model the random motion such as a swerving trainer here.
1:13:33Phil (Tesla)This network is currently running in all Teslas with FSD computers and it is incredibly efficient for runs about every 10 milliseconds with our neural net accelerator. So how does this work?
1:13:46Phil (Tesla)Let's take a look at the architecture. First we rectify each camera images with the camera calibration. And the images we're shown here were given to the network. It's actually not the typical 8 bit RGB image as you can see from the first image on top.
1:14:02Phil (Tesla)We're giving the 12 bit raw photocount image to the network. Since it has 4 bits more information, it has 16 times better dynamic range as well as reduced latency.
1:14:14Phil (Tesla)Since we don't have to run ISP in the loop anymore, we use a set of reglets and biofpns as a backbone to extract image space features. Next we construct a set of 3D position query along with the image space features.
1:14:30Phil (Tesla)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 temporarily using vehicle odometry to derive motion.
1:14:49Phil (Tesla)Last, these spatial temporal features go through a set of deconvolution to produce the final occupancy and occupancy flow output. They're formed as fixed size voxel grid which might not be precise enough for planning and control in order to get a higher resolution.
1:15:06Phil (Tesla)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.
1:15:22Phil (Tesla)Here we have an articulated bus parked on right side road highlighted as an L shaped voxel. Here as we approach the bus start to move, the front of the cart turns blue first indicating the model predicts the front bus has a non zero occupancy flow and 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.
1:15:52Phil (Tesla)Well, this is a very complicated problem for traditional object detection network as you have to see whether I'm going to use one cuboid or perhaps two to fit the curvature.
1:16:02Phil (Tesla)But for occupancy network, since all we care about is the occupancy in the visible space and we'll be able to model the curvature precisely. Besides the voxel gray, the occupancy network also produces a driver surface.
1:16:18Phil (Tesla)The drywall 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.
1:16:31Phil (Tesla)Instead the voxel grid actually aligns with the surface implicitly. Here we are at a hillcrest where you can see the 3D geometry of the surface being predicted nicely. Planar can use this information to decide perhaps we need to slow down more for the hill crest and as you can also see the voxel gray aligns with the surface consistently.
1:16:58Phil (Tesla)Besides the voxels and the surface, we're also very excited about the recent breakthrough in neural radiance field or lerf. We're looking into both incorporate some of the lightslrf features in into occupancy network training as well as using our network output as the input state for nerf.
1:17:19Phil (Tesla)As a matter of fact, Ashok is very excited about this. This has been his personal weekend project for a while.
1:17:29Ashok ElluswamyThese nerves, because I think academia is building all of these foundation models for language using like tons of large data sets for language. But I think for vision nerves are going to provide the foundation models for computer vision because they are grounded in geometry.
1:17:45Ashok ElluswamyAnd geometry gives us a nice way to supervise these networks and frees us of the requirement to define an ontology. And the supervision is essentially free because you just have to differentially render these images.
1:17:56Ashok ElluswamySo 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 differentiable rendered into any image that was observed, I personally think is the future of computer vision.
1:18:15Ashok ElluswamyAnd 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 full volumetric occupancy will be the future.
1:18:29Ashok ElluswamyThat's my personal bit.
1:18:32Elon MuskThanks, Rachel.
1:18:34Phil (Tesla)So here's an example. Early result of a 3D reconstruction from our fleet data. Instead of focusing on getting perfect RGB reprojection in imaging space, our primary goal here is to accurately represent the world in 3D space for driving.
1:18:49Phil (Tesla)And we want to do this for all our fleet data all over the world in all weather and lighting conditions. And obviously this is a very challenging problem and, and we're looking for you guys to help.
1:19:00Phil (Tesla)Finally, the occupancy network is trained with large auto labeled data set without any human in the loop. And with that, I'll pass to Tim to talk about what it takes to train this network.
1:19:12Tesla PresenterThanks, Phil.
1:19:18Tesla PresenterAll right, hey everyone, let's talk about some training infrastructure. So we've seen a couple videos, four or five I think, and care more and worry more about a lot more clips than that.
1:19:32Tesla PresenterSo 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 will take one hour, but if you have one GPU, it would take 100,000 hours.
1:19:52Tesla PresenterSo 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.
1:20:00Tesla PresenterSo you need 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 run 4,000 GPUs for auto labeling.
1:20:18Tesla PresenterAll 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.
1:20:34Tesla PresenterSo 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.
1:20:49Tesla PresenterSo that is, that's a lot of calls. We are 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.
1:21:04Tesla PresenterSo you cannot just buy, you know, 14,000 GPUs and then 30 petabytes of flash NVMe and you still put it together and let's go train. It actually takes a lot of work.
1:21:14Tesla PresenterAnd 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.
1:21:26Tesla PresenterAnd 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.
1:21:37Tesla PresenterAnd 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.
1:21:46Tesla PresenterThese 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.
1:21:58Tesla PresenterBut 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.
1:22:10Tesla PresenterWe'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.
1:22:27Tesla PresenterVideo 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 and optimize every single piece of that.
1:22:39Tesla PresenterBecause 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.
1:22:53Tesla PresenterWe load those in, including the frames that they depend on. So these are your iframes or your key keyframes. We package those up, move them into shared memory, move them into a double buffer on the GPU and then use the hardware decoder that's only accelerated to actually decode the video.
1:23:09Tesla PresenterSo we do that on the GPU natively. And this is all in a very nice Python Pytorch extension. Doing so unlocked more than 30% training speed increase for the occupancy networks and freed up basically the whole CPU to do any other thing.
1:23:26Tesla PresenterYou 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.
1:23:49Tesla PresenterBecause 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.
1:24:01Tesla PresenterWe 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 these are your is your table that you want to store.
1:24:13Tesla PresenterThen 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.
1:24:26Tesla PresenterThen 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.
1:24:37Tesla PresenterThen you check out which one was most optimal and then store that one. This is actually a huge step. If you do feature caching unintelligible output from the machine learning network, rotate around the dimensions a little bit, you can
1:24:49Pete Bannonget up to 20% increase in efficiency of storage.
1:24:53Tesla PresenterThen 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 the read on more values which you'll use later so that you don't need to do another file system operation.
1:25:12Tesla PresenterSo I could go on and on, I just went on touched on two projects that we have internally. But this is actually part of a huge continuous effort to optimize the compute that we have in house.
1:25:24Tesla PresenterSo accumulating and aggregating through all these optimizations, we now train our occupancy networks twice as fast, just because it's twice as efficient. And and now if we add in bunch more compute and go parallel, we can now train this in hours instead of days.
1:25:38Tesla PresenterAnd with that, I'd like to hand it off to the biggest user of compute. John.
1:25:50John EmmonsHi 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.
1:26:04John EmmonsIn the early days of Autopilot, we modeled the lane detection problem as an image based instance segmentation task. Our network was super simple though. In fact, it was only capable of printing lanes from a few different kinds of geometries.
1:26:18John EmmonsSpecifically, it would segment the eagle 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.
1:26:31John EmmonsBut 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.
1:26:41John EmmonsWhen 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.
1:26:54John EmmonsAnd 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 connectivities between these lanes.
1:27:05John EmmonsSo 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.
1:27:23John EmmonsWe then enhance this visual representation with a coarse road road level map data, which we encode with a set of additional neural network layers that we call the lane guidance module.
1:27:35John EmmonsThis 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.
1:27:46John EmmonsThe 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 smart set of lanes and their connectivities.
1:27:57John EmmonsWe approach this problem like an image captioning task where the input is this dense tensor and the output text. It's predicted into a special language that we developed at Tesla for encoding lanes in their connectivity.
1:28:09John EmmonsIn this language of lanes, the words and tokens are the lane positions in 3D space in the ordering of the tokens. Interdicted modifiers in the tokens encode the connective relationships between these lanes.
1:28:21John EmmonsBy modeling the task as a language problem, we can capitalize on recent auto regressive architectures and techniques from the language community for handling the multiple dality of the problem.
1:28:30John EmmonsWe're not just solving the computer vision problem at Autopilot, we're also applying the state of the art and language modeling to machine learning more generally. I'm now going to dive into a little bit more detail this language component.
1:28:42John EmmonsWhat I have depicted on the screen here is a satellite image which sort of represents the local area around the vehicle. The set of nodes and edges is what we refer to as the lane graph.
1:28:52John EmmonsAnd 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.
1:29:03John EmmonsThis green dot's position is encoded as an index into a coarse grid which discretizes the 3D world. Now, we don't predict this index directly because it would be too computationally expensive to do so.
1:29:13John EmmonsThere'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.
1:29:22John EmmonsFirst, we predict a 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.
1:29:34John EmmonsNow we know where the position of this token is, but we don't know its type. In this case, though, it's the beginning of a new lane. So we predict it as a start token.
1:29:44John EmmonsAnd because it's a start token, there's no additional attributes in our language. We then take the Predictions from this first forward path, and we encode them using a learned conditional embedding, which produces a set of tensors that we combine together, which is actually the first word in our language of lanes.
1:30:00John EmmonsWe 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.
1:30:12John EmmonsIt'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.
1:30:23John EmmonsWe 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.
1:30:34John EmmonsWe 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.
1:30:44John EmmonsNow 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 the fork originates.
1:31:00John EmmonsSo you can see here the fork point predictor is actually the index zero. So it's actually referencing back to tokens that it's 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.
1:31:18Ashok ElluswamyYeah, 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.
1:31:25Ashok ElluswamyHere with neural networks though, is that we tried simpler 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.
1:31:40Ashok ElluswamyA 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.
1:31:51John EmmonsYeah, and the problems don't end there. I mean, you need to predict these connective, like these connective lanes inside of intersections, which it's just not possible with the approach that Ashok's mentioning, which is why we had to upgrade to this sort of.
1:32:01Ashok ElluswamyYeah. When it like overlaps like this segmentation would just go haywire. But even if you try very hard to, you Know, put them on separate layers. It's just a really hard problem.
1:32:09Ashok ElluswamyBut language just offers a really nice framework for getting a sample from a posterior as opposed to, you know, trying to do all of this in post processing. But this doesn't actually stop for just autopilot, right, John?
1:32:21Ashok ElluswamyThis can be used for Optimus.
1:32:23John EmmonsYeah. You know, I guess they wouldn't be called lanes, but you could imagine, you know, sort of in this, you know, stage here that you might have sort of paths that sort of, you know, encode the possible places that people could walk.
1:32:33Elon MuskYeah.
1:32:34Ashok ElluswamyBasically, 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.
1:32:44Ashok ElluswamyAnd then we predict a set of pathways that would go through the aisles. Take the robot and say, ok, 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.
1:33:00John EmmonsAll right, so ultimately, what we get from this lane detection network is a set of lanes and their connectivities, which comes directly from the network. There's no additional step here for sparsifying these, you know, dense predictions into sparse ones.
1:33:12John EmmonsThis 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, future paths and other semantics on objects.
1:33:26John EmmonsSo 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.
1:33:34John EmmonsWhat 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 braking and steering action is required to avoid a collision.
1:33:48John EmmonsIn 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.
1:33:57John EmmonsBut 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.
1:34:06John EmmonsBut 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.
1:34:19John EmmonsPredicting 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.
1:34:30John EmmonsEvery millisecond really matters here. To minimize the inference latency, our neural network is split into two phases. In the first phase, we identified locations in 3D space where agents exist.
1:34:42John EmmonsIn 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, you know, do the rest of the processing.
1:34:51John EmmonsThis sparsification 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.
1:35:06John EmmonsIt 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.
1:35:15John EmmonsThank you.
1:35:24Sri (Tesla)Hi everyone, I'm sri. Today I'm going to give a glimpse of what it takes to run these FSC networks in the car and how do we optimize for the inference latency.
1:35:34Sri (Tesla)Today I'm going to focus just on FSG lanes network that John just talked about. So when we started this track, we wanted to know if we can run this FSG lanes network natively on the TRIP engine, which is our in house neural network accelerator that we built in the FSD computer.
1:35:53Sri (Tesla)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 auto regressive and iterative where it crunches through multiple attention blocks in the inner loop producing sparse points directly at every step.
1:36:14Sri (Tesla)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.
1:36:25Sri (Tesla)So the network predicts the heat map of most probable spatial locations of the point. Now we do a ARGMAX and a one hot operation which gives the one hot encoding of the index of the spatial location.
1:36:41Sri (Tesla)Now we need to select the embedding associated with this index from an embedding table that is learned during training. 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.
1:37:02Sri (Tesla)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.
1:37:12Sri (Tesla)Again, we pulled 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.
1:37:31Sri (Tesla)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 intake accuracy as well as to optimize performance.
1:37:45Sri (Tesla)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.
1:37:59Sri (Tesla)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 1,000 neural network signals.
1:38:14Sri (Tesla)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.
1:38:32Sri (Tesla)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 sub graphs and compiles each of those sub graphs natively for the inference devices.
1:38:49Sri (Tesla)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 bandwidth constraints so that it comes with an optimized schedule that gets executed in the car on the runtime.
1:39:11Sri (Tesla)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.
1:39:24Sri (Tesla)To get 100 tops of compute utilization, we need to optimize across all 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.
1:39:40Sri (Tesla)And in fact going even deeper to understanding and optimizing the cache coherent and non coherent data parts 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.
1:39:59Sri (Tesla)And this is just the, this is 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.
1:40:17Sri (Tesla)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.
1:40:36Tesla PresenterThank you sir. Hi everyone, I'm Yeagan Zhang and I'm leading 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.
1:40:54Tesla PresenterBut today I'd like to focus on the awesome lanesnet here. So to successfully train and generalize this network to everywhere we think we went tens of millions of trips from probably 1 million intersection or even more.
1:41:12Tesla PresenterSo then 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.
1:41:28Tesla PresenterHowever, combining 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.
1:41:50Tesla PresenterBut still we want an even better auto labeling machine that can provide us 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.
1:42:15Tesla PresenterSo 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 structure recovery by multi camera visual inertial odometry.
1:42:34Tesla PresenterSo 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 drift rate of this trajectory in car is like 1.3 centimeter per meter and 0.45 milli region per meter, which is pretty decent considering its compact compute requirement.
1:42:57Tesla PresenterThen the recovery service 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 pre processed trajectories and structures along with the trip data.
1:43:16Tesla PresenterThe second step is multi tool 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 vehicle, not the same vehicle.
1:43:32Tesla PresenterSo this is done by multiple Internal steps like coarse alignment, pairwise matching, joint optimization, then further surface refinement. In the end, the human analyst comes in and finalizes the label.
1:43:46Tesla PresenterSo each habit 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.
1:44:01Tesla PresenterSo 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.
1:44:15Tesla PresenterThat'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.
1:44:31Tesla PresenterThis 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.
1:44:43Tesla PresenterSo 53 from different vehicles. So this is how we capture and transform the space time slices of the world into the network.
1:44:53Ashok ElluswamySupervision. Yeah. One thing I'd like to note is that Yeagen just talked about how we auto label our lens. But we have auto labels for almost every task that we do, including our planner.
1:45:04Ashok ElluswamyAnd many of these are fully automatic. There's no humans involved. For example, for objects, all of the kinematics, the shapes, their futures, everything just comes from auto labeling. And the same is true for occupancy too.
1:45:15Ashok ElluswamyAnd we have really just built a machine around this.
1:45:17Tesla PresenterYeah. So if you can go back one,
1:45:19Pete Bannonslide
1:45:21Tesla Presenterone 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.
1:45:37Tesla PresenterAnd then someone makes a script, it starts to work, it starts working better until we 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.
1:45:51Tesla PresenterWhat 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.
1:46:02Tesla PresenterAnd 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 100,000 clips per day or 100,000 items.
1:46:12Tesla PresenterThat seems good. And so the engineer said, well, we can do a bit of postgres 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.
1:46:28Tesla PresenterAgain, 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.
1:46:39Ashok ElluswamyYeah, it really is like a factory that produces labels and like production lines yield quality, inventory. Like all of the same concepts apply to this label factory. That applies for, you know, the factory for our cars.
1:46:52Tesla PresenterThat'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.
1:47:08Tesla PresenterSo 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 raindrops on camera lenses.
1:47:22Tesla PresenterThese 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.
1:47:34Tesla PresenterSo 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.
1:47:47David (Tesla)Thank you, Yagan. 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.
1:47:58David (Tesla)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.
1:48:16David (Tesla)However, I'm going to talk about using Yagen'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.
1:48:27David (Tesla)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.
1:48:43David (Tesla)Starting with road boundary labels, we can generate a solid road mesh and retopologize it with the lane graph labels. This helps inform important road details like crossroad slope and detailed material blending.
1:48:57David (Tesla)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 spawn island geometry and populate it with randomized foliage.
1:49:13David (Tesla)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 curbs, while trees can drop leaves below it, obscuring lines or edges.
1:49:33David (Tesla)Next, we can bring in map data to inform positions of things like 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.
1:49:48David (Tesla)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.
1:50:06David (Tesla)And again, this is all automatic. No artists 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 of the single ground truth.
1:50:23David (Tesla)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.
1:50:43David (Tesla)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.
1:50:59David (Tesla)We can start to procedurally alter our lane graph inside the simulator to help focus to create entirely new flows through this intersection to help focus the network's attention to the road markings to create more accurate predictions.
1:51:13David (Tesla)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.
1:51:28David (Tesla)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 into geohash tiles about 150 meters square in size.
1:51:43David (Tesla)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.
1:51:56David (Tesla)Then using a tile loader tool, we can summon any number of those cache tiles using a geohash ID. Currently we're doing about these 5 by 5 tiles or 3 by 3, usually centered around fleet hotspots or interesting lane graph locations.
1:52:11David (Tesla)And the tile Loader also converts these tile sets into uassets for consumption by the Unreal Engine and gives you a finished product from what you saw on the first slide.
1:52:23David (Tesla)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.
1:52:32David (Tesla)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.
1:52:44David (Tesla)This allows us to throw compute at it and regenerate all these tile sets overnight. This ensures all environments are of consistent quality and features, which is super important for training since new ontologies and signals are constantly released.
1:53:03David (Tesla)And now to come full circle, because we generated all these tile sets from ground truth data that contain all the weird intricacies from the real world. And we can combine that with the procedural, visual and traffic variety to create limitless targeted data for the network to learn from.
1:53:20David (Tesla)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.
1:53:36Kate Park (Tesla)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.
1:53:46Kate Park (Tesla)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.
1:54:04Kate Park (Tesla)In reality, there's nobody in the car, it's just awkwardly parked. We built this tooling to identify the mispredictions, correct the label, and categorize this clip into an evaluation set.
1:54:17Kate Park (Tesla)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.
1:54:34Kate Park (Tesla)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 infrastructure, request data, and use the tooling shown previously to correct the labels.
1:54:50Kate Park (Tesla)By surgically targeting the mispredictions of the current model, we're only adding the most valuable examples to our training set. We surgically fixed 13,900 clips, and because those were examples where the current model struggles, we don't even need to change the model architecture.
1:55:09Kate Park (Tesla)A simple weight update with this new valuable Data is enough to solve the chall challenge case. So you see, we no longer predict that crossing vehicle as stopped as shown in orange, but parked as shown in red.
1:55:21Kate Park (Tesla)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.
1:55:37Kate Park (Tesla)We just showed you the data engine loop for one challenge case, namely these park cars at turns. But there are many challenge cases, even for one signal of vehicle movement.
1:55:47Kate Park (Tesla)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.
1:56:01Kate Park (Tesla)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.
1:56:12Kate Park (Tesla)This data engine framework applies to all our signals, whether they're 3D multicam video, whether the data is human labeled, auto labeled or simulated, whether it's an offline model or an online model.
1:56:26Kate Park (Tesla)And Tesla's able to do this at scale because of the fleet advantage, the infra that our ENG team has built, the labeling resources that feed our networks. To train on all this data, we need a massive amount of compute.
1:56:39Kate Park (Tesla)So I'll hand it off to Pete and Ganesh to talk about the Dojo supercomputing platform. Thank you.
1:56:49Tesla PresenterThank you, Katie.
1:56:55Pete BannonThanks 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.
1:57:04Tesla PresenterAnd my name is Ganesh Venkat.
1:57:06Elon MuskI run the Doja program.
1:57:14Tesla PresenterThank you.
1:57:15Pete BannonI'm frequently asked, why is a car company building a supercomputer for training? And this question fundamentally misunderstands the the nature of Tesla. At its heart, Tesla is a hardcore technology company.
1:57:31Pete BannonAll 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.
1:57:50Phil (Tesla)World.
1:57:50Pete BannonIt'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.
1:58:04Pete BannonBut 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.
1:58:17Pete BannonSome 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, you know, a 30x speed up would be really nice if we could provide it at a cost competitive and energy competitive way to do that.
1:58:35Pete BannonWe wanted to build a chip with a lot of arithmetic units that, 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.
1:58:52Pete BannonAnd 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.
1:59:03Pete BannonSRAM provides unfortunately, a modest amount of capacity, but extremely, extremely high bandwidth and very low latency. And that enables us to achieve high utilization with the arithmetic units. Those choices.
1:59:17Pete BannonThat 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.
1:59:24Pete BannonWe didn't have space, so no virtual memory. 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.
1:59:41Pete BannonSo 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 capacity, which we obviously don't have.
2:00:01Pete BannonSo 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.
2:00:14Pete BannonSo 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 you know, pretty much if your model was 2 too big for the computer, you just have to go buy a bigger computer.
2:00:29Pete BannonThat's what we were looking for today. The way package machines are packaged, there's a pretty fixed ratio of, for example, GPUs, 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 needs of the autopilot.
2:00:53Tesla PresenterTeam. Yeah, and it's so true, Pete, like
2:00:56Elon Muskno limits philosophy was our guiding star all the way.
2:01:00Felix (Tesla)All of our choices were centered around
2:01:03Tesla Presenterthat and to the point that we
2:01:06Elon Muskdidn't want traditional data center infrastructure to
2:01:09Felix (Tesla)limit our capacity to execute these programs at speed.
2:01:14Paril JainSo that's why we. That's why we.
2:01:18Elon MuskSorry about that. That's why we integrated vertically our data
2:01:24Felix (Tesla)center, the entire data center.
2:01:26Elon MuskBy doing a vertical integration of the data center, we could extract new levels of efficiency. We could optimize power delivery, cooling, and
2:01:37Felix (Tesla)as well as system management across the
2:01:40Elon Muskwhole data center stack.
2:01:42Paril JainRight.
2:01:42Elon MuskRather than doing box by box and integrating those boxes into data centers.
2:01:49Tesla PresenterAnd to do this, we also wanted
2:01:53Elon Muskto 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.
2:02:06Elon MuskAnd 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
2:02:21Felix (Tesla)as go over some of our cool results.
2:02:25Elon MuskGreat.
2:02:31Tesla PresenterThanks, 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.
2:02:49Tesla PresenterNow, 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.
2:03:08Tesla PresenterNow, 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 system.
2:03:23Tesla PresenterNow, silicon technology has, has used this, has done this for decades. Chips had followed Moore's law for density integration to get performance scaling. Now, a key step in realizing that vision was our training tile.
2:03:41Tesla PresenterNot only 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.
2:04:03Tesla PresenterAnd 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.
2:04:19Tesla PresenterBut 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.
2:04:38Tesla PresenterAnd 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 1amp per millimeter squared.
2:04:54Tesla PresenterAnd 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.
2:05:12Tesla PresenterNow 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.
2:05:27Tesla PresenterSo we started this effort by working with vendors to deliver 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 and software development, and also to find the optimal design and stack up that would meet our final production goals.
2:05:52Tesla PresenterAnd in the end, we were able to reduce CTE over 50% and meet our performance by by 3x over our initial version. Now, needless to say, finding this optimal material stackup while maximizing performance at density is extremely difficult.
2:06:14Tesla PresenterNow, 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.
2:06:29Tesla PresenterAnd 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.
2:06:45Tesla PresenterNow 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.
2:07:00Tesla PresenterNow, 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.
2:07:14Tesla PresenterNow that includes our clock, our 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 on nearby capacitors.
2:07:39Tesla PresenterNow, singing caps are not a new phenomenon and 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.
2:07:52Tesla PresenterBut 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.
2:08:05Tesla PresenterIt 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.
2:08:22Tesla PresenterWe 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.
2:08:40Tesla PresenterNow, 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 re examine every aspect of the data center infrastructure in order to support our unprecedented power and cooling density.
2:08:59Tesla PresenterWe 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.
2:09:13Tesla PresenterAnd 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.
2:09:31Tesla PresenterAnd earlier this year we started load 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.
2:09:46Tesla PresenterNow, last year we introduced only a couple of components of our system. The custom D1 die and the training tile. But we tease the exit pod as our end goal.
2:09:57Tesla PresenterWe'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.
2:10:09Tesla PresenterIt enables us to seamlessly connect tiles together, not only within the cabinet but between cabinets. We can connect these tiles at very tight space spacing across the entire accelerator.
2:10:23Tesla PresenterAnd 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.
2:10:36Tesla PresenterIt's 75 millimeters in height and supports six tiles at 135 kilograms. This is the equivalent of three to four fully loaded high performance racks. Next, we need to feed data to the training tiles.
2:10:55Tesla PresenterThis 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 create communicate across our entire accelerator.
2:11:15Tesla PresenterIt 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.
2:11:27Tesla PresenterAnd lastly, we can connect to it through a standard gen 4 PCIe interface. Now we pair 20 of these cards per tray and that gives us 640 gigabytes of high bandwidth DRAM.
2:11:43Tesla PresenterAnd this provides our disaggregated memory layer for our training tiles. These cards are our high bandwidth ingest path both through PCIe and Ethernet. They also provide a high Radex Z connectivity path that allows shortcuts across our large dojo accelerator.
2:12:04Tesla PresenterNow 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.
2:12:23Tesla PresenterAnd 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 3 phase 480 volt AC power to 52 volt DC power.
2:12:52Tesla PresenterNow, 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.
2:13:17Tesla PresenterAnd altogether we can house two full accelerators in our exit pod for a combined one. One exaflop of ML Compute. Now altogether. This amount of technology and integration has only ever been done a couple of times in the history of compute.
2:13:35Tesla PresenterNext we'll see how software can leverage this to accelerate their performance.
2:13:48Eric (Tesla)Thanks, Bill. My name is Rajeev 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.
2:14:01Eric (Tesla)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 seen how the hardware provides peak performance.
2:14:16Eric (Tesla)It's 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.
2:14:29Eric (Tesla)So let's talk about why communication bound models are difficult to scale. But before that, let's look at why Resnet50 like models are easier to scale. We start off with a single accelerator, run the forward and backward Passes followed by the optimizer.
2:14:44Eric (Tesla)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 pipelined with the backward pass.
2:15:00Eric (Tesla)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.
2:15:15Eric (Tesla)So to get around this, researchers typically run the 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.
2:15:29Eric (Tesla)And while there are ways to get around this, they usually involve tedious manual will 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.
2:15:44Eric (Tesla)And even with significant efforts from our ML engineers, we see such models don't scale linearly. The Dojo system was built to make such models work at high utilization. The high density integration is 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 reduce or a parameter all gather.
2:16:12Eric (Tesla)A slice of the Dojo mesh can be carved out to run any model. The only thing you just need to do is to make the slice large enough to fit a batch room surface for their particular model.
2:16:25Eric (Tesla)After that, the partition presents itself as one large accelerator, freeing the users from having to worry about the internal details of execution, and it's the job of the compiler to maintain this abstraction.
2:16:38Eric (Tesla)Fine grained synchronization primitives and 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 layers execution.
2:16:52Eric (Tesla)We depend on the high Dojo 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,
2:17:07Tesla Presenterwe
2:17:07Eric (Tesla)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 I out of the box.
2:17:17Eric (Tesla)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.
2:17:36Eric (Tesla)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 acid test of a compiler.
2:17:48Eric (Tesla)And the underlying hardware is executing a cross type 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.
2:18:07Elon MuskParallel.
2:18:09Eric (Tesla)So what would an ideal batch form look like on 25 dojo dice? Let's say the previous list activations are already split across dice. We would expect the 350 nodes on each die to coordinate and produce die local mean and standard deviation values.
2:18:26Eric (Tesla)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.
2:18:38Eric (Tesla)Let's see how the compiler actually executes a real batch drum operation across 25 dice. The communication trees were extracted from the compiler and the timing is from a real hardware run.
2:18:49Eric (Tesla)We're about to see 8,750 nodes on 25 dies coordinating to reduce and then broadcast the Bastrum mean and standard deviation values die local reduction followed by global reduction towards the middle of the tie.
2:19:05Eric (Tesla)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 dice. The same operation takes 150 microseconds on 24 GPUs.
2:19:24Eric (Tesla)This is an orders of magnitude improvement over GPUs. And while we talked about an all reduced operation in the context of a batch norm, it's important to reiterate that the same advantages apply to all other communication primitives.
2:19:38Eric (Tesla)And these primitives are essential for large scale training. So how about full model performance? So while we think that Resnet50 is not a good representation of real world Tesla workloads, it is a standard benchmark.
2:19:51Eric (Tesla)So let's start there. We are already able to match the A100 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 eight per die.
2:20:04Eric (Tesla)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 the two models stood out.
2:20:16Eric (Tesla)The auto labeling networks, a class of offline models that are used to generate ground truth, and the occupancy networks that you heard about the Auto labeling networks are large models that have high arithmetic intensity, while the occupancy networks can be ingest bound.
2:20:31Eric (Tesla)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 did we do on these two networks?
2:20:44Eric (Tesla)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 auto labeling network, we're already able to surpass the performance of an A100 with our current hardware running on our older generation VRMs.
2:21:01Eric (Tesla)On our production hardware with our new VRMs, that translates to doubling the throughput of 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.
2:21:15Eric (Tesla)We see even bigger leaps on the occupancy network, almost 3x with our production hardware with room for more.
2:21:33Eric (Tesla)So what does that mean for Tesla? With the 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.
2:21:55Eric (Tesla)And this Dojo tile costs less than 1 of these GPU boxes.
2:22:04Eric (Tesla)What it really means is that networks that took more than a month to train now take less than a week. Alas, when we measured things, it did not turn out so well.
2:22:16Eric (Tesla)At 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.
2:22:29Eric (Tesla)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 host.
2:22:41Eric (Tesla)The data loaders 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.
2:22:54Eric (Tesla)The Tesla transport protocol moves data seamlessly across host tiles and ingest processors. So we extended the Tesla transport protocol to work over Ethernet. We then built the Dojo network interface card, the dnic, to leverage TTP over Ethernet.
2:23:09Eric (Tesla)This allows any host with a DNIC card to be able to DMA to and from other T TTP endpoints. So we started with the Dojo mesh. Then we added a tier of data loading hosts equipped with the DNIC card.
2:23:25Eric (Tesla)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.
2:23:39Eric (Tesla)After these optimization, our occupancy went from 4% to 97%. So the data loading sections have reduced,
2:23:53Eric (Tesla)The data loading sections have reduced drastically and the ML hardware is kept busy. We actually expect this number to go to 100% pretty soon after these changes went in.
2:24:03Eric (Tesla)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.
2:24:17Eric (Tesla)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.
2:24:28Eric (Tesla)Our high arithmetic intensity auto labeling network today that occupies 4,000 GPUs, over 72 GPU racks. With our dense compute and our high performance, we expect to provide the same throughput with just four dojo cabinets.
2:24:54Eric (Tesla)And these four dojo cabinets will be part of our first exopod that we plan to build by quarter one of 2023. This will more than double Tesla's auto labeling capacity.
2:25:10Eric (Tesla)The first exapot is part of a total of seven exit parts that we plan to build in Palo Alto. Right here across the wall. And we have a display cabinet from one of these exapods here for everyone to look at.
2:25:27Eric (Tesla)Six 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.
2:25:51Eric (Tesla)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.comai thank you.
2:26:27Elon MuskAll right, so hopefully that was enough detail and now we can move to questions. And guys, I think the team come out on stage and 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.
2:27:04Elon MuskA lot of people think we're like just a car company or we make cool cars, whatever, but they don't have. 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 first, the most radical computer architecture since the crayon supercomputer computer.
2:27:32Elon MuskAnd I Think if you're interested in developing some, 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.
2:27:43Elon MuskSo, yeah, let's fire away with some questions. I think there's, there's a mic at the front and a mic at the back.
2:27:56Kate Park (Tesla)On this side.
2:27:57Elon MuskJust throw mics at people, jump. All for the mic.
2:28:02Tesla PresenterYeah. Hi.
2:28:04Elon MuskThank you very much.
2:28:05Ashok ElluswamyI was impressed here.
2:28:08Elon MuskYeah, I was impressed very much by Optimus.
2:28:11Ashok ElluswamyBut I wonder why tendon driven the hand?
2:28:14Elon MuskWhy did you choose a tendon driven approach for the hand?
2:28:18Tesla PresenterBecause tendons are not very durable.
2:28:21Elon MuskAnd why spring loading?
2:28:25Sri (Tesla)Yeah.
2:28:26Tesla PresenterLou, is this cool?
2:28:27Felix (Tesla)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.
2:28:38Elon MuskJust keep the mic close to your
2:28:39Tesla Presentermouth, a little bit closer.
2:28:40Elon MuskYeah.
2:28:40Tesla PresenterHear me. Cool.
2:28:42Eric (Tesla)Yeah.
2:28:43Felix (Tesla)So, 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.
2:28:55Felix (Tesla)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.
2:29:09Felix (Tesla)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, stuttery motion in your fingers.
2:29:26Felix (Tesla)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, to have the tendon drive them close and then the springs passively extend.
2:29:43Felix (Tesla)And this is something that's seen in our hands as well.
2:29:45Tesla PresenterRight.
2:29:46Felix (Tesla)We have the ability to actively flex and then we also have the ability to extend.
2:29:51Tesla PresenterYeah.
2:29:52Elon MuskOur 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 probably not barking up the right tree on all the technical solutions.
2:30:09Elon MuskAnd 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.
2:30:29Elon MuskWe're trying to follow the. 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?
2:30:44Elon MuskBecause you have to have a. You're going 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.
2:31:01Elon MuskBut for sure there'll be hand version 2, version 3, and we may change the architecture quite significantly over time.
2:31:12Lizzie MiskovetzHi. The Optimus robot is really impressive. You did a great job. Bipedal robots are really difficult. But what I noticed is, might be missing from your plan is to acknowledge the utility of the human spirit.
2:31:30Lizzie MiskovetzAnd I'm wondering if Optimus will ever get a personality and be able to laugh at our jokes while they. While it folds our clothes.
2:31:38Elon MuskYeah, absolutely. I think we want to have really fun versions of Optimus and so that Optimus can both to be utilitarian and do tasks, but can also be kind of like a friend and a buddy and hang out with you.
2:31:59Elon MuskAnd I'm sure people will think of all sorts of creative uses for this robot. And, you know, 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.
2:32:21Elon MuskI mean, you can make the robot
2:32:22Tesla Presenterlook,
2:32:25Elon Muskyou can skin the robot in many different ways and I'm sure people will find very interesting ways to. Yeah. Versions of Optimus. So.
2:32:40Elon MuskThanks for the great presentation.
2:32:42Felix (Tesla)I wanted to know if there was an equivalent to interventions in Optimus.
2:32:46John EmmonsIt seems like labeling through moments where
2:32:49Elon Muskhumans disagree with what's going on is important. And in a humanoid robot, that might
2:32:54John Emmonsbe also a desirable source of information.
2:33:05Ashok ElluswamyYeah, 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.
2:33:19Ashok ElluswamyIf it's going to hit something, we can just like hold it and then it will stop. It won't like, you know, crush your hand or something. And those are all intervention data.
2:33:27Elon MuskYeah.
2:33:28Ashok ElluswamyAnd we can learn a lot from our simulation systems too, where we can check for collisions and supervise that those are bad actions.
2:33:35Elon MuskYeah, I mean, so Optimus, we want 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.
2:33:47Elon MuskBut obviously we could program the robot to be less robot like and more friendly. And, you know, it can obviously learn to emulate humans and feel very natural. So as AI in general improves, we can add that to the robot and it should be obviously able to do simple instructions or even intuit what it is that you want.
2:34:13Elon MuskSo you could give it a high level instruction and then it can break that down into a series of actions and take those actions.
2:34:25Tesla PresenterHi. Yeah, it's exciting to think that with the Optimus you will think that you
2:34:30Elon Muskcan achieve orders of magnitude of improvement in economic output. That's really exciting.
2:34:38Tesla PresenterAnd when Tesla started, the mission was
2:34:40Elon Muskto accelerate the advent of renewable energy or sustainable transport. So with the Optimus, do you still
2:34:47Tesla Presentersee that mission being the mission statement of Tesla or is it going to
2:34:53Elon Muskbe updated with, you know, mission to accelerate the advent of, I don't know,
2:34:57Tesla Presenterinfinite abundance or limitless economy?
2:35:03Elon MuskYeah, I mean, it is not strictly speaking, Optimus is not strictly speaking directly in line with accelerating sustainable energy
2:35:17Tesla Presenterto
2:35:17Elon Muskthe 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, I don't know, making the future awesome.
2:35:35Elon MuskSo 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,
2:35:51Sri (Tesla)are
2:35:51Elon Muskyou, do you want to see what it's like in a year, 2 years, 3 years, 4 years, 5 years, 10? I'd say for sure. You definitely want to see what's happening with Optimus, whereas, you know, a bunch of other technologies are, you know, sort of plateaued.
2:36:09Elon MuskDon't name names here, but.
2:36:15Elon MuskYou know, so I think Optimus is going to be incredible in like 5 years, 10 years, like mind blowing. And I'm really interested to see that happen and I hope you are too.
2:36:30Elon MuskI have a quick question here.
2:36:32Eric (Tesla)I'm Justin and I was wondering, are you planning to extend conversational capabilities for the robot? And my second follow up question to
2:36:43John Emmonsthat is, what's the end goal?
2:36:45Elon MuskWhat'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.
2:37:00Elon MuskSo 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 someplace interesting for sure. You know, we always have to be careful about the, you know, don't go down the Terminator path.
2:37:21Elon MuskThat's, you know, I thought we might, maybe we should start off with a video of like the Terminator starting off with this, you know, skull crushing. But that might be kind of.
2:37:30Elon MuskPeople might take that 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 control ROM that you can't update over the Internet, which I think that's quite important, essential frankly.
2:37:59Elon MuskSo like a localized stop button
2:38:04David (Tesla)or
2:38:04Elon Muskremote control, something like that, that cannot be changed. But it's definitely going to be interesting. It won't be boring.
2:38:22Elon MuskOkay.
2:38:22Tesla PresenterYeah.
2:38:23Eric (Tesla)I see you today.
2:38:24Tesla PresenterYou have very attractive product with Dojo and its applications. So I'm wondering what's the future for Dojo platform? Will you like provide like infrastructure and service like aws or you will like sell the chip like the Nvidia?
2:38:39Tesla PresenterSo basically what the future because I say you use 7nm, so the developer
2:38:44Phil (Tesla)cost is like easily over US$10 million. How do you make the business like business wise?
2:38:51Elon MuskYeah, I mean Dojo is a very big computer and actually will use a lot of power and needs a lot of cooling. So I think it's probably going to make more sense to have Dojo operate in like a Amazon Web Services manner than to try to sell it to someone else.
2:39:10Elon MuskSo 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.
2:39:26Elon MuskAnd as the world transitions to software 2.0,
2:39:34Ashok Elluswamyand that's on the bingo card
2:39:37Elon Muskas someone I know has to know how to drink five tequilas. 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.
2:40:07Elon MuskSo I think there's a lot of opportunity in that direction.
2:40:14Elon MuskHi, 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 humanity? Robots that understand our emotions and art
2:40:34Phil (Tesla)and can contribute to our creativity.
2:40:40Elon MuskWell, I think there's this. You're already seeing robots that at least are able to generate very interesting art with like, like Dall EE and Dall E2. And I think we'll start seeing AI that can actually generate even movies that have coherence, like interesting movies and tell jokes.
2:41:04Elon MuskSo 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?
2:41:21Ashok ElluswamyYeah, I guess the optimist robot can come up with physical art, not just digital art. You can, you know, you can ask for some dance moves in text or voice, and then you can produce those in the future.
2:41:33Ashok ElluswamySo it's a lot of like physical art, not just digital art.
2:41:38Tesla PresenterOh yeah, yeah.
2:41:38Elon MuskComputers can absolutely make physical art. Yeah, yeah, 100%.
2:41:41John EmmonsYeah.
2:41:42Ashok ElluswamyLike dance, play soccer, or what have you. It needs to get more agile, but over time for sure.
2:41:50David (Tesla)Thanks so much for the presentation.
2:41:53John EmmonsFor 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
2:42:02Elon Muskof that was and how much of
2:42:04David (Tesla)an improvement it gave.
2:42:05Tesla PresenterI thought the.
2:42:05David (Tesla)That was a really interesting, curious choice
2:42:07John Emmonsto use language models for the lane transitioning. So there's sort of two aspects for why we transition to language modeling.
2:42:14Elon MuskSo the first, talk loud and close.
2:42:17Tesla PresenterOkay.
2:42:17Elon MuskIt's not coming through very clearly.
2:42:19John EmmonsOkay, got it.
2:42:21David (Tesla)Yeah.
2:42:21John EmmonsSo 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 Shook mentioned earlier, basically when we predicted lanes and sort of of a dense 3D fashion, you can only model certain kinds of lanes.
2:42:35John EmmonsBut we want to get those crisscrossing 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.
2:42:45John EmmonsAlso, 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, you know, coherent predictions.
2:42:59John EmmonsYou don't want to be predicting two lanes and three lanes at the same time. You want to commit to one in a generative model like these language models provides that.
2:43:10Tesla PresenterHi.
2:43:12Phil (Tesla)Hi, My name is Giovanni.
2:43:15John EmmonsYeah, thanks for the presentation.
2:43:17Elon MuskIt's really nice.
2:43:18Phil (Tesla)I have a question for FST team.
2:43:21Tesla PresenterSo for the neural networks, how do you test? Like, how do you do unit test Software unit tests on that, like do
2:43:31Phil (Tesla)you have like a bunch or I
2:43:33Tesla Presenterdon't know, mid thousands or.
2:43:36Phil (Tesla)Yes, cases where the neural network that after you train it, you have to pass it before you release it to
2:43:44Elon Muskas a product, Right?
2:43:45Tesla PresenterYeah.
2:43:46Phil (Tesla)What's your software unit testing strategies for this?
2:43:49Eric (Tesla)Basically?
2:43:50Ashok ElluswamyYeah, glad you asked. There's like a series of tests that we have defined starting from, you know, unit test for software itself. But then for the neural network models we have VIP sets defined where you know, you can define if you just have a large test set, that's not enough.
2:44:04Ashok ElluswamyWhat we find we need like sophisticated VIP sets for different failure modes and then we curate them and grow them over the time of the product. So over the years we have hundreds of thousands of examples where we have been failing in the past that we have curated.
2:44:20Ashok ElluswamyAnd 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.
2:44:36Ashok ElluswamyAnd there's an extensive QA program. It's very hard to ship a regression. There's like nine levels of filters before it hits customers. But then we have really good infra to make this all efficient.
2:44:49Elon MuskI'm one of the QA testers, so I QA the car. Yeah, like QA tester.
2:44:54Sri (Tesla)Yeah.
2:44:55Elon MuskSo I'm constantly in the car just being qa like whatever the latest alpha bolt is that doesn't totally crash, finds
2:45:04Ashok Elluswamya lot of bugs.
2:45:08Tesla PresenterHi, great event. I 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, right from GPT3 to palm it can actually now do reasoning.
2:45:25Tesla PresenterDo you see that it's essential scaling up foundation models with data and size and then at least you can get a teacher model, right? That potentially can solve all the problems and then you distill to a student model.
2:45:42Tesla PresenterIs that how you see foundational models relevant for autonomous learning?
2:45:46Ashok ElluswamyThat's quite similar to our auto labeling model. 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 the car.
2:45:57Ashok ElluswamySo 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.
2:46:11Ashok ElluswamyWe are building some really, really large Data sets that are multiple petabytes. And we are seeing that some of these tasks work really well when we have these large data sets like the kinematics, like I mentioned video in all the kinematics, out of all the objects and up to the fourth derivative.
2:46:27Ashok ElluswamyAnd 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 data sets and large models.
2:46:41Ashok ElluswamySo we're seeing the equivalent of foundation models in our own way for geometry and kinematics and things like those. You want to add anything, John?
2:46:52John EmmonsYeah, I'll keep it brief. Basically whenever we train on a larger data set we see big.
2:46:58Tesla PresenterOkay.
2:46:59John EmmonsBasically whenever we train on a larger data set we see big improvements in our model performance. And basically whenever we initialize our networks with, you know, some pre training step from some other auxiliary task, we basically see improvements.
2:47:10John EmmonsThe self supervised or supervised with large data sets, both help a lot.
2:47:17Tesla PresenterHi.
2:47:18Elon MuskSo at the beginning Elon said that Tesla is potentially interested in building artificial general intelligence systems. Given the potentially transformative impact of technology
2:47:28John Emmonslike that, it seems prudent to invest
2:47:30Elon Muskin technical AGI safety expertise specifically. I know Tesla does a lot of technical narrow AI safety research.
2:47:39John EmmonsI was curious if Tesla was intending
2:47:41Phil (Tesla)to try to build expertise in technical
2:47:44Elon Muskartificial general intelligence safety specifically. Well, I mean, if it starts looking like we're going to be making a significant contribution to artificial general intelligence, then we'll for sure invest in safety.
2:47:59Elon MuskI'm a 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.
2:48:12Elon MuskSo 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, but I think, I think there should be a referee that is ensuring or doing, trying to ensure public safety for AGI.
2:48:39Elon MuskAnd you think of like, well, what are the elements that are necessary to create AGI? Like the accessible data set 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 data set.
2:49:08Elon MuskIt probably is the biggest data set 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 of cameras in the real world and like I said with audio and other sensors as well.
2:49:29Elon MuskSo 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.
2:49:48Elon MuskHey, I noticed the semi was back
2:49:51John Emmonsthere, but we haven't talked about it too much.
2:49:53Elon MuskI was just wondering for the semi truck, what are the changes you're thinking about from a sensing perspective?
2:49:59David (Tesla)I imagine there's very different requirements obviously than just a car.
2:50:03Eric (Tesla)And if you don't think that's true,
2:50:04Elon Muskwhy is that true? No, I think basically you can drive a car. I mean think about what drives any vehicle. A biological neural net with, with eyes, with cameras essentially.
2:50:18Elon MuskSo if and really what is your primary sensors are two cameras on a slow gimbal, a very slow gimbal, that's, that's your head. So if you know a biological neural net with two cameras on a slow gimbal can drive a semi truck, that then if you've got like eight cameras with continuous 360 degree vision operating at a higher frame rate and much higher reaction rate, then I think it is obvious that you should be able to drive a semi or any, any vehicle much better than a human.
2:50:56Ashok ElluswamyHi, my name is Akshay.
2:50:58Tesla PresenterThank you for the event.
2:51:00Ashok ElluswamyAssuming, you know, Optimus would be used for different use cases and would evolve at different pace for these use cases, would it be possible to sort of develop and deploy different software and hardware components independently and deploy them, you know, in the, in Optimus so that the overall, you know, feature development is faster for Optimus?
2:51:30Elon MuskOkay. All right. We did not comprehend, unfortunately our neural net did not comprehend the question. So next question.
2:51:43Tesla PresenterHi, I want to switch a gear to the autopilot. So when you guys plan to roll out the FSD Beta to countries other than U.S. and Canada. And also my next question is what's the biggest the bottleneck or the technological barrier you think in the current autopilot stack and how you envision to solve that to make the autopilot is considerable better than human in terms of performance matrix like safety assurance and the human confidence.
2:52:11Tesla PresenterAnd 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 explain expand a bit on that?
2:52:23John EmmonsThank you.
2:52:24Elon MuskWell, that's a whole bunch of questions. We're hopeful to be able to, I think from a technical standpoint FSD Beta should be, should be possible to roll out FSD Beta worldwide by the end of this year.
2:52:43Tesla PresenterBut we, you know, for a lot
2:52:44Elon Muskof countries we need regulatory approval and so we are somewhat gated by the regulatory approval in other countries. But you know, but I think from technical standpoint it will be ready to go to a worldwide beta by the end of this year.
2:53:02Elon MuskAnd there's quite a big improvement that we're expecting to release next month that will be especially good at assessing the velocity of fast moving cross traffic and a bunch of other things.
2:53:16Elon MuskSo, anyone elaborate?
2:53:22John EmmonsYeah, 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.
2:53:31John EmmonsI 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.
2:53:45John EmmonsSo 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.
2:54:01Elon MuskYeah, and the version of FSD beta that I drive as 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.
2:54:14Elon MuskBut 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.
2:54:29Elon MuskBut we're pretty close to the that. I mean, I think it's. I don't know, maybe it'll definitely be before the end of the year and maybe November.
2:54:41Paril JainYeah, 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 FSC stack before the end of this year.
2:54:53Paril JainSo 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.
2:55:02Elon MuskAnd in terms of the like, the fundamental metric to optimize against is how many miles per between a necessary intervention. So just massively improving the how many miles the car can drive on in full autonomy before an intervention is required.
2:55:22Elon MuskThat is, is safety critical. So yeah, that's, that's the fundamental metric that we're measuring every week and we're making radical improvements on that.
2:55:36Kate Park (Tesla)Oh hi thank you.
2:55:38Tesla PresenterHi.
2:55:38Lizzie MiskovetzThank you so much for the presentation. Very inspiring.
2:55:42Kate Park (Tesla)My name is Daisy.
2:55:43Tesla PresenterI actually have a non technical question for you.
2:55:46Kate Park (Tesla)I'm curious if you, you are back to your 20s, what are some of
2:55:50Elon Muskthe things you wish you knew back then?
2:55:52Kate Park (Tesla)What are some advice you would give to your younger self?
2:56:07Elon MuskWell, I'm trying to figure out something useful to say. Yeah, join Tesla would be one thing. Yeah, I think just trying to try to expose yourself to as many smart people as possible.
2:56:28Elon MuskI don't read a lot of books,
2:56:34Elon MuskYou know, I do, I did do that though.
2:56:37Pete BannonSo.
2:56:42Elon MuskI think there's some merit to just also like not being like necessarily too intense and like enjoying the moment a bit more. I would say to 20 or 20 something me just, you know, stop and smell the roses occasionally would probably be a good idea.
2:57:02Elon MuskYou know, it's like when we were developing the Falcon 1 rocket and on the Kwajalein Atoll and we had this beautiful little island that we're developing the rocket on and not once during that entire time did I even have a drink on the beach.
2:57:18Elon MuskI'm like, I should have had a drink on the beach. That would have been fine.
2:57:26Tesla PresenterThank you very much.
2:57:28Phil (Tesla)I think you have excited all of
2:57:29Elon Muskthe robotic people with optimization.
2:57:32Eric (Tesla)This feels very much like 10 years ago in driving.
2:57:35Phil (Tesla)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
2:57:44Tesla Presentermake for example AGI on a humanoid come faster?
2:57:49Elon MuskWell, 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.
2:58:11Elon MuskIt's, it's able to create extremely impressive art, engage in conversations that are very sophisticated, you know, write essays and these, these just keep improving. And there's, there's so much more, so, so many more talented people working on AI and the hardware is getting better.
2:58:38Elon MuskI think it's, it's a, AI is on a super, like a strong exponential curve of, of improvements independent of what we do at Tesla. And obviously we'll benefit somewhat from that exponential curve of improvement with AI.
2:58:57Elon MuskTesla just also happens to be very good at actuators that motors, you know, motors, gearboxes, controllers, power electronics, batteries, sensors. And you know, really like I say that, you know, the biggest difference between the robot on four wheels and the, the robot with arms and legs is getting the actuators right.
2:59:19Elon MuskIt'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, it's. I don't know, we have to have like the ingredients necessary to create a compelling robot.
2:59:37Elon MuskAnd we're doing it.
2:59:42Ashok ElluswamyHi, Elan. You are actually bringing the humanity to the next level, literally Tesla, and you are bringing the humanity to the next level.
2:59:52Tesla PresenterSo you said Optimus Prime.
2:59:54Ashok ElluswamyOptimus will be used in next Tesla factory.
2:59:57Tesla PresenterMy question is, will a new Tesla
3:00:00Ashok Elluswamyfactory will be fully run by Optimus program?
3:00:04Tesla PresenterAnd, and when can general public order a humanoid?
3:00:10Elon MuskYeah, 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, loading apart, you know, carrying a part from one place to another or loading a part into
3:00:28Paril Jainone of our
3:00:28Elon Muskmore conventional robot cells to, you know, that welds buddy together. So we'll start, you know, just trying to how do we make it useful at all? And then, and then gradually expand the number of situations where it's useful.
3:00:45Elon MuskAnd I think that, that the number of situations where Optimus is useful will, will grow exponentially, like really, really fast. In terms of when people can order one, I don't know.
3:00:59Elon MuskI think it's not that far away. Well, I think you meant when can people receive one? So I don't know, I'm like, I'd say probably within three years, not more than five years.
3:01:15Elon MuskWithin three to five years you could probably receive an optimus.
3:01:24Tesla PresenterI 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 human or research at the moment, would it make sense for the kind of Tesla to sort of open source some of the simulation hardware parts?
3:01:44Tesla PresenterI think Tesla can still be the dominant platformer where it can be something
3:01:48Elon Musklike Android OS or like iOS stuff
3:01:51Tesla Presenterfor the entire humanoid research. Would that be something that rather than keeping the optimist to just Tesla researchers or the factory itself can open it and let the whole world export humanoid research?
3:02:09Elon MuskI 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.
3:02:19David (Tesla)So
3:02:22Elon MuskI think, you know, we'd provide optimists where you can provide instructions to optimists, but where those instructions are, you know, governed by some laws of robotics that you cannot overcome.
3:02:41Elon MuskSo, you know, not Doing harm to others and I think probably quite a few safety related things with Optimus.
3:02:51John EmmonsYeah.
3:02:52Elon MuskSo. All right, we'll just take maybe a few more questions and then, 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, the ideal controller bandwidth?
3:03:09Elon MuskAnd then in the broader question, there's this big advertisement for the depth and
3:03:13Phil (Tesla)breadth of the company.
3:03:15John EmmonsWhat is it uniquely about Tesla that enables that.
3:03:21Elon MuskAnyone want to tackle the bandwidth question?
3:03:23Tesla PresenterYeah, so the technical bandwidth of the.
3:03:27Elon MuskClose to your mouth and loud.
3:03:29Tesla PresenterOkay. For the bandwidth question you have to understand or figure out what is the task that, that you wanted to do and what is the, if you took a frequency transform of that task, what is it that you want your limbs to do?
3:03:41Tesla PresenterAnd 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 from, that's where the bandwidth comes from.
3:03:50Tesla PresenterWhat is the broad question?
3:03:52Elon MuskI don't quite remember the breadth and depth thing. I can answer the breadth and depth.
3:04:05Elon MuskI mean, it's interesting on the bandwidth question, I think we probably will just end up increasing the bandwidth or you know, which translates to the effective dexterity and reaction time of the, of the robot.
3:04:19Elon MuskLike you can safe say it's not one hertz and it's. Maybe you don't need to go all the way to 100 hertz, but I don't know, maybe 1025. I don't know.
3:04:31Elon MuskOver 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. Yeah, minimize latency, maximize dexterity in terms of breadth and depth.
3:04:50Elon MuskI guess we've got, we've got, 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 autonomous or in order to make electric cars.
3:05:01Elon MuskAnd then in order to make autonomous electric cars, we've, we've just, I mean, Tesla is like a whole series of startups basically, and so far they've almost all been quite successful.
3:05:17Elon MuskSo we must be doing something right. And I consider one of my core responsibilities 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 it's, you know, and some of the companies that engineering talent is suppressed in a way that is maybe not obviously bad, but where it's just so comfortable and you're paid so much money and the output you actually have to produce is so low that it's like a honey trap, you know.
3:06:06Elon MuskSo like 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?
3:06:19Elon MuskAnd the output of that engineering talent seems very low, even though they seem to be enjoying themselves. That's why I call it there's a few honey trap companies in Silicon Valley.
3:06:32Elon MuskTesla is not a honey trap. And we're demanding and it's like going to get a lot of shit done and it's going to be really cool and it's not going to be easy.
3:06:44Elon MuskBut 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. So.
3:07:04Ashok ElluswamyHi Elan, I have two questions, so both to the Autopilot team.
3:07:09Tesla PresenterSo the thing is like I have
3:07:10Ashok Elluswamybeen following your progress for the past few years. So today you have made changes on
3:07:14Tesla Presenterlike the lane detection.
3:07:15Ashok ElluswamyLike you said that like previously you
3:07:17Tesla Presenterare doing instant semantic segmentation now you
3:07:19Ashok Elluswamyguys have built transfer models for like building the lanes. So what are another, some, some other common challenges which you guys are facing
3:07:26Tesla Presenterright now, like which you are solving in future as a curious engineer so
3:07:30Ashok Elluswamythat like we as a researcher can work on those.
3:07:33Tesla PresenterStart working on those.
3:07:34Elon MuskAnd the second question is like I'm
3:07:35Ashok Elluswamyreally 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
3:07:46Tesla Presenterthe data which you have like so little bit more on the data engine would be great. So that's it. Okay, I'll start.
3:07:52Phil (Tesla)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.
3:08:03Phil (Tesla)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 in all weather conditions, it's actually really, really challenging.
3:08:17Phil (Tesla)So only over a year ago we're kind of like driving a 2D world. If there's a wall and if there's curve, we kind of represent with the same static edge which is obviously, you know, not ideal.
3:08:30Phil (Tesla)Right. There's a big difference between a curve and a wall. When you drive, you make different choices.
3:08:34Tesla PresenterRight.
3:08:34Phil (Tesla)So after we realize 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 challenges we have, we have conquered in the past year.
3:08:51Kate Park (Tesla)Yeah. To answer the question about how we actually source examples of those 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.
3:09:03Kate Park (Tesla)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.
3:09:16Kate Park (Tesla)So these are just different, like various trigger logic that allows us to get those data campaigns back.
3:09:25Tesla PresenterHi, 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, one of the reasons why it's such a hard problem is because the problem itself is so hard to define.
3:09:40Tesla PresenterSeveral companies have several different definitions. They focus on different things. So what is Tesla, how is Tesla
3:09:46Paril Jaindefining the AGI problem?
3:09:47Tesla PresenterAnd what are you focusing on specifically?
3:09:51Elon MuskWell, we're not actually specifically focused on AGI. I'm simply saying that AGI seems likely to be an emergent property of, of what we're doing because we're creating all these autonomous cars and autonomous humanoids that are actually within a truly gigantic data stream that's coming in and being processed.
3:10:18Elon MuskIt'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, people and interacting with the roads and just, you know, Earth is a big place and reality is messy and complicated.
3:10:35Elon MuskSo I think it's sort of like you just. It just seems likely to be an emergent property of if you've got, you know, tens or hundreds of millions of autonomous vehicles and maybe even a comparable number of humanoids, maybe more than that.
3:10:50Elon MuskOn the humanoid front, well, that's just the most amount of data. And if that video is being processed, it just seems likely that, you know, the cars will definitely get way better than human drivers and the humanoid robots will become increasingly indistinguishable from humans, perhaps.
3:11:16Elon MuskAnd 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.
3:11:37Elon MuskThat seems to be way back in the early days of 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 knowledge, certainly a huge part of it, Whereas previously we would exchange information by osmosis in order to transfer data.
3:12:03Elon MuskSo 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, insanely slow when you think about it.
3:12:19Elon MuskAnd 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.
3:12:30Elon MuskSo, I mean, one of the great sort of equality elements, like the Internet, has been the most. The biggest equalizer in history in terms of access to information and knowledge.
3:12:48Elon MuskAnd any student of history, I think, would agree with this because 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 small number of people even had a book.
3:13:03Elon MuskNow look at it like you can access any book instantly. You can learn anything for 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?
3:13:25Elon MuskAnd my answer was, right now, this is the most interesting time in history, and I read a lot of history. So let's. Let's do our best to keep that going.
3:13:38Elon MuskYeah. And to go back to one of the earlier questions, I would answer like you can. The thing that's happened over time with respect to Tesla Autopilot is that we've just.
3:13:49Elon MuskThe neural nets have gotten. 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.
3:14:09Elon MuskAnd in principle, you could train with nothing in between, because that's what humans are doing with a 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.
3:14:30Elon MuskWe're not there yet, but it's gradually going in that direction. All right, maybe last question.
3:14:40John EmmonsHow you going?
3:14:42Elon MuskI think we've got a question at the front here. Hello. They're Right there. We'll do two questions. Fine.
3:14:48Tesla PresenterOver there. Hi. Thanks for such a great presentation.
3:14:52Elon MuskWe'll do your question last.
3:14:54Tesla PresenterOkay, cool. With FSD being used by so many people, do you think, what's the. 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 how.
3:15:08Tesla PresenterWhat's good enough? And do you. Defining, like thresholds for performance across so many miles?
3:15:16Pete BannonSure.
3:15:17Elon MuskWell, the, you know, the number one design requirement at Tesla is safety. So. 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.
3:15:36Elon MuskGovernment for just a passive mechanical safety, essentially crash structure and airbags and whatnot. We have the best, 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 like just absurdly better than a human.
3:16:02Elon MuskAnd then with respect to Autopilot, we do publish this, broadly speaking, the statistics on miles driven with cars that have no autonomy, or Tesla cars with no autonomy, with kind of hardware 1, hardware 2, hardware 3, and then the ones that are in FSD beta.
3:16:23Elon MuskAnd we see steady improvements all along the way. And, you know, sometimes there's this dichotomy of, you know, should you wait until the car is like, I don't know, three times safer than a person before deploying any technology.
3:16:38Elon MuskBut I think that's, that is actually morally wrong. At the point at which we believe that adding autonomy reduces injury and death, I think you have a more 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.
3:17:02Elon MuskAnd the people, the people who do occasionally die or get injured, they definitely know or their state does that it was, you know, whatever. There was a problem with Autopilot.
3:17:14Elon MuskThat'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, you know, we've got well over 3 million cars on the road.
3:17:25Elon MuskSo this, it's, that's a lot of miles driven every day. It's not going to be perfect. But what matters is that it is very clearly safer than not deploying it.
3:17:38Elon MuskYeah. So I think. Last question.
3:17:46Tesla PresenterI think. Yeah.
3:17:46Elon MuskSo thanks. Well, the last question here.
3:17:56Kate Park (Tesla)Okay. Hi. So I do not work on hardware,
3:18:01Elon Muskso maybe the hardware team and you
3:18:03Paril Jainguys can enlighten me.
3:18:05Kate Park (Tesla)Why is it required that there be symmetry in the design design of Optimus? Because humans, we have handedness, right?
3:18:15Elon MuskWe are, we use some set of
3:18:17Tesla Presentermuscles more than others.
3:18:19Kate Park (Tesla)Over time there is wear and tear, right?
3:18:22Paril JainSo maybe you'll start to see some
3:18:23Elon Muskjoint failures or some actuator failures more over time.
3:18:28Kate Park (Tesla)I understand that this is extremely prestige also. We as humans have based so much
3:18:36Paril Jainfantasy and fiction or superhuman capabilities. Like all of us don't want to
3:18:41Kate Park (Tesla)walk right over there. We want to extend our arms and
3:18:44Elon Musklike we have all these, you know,
3:18:46Paril Jaina lot of fantasy fantastical designs.
3:18:49Kate Park (Tesla)So considering everything else that is going on in terms of batteries and intensity
3:18:56Elon Muskof compute, maybe you can leverage all
3:18:59Kate Park (Tesla)those aspects into coming up with, with something. Well, I don't know, more interesting in
3:19:05Elon Muskterms of your, the robot that you're building.
3:19:08Kate Park (Tesla)And I'm hoping you're able to explore those directions.
3:19:13Elon MuskYeah, I mean, 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 a basic humanoid work well.
3:19:25Elon MuskAnd our goal is fastest 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 of like how many people did you help?
3:19:51Elon MuskHow 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.
3:20:11Elon MuskSo I could say like, man, there's a hell of a difference between a company that has shipped product and one has not shipped product. To get this is night and day.
3:20:17Elon MuskAnd 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.
3:20:28Elon MuskSo But I think over time I think 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.
3:20:43Elon MuskLike companies that are able to add things to an Optimus. Like maybe we've, you know, add a power port or something like that or attach them, you can add, you know, add attachments to your Optimus.
3:20:55Elon MuskLike you can add them to your phone. It could be a lot of cool things that could be done over time and it could be maybe an ecosystem of small companies that or big companies that make add ons for Optimus.
3:21:07Elon MuskSo with that, I'd like to thank the team for their hard work. You guys are awesome. And.
3:21:21Elon MuskThank 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, 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.
3:21:45Elon MuskSo 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'd be great to sort of bring you along for the ride and show you what cool things are happening and, and yeah, thank you.
3:22:04Tesla PresenterAll right. Thanks,