[00:00] I'm here today to answer your questions from the internet. [upbeat music] of these food delivery robots."
[00:15] They obviously work in some scenarios I think the bigger thing they're trying to solve like automated package delivery,
[00:28] A lot of these are first steps which is actually a really hard problem. has taught us a lot about both what works
[00:40] They can hit things, they can fall off the road. And I think that's sort of something to keep in mind that are actually in the field today
[00:54] And it tells you we got a lot of work to do. From Paradise Nights, I thought they were supposed to be our slaves,
[01:09] They're just partying." First and foremost, why are they having robots dance? that's happened in humanoids,
[01:24] I mean really the behaviors we've achieved is in the last year or two has been absolutely remarkable. And the way you do it is you start
[01:37] The reason why the dancing looks so human-like very beautiful and advanced engineering puppetry and the human dances, not me 'cause I'm a very bad dancer,
[01:53] and you get the trajectories from that data, And then you train a reinforcement learning algorithm or copies that human data as much as possible
[02:06] And the end result is it dances like the person meaning the environments like it was But you asked about this,
[02:19] And the answer to that is that we still don't know That's sort of the not-very-secret secret to be truly autonomous and in our homes
[02:32] The next question is from Projectguy111, in a Terminator future?" One is, what's the chance that AI in learning
[02:46] And I think that probability's actually fairly high If you trust AI to be the decision maker, for that AI, it will make bad decisions.
[02:59] you see that it can produce really nice answers sometimes, So we cannot trust AI. but you can use AI as a powerful tool.
[03:13] you get a lot of results back, So I think if we put, like, AI in charge then, you know, bad things will happen.
[03:27] It actually learned to think on its own, So I have no concern of sentient AI. They say AI, artificial intelligence,
[03:41] It has no notion of what it's saying or doing. we've never pattern-matched before. "What benefits do legged robots have over wheeled
[03:53] Legs are inherently beneficial if you wanna operate and more importantly, where things are not flat. as long as there's not uneven terrain.
[04:08] or wheeled someone out around in a wheelchair, Even what you perceive as flat, you realize there's curbs that don't dip enough,
[04:22] They become sort of sticking points to walk more robustly over multiple terrain types. I mean, where your dog or your cat can go on four legs
[04:36] Bipeds, of course, are sort of the ultimate expression So if you need to get into a small space or climb up a ladder, only a biped can do that.
[04:50] Give a ladder to a human, From SamMee514, So we've actually come a long ways in robot dogs,
[05:05] 'cause they have four legs. where we've gone from these robot dogs and research environments
[05:18] The hardware has made immense progress. I think the practical use cases are still a little thin. sending robots ahead in disastrous scenarios, right?
[05:33] The next question is from Leovan21, Yeah, there's lots of attempts. that have that as a layer in their humanoid robots.
[05:48] is that as we're making robots do things, So, imagine your body, you have a brain, So you sort of intrinsically have multiple computers
[06:03] Like, your spinal cord is a computer in its own right So at the highest level, at our cognitive level, So you wanna ask the robot a question
[06:17] So ask the robot, you know, It can use an LLM to parse that language into code, on the table and return based on that an estimate
[06:32] an LLM is not gonna reach for the apple. Or where reinforcement learning will come in, And those are what's actually running on the robot, right?
[06:46] is really what moves your body most of the time "Are autonomous vehicles mobile robots?" what the most advanced robot is today,
[07:01] and mobile warehouse robots like Amazon has. Autonomous cars, I think, are one of the prime examples So yeah, it's absolutely a robot,
[07:16] From Vnightpersona, I think what they're asking is how does the Roomba is the way we would technically phrase this question.
[07:31] that it sees and correctly identifying them. a bunch of examples of pictures on the internet, By teach, what we mean is we can train up
[07:47] basically this big neural network that takes images in And as a result, we can now correctly identify And so what Roomba does is, simply, it has a camera
[08:01] It's checking with the internet based on these large models and using that information to tell you what's going on. "Why make robots humanoid shaped
[08:14] Humanoid robots are the most suited to do the most things, Again, the world is built for us. We've built the robot for something of our shape
[08:29] Doors, stairs, narrow corridors, all these things. where a human can go, you want somebody to start making you sandwiches
[08:43] that can be dropped into an existing factory I wanna say all that as a preface to the fact for, again, a given application.
[08:57] the largest owner of robots is Amazon. one of the largest sector of robotics is warehouse robots. and then move them around warehouses.
[09:12] with the thing you order from Amazon, So the human can just stand there, Now you could theoretically have a humanoid robot do this,
[09:26] Like, these robots are very good to work synergistically with the robot. if you had an application that determined
[09:40] and you find that two arms is not enough and you wanna hold fire extinguisher. A Reddit user asks, "When it comes to automation,
[09:55] if not all, warehouse work with robots?" and then other ones are further away. Amazon is by far the leader in this domain.
[10:11] has been working on warehouse robots, to move pallets around, for, you know, 20-plus years now.
[10:23] so they maximally and efficiently can move packages So the question becomes what remains? which is grabbing these things off the shelf
[10:37] And so this is much more an open-ended work, So for that, you can use things called local manipulators, and maybe you use suction cups or soft graspers
[10:52] And you can do this with varying success levels right now, where you don't need a person present for all objects. all the different geometries and how they feel and look,
[11:08] to automatically and autonomously load all these objects. KaleidoscopeInside asks, with no financial barriers, what would your robot do?"
[11:22] I would try to sort of solve the exoskeleton problem. a billion in my mind could develop something
[11:34] A Reddit user asks, have their knees look backwards?" There's potentially mathematical advantage
[11:49] It's not just that they're inverted, So it turns out the way you control robots on robots is by having very light legs
[12:02] There's lots of mathematical reasons for that, And what that means is as you're walking, to catch yourself as you fall.
[12:14] And in addition, 'Cause birds actually have this type of morphology birds have some of the most robust locomotion out there.
[12:29] and then step up out of the hole And so if you can get that kind of behavior on robots, A Reddit user asks,
[12:43] "What is so special about the Mars rover Curiosity?" Pretty much everything, it was on Mars. are incredible engineering feats.
[12:57] And then the rover itself, they had these amazing engineering sort of gems in them You know, their suspension system is specifically built
[13:11] Their wheels are specifically built so they'll be robust So the entire structure is built to maximally be robust which have to deal with this very sandy
[13:27] where there's dust storms, you know. I mean, the sun doesn't come through at the same intensity. Electronics have to be battery-efficient, right?
[13:41] you have to talk back with Earth at the same time But to build a system like that, is an immense and remarkable engineering feat
[13:55] that sort of just makes me happy as an engineer. "Wait, how do autonomous cars even work? Like, are we trusting robots with our lives now?
[14:09] have gone through some things. there were all these startup companies that formed, And then exactly what you're worried about happened.
[14:25] Despite the fact that they'd already been working and I think that the entire autonomous car industry there was really rigorous safety methods.
[14:38] software, architectures, et cetera, And so in that way, there's a lot of evidence to show have very rigorous safety standards,
[14:52] with the limited number of crashes that they've had. but, you know, after a decade, A decade ago, we could have autonomous cars
[15:04] for, you know, a demo and actually driving around and a kid chases after it, right? meaning if you're not safe,
[15:18] you die, both as a company, but also as a technology. Why does Elon Musk think LiDAR is not a good idea?" but in my mind, LiDAR is awesome.
[15:34] autonomous cars, et cetera. the two most popular things, Now, LiDAR sends out a bunch of laser pulses
[15:48] So it's sort of 3D radar, but now it's using lasers, LiDAR. They can tell you not just how far objects are, but what objects there are
[16:03] the environment and things like that. of precisely identifying everything in the environment so you know exactly where all the objects are,
[16:15] LiDAR is super effective on robots and on autonomous cars It doesn't matter if you're sort of hitting another car
[16:28] You don't wanna hit any of that stuff. to do things like dynamic collision avoidance. And so I imagine that what they're thinking at Tesla,
[16:43] and make them as good as LiDAR, But I think what I found on robots And so for that, for safety critical applications,
[16:57] you want LiDAR in the loop to dynamic changes in the environment that's present in a lot of perception-based representations.
[17:11] "Why is a surgical robot better than a surgeon?" at certain things. They're very, very good at precise small motions
[17:24] repeatedly again and again and again. Basically, any place where you have to interact with the environment in a soft and tactile way.
[17:37] because our skin is soft, our organs are soft, to perform surgery, and sort of compliant interactions with the human body.
[17:51] The surgical robot can do those precise motions and they get all this haptic and tactile feedback while still being able to do what humans do really well,
[18:06] This is from theballisrond, The problem with clothes They move around, there's fabric.
[18:20] So a robot has to interact with something in the environment And so that's why this is a big challenge to try to understand this clothes folding problem.
[18:35] like you teleoperate the robot and then you try to teach the robot that task of the environment itself.
[18:47] that humans generated through teleoperation. "Whatever happened to the NEO robot?" that you could buy like a humanoid robot
[19:00] Now, to give them credit, they completely were transparent about the fact There's even in the app,
[19:14] with a teleoperator. One is it's a tacit, at least, acknowledgement I mean, that's the clear and obvious ramification.
[19:29] meaning it can't do a lot of tasks on its own, very few? is that the only thing that's missing is data. and that's what made LLMs possible.
[19:44] is if we can get sort of internet-scale humanoid robot data, just like ChatGPT solve the problems for language, is not so much that they think
[19:57] but rather this is an amazing way to do data collection. and then people agree to let you teleoperate them and then they're gonna use this data to train a large model
[20:12] If you can get early adopters to sign up for that, to actually train large models is that I don't think it's gonna work.
[20:26] you can collect in this way People tend to confuse LLMs and text with robots, In particular, the amount of data needed to understand
[20:41] think about them as trajectories as the language of robots. all this rich dynamic information. that it makes language seem simple.
[20:57] We developed language in a very small fraction was to do these other basic things, right? That's why they say you only use 10% of your brain.
[21:12] for these very complicated motions that we do, right? is I don't think that data's gonna work. not that it can't be helpful or help to polish things,
[21:26] You need to merge physics with human data the general intelligence problem on humanoid robots, That's it. That's all the questions.
[21:41] [gentle bright music]