WEBVTT - Building a Robot That Can Walk the Walk 0:00:15.356 --> 0:00:23.156 Pushkin. So how long have you been trying to make 0:00:23.196 --> 0:00:23.916 a robot walk? 0:00:25.396 --> 0:00:27.796 It's been my entire career, starting from why I went 0:00:27.836 --> 0:00:29.396 to college in the first place. 0:00:29.636 --> 0:00:34.996 Why why that particular problem? Why is that your life's work? 0:00:35.556 --> 0:00:38.436 You know, there's few things more interesting and more dynamically 0:00:38.436 --> 0:00:41.436 complex and more elegant than the way animals move in 0:00:41.476 --> 0:00:43.876 the world. And to be able to get machines that 0:00:43.996 --> 0:00:46.436 can move that way, they can interact physically with the 0:00:46.436 --> 0:00:49.996 world the way humans and animals do. What a fun 0:00:50.076 --> 0:00:51.796 and interesting thing to work on for a career. 0:00:57.756 --> 0:01:00.076 I'm Jacob Goldstein and this is What's Your Problem? The 0:01:00.116 --> 0:01:02.036 show where I talk to people who are trying to 0:01:02.116 --> 0:01:06.796 make technological progress. My guest today is Jonathan Hurst. He's 0:01:06.796 --> 0:01:10.236 a professor at Oregon State University and founder and chief 0:01:10.436 --> 0:01:15.596 robot officer at Agility Robotics. Agility Robotics has built a 0:01:15.716 --> 0:01:19.196 robot called Digit. Digit looks kind of like a person. 0:01:19.316 --> 0:01:21.996 It walks around on two legs. It's got this flat, 0:01:21.996 --> 0:01:24.596 rectangular head, and it has two arms that it can 0:01:24.676 --> 0:01:28.836 use to pick stuff up. Jonathan's problem is this, how 0:01:28.876 --> 0:01:31.636 can you make a walking robot that can do useful 0:01:31.676 --> 0:01:36.316 work and that companies will actually pay for. Jonathan says 0:01:36.356 --> 0:01:40.276 that robot Digit is already being tested out in warehouses 0:01:40.356 --> 0:01:41.276 in the real world. 0:01:43.956 --> 0:01:46.676 We are deploying robots with cost We have two announced. 0:01:46.676 --> 0:01:49.596 We've announced a couple with Amazon and with GXO. You know, 0:01:50.036 --> 0:01:52.996 you place an order and Digit handles that order as 0:01:53.036 --> 0:01:56.516 part of the workflow that has happened is happening right now, and. 0:01:56.436 --> 0:02:02.876 So specifically, what are your robots doing well? First at Amazon, Yeah, 0:02:02.876 --> 0:02:05.636 the first use case or the first class of use 0:02:05.676 --> 0:02:08.836 cases that works for us is picking up these plastic 0:02:08.876 --> 0:02:11.996 tote plastic bins and putting them somewhere else. And then 0:02:12.556 --> 0:02:18.316 in warehouses, Yeah, but anywhere warehouses, in logistics, in manufacturing, 0:02:18.436 --> 0:02:20.796 you know, the whole environments like that that are a 0:02:20.796 --> 0:02:24.236 bit structured. They're kind of like there's these islands of 0:02:24.276 --> 0:02:27.156 automation they call them, you know, where one machine is 0:02:27.196 --> 0:02:30.076 putting things in a bin, another machine sorts bins and 0:02:30.116 --> 0:02:33.596 sends them different parts of the warehouse. Right now, sometimes 0:02:33.636 --> 0:02:36.156 a person will stand there, the robot will tell them 0:02:36.156 --> 0:02:38.036 which bin to pick up, and then all they are 0:02:38.036 --> 0:02:41.156 basically is a manipulator for the robots system. They pick 0:02:41.236 --> 0:02:42.596 up the bin and put it on the conveyor belt 0:02:42.636 --> 0:02:44.076 and wait for the robot to tell them the next 0:02:44.076 --> 0:02:46.836 thing to do. And it's really hard to hire people 0:02:46.876 --> 0:02:50.116 for that. There are a lot of open jobs in that, 0:02:50.276 --> 0:02:52.596 and so it's kind of a perfect place for Digit 0:02:52.676 --> 0:02:56.996 to walk in in this relatively structured first use case. Now, 0:02:57.116 --> 0:03:01.236 digits of course going to evolve towards picking up boxes, depalletizing, 0:03:01.636 --> 0:03:04.876 loading and unloading you know, tractor trailers and eventually getting 0:03:04.876 --> 0:03:08.636 out to things like retail and stocking shelves and you 0:03:08.676 --> 0:03:12.036 know sticks in hospitals carrying things around and eventually become 0:03:12.036 --> 0:03:15.876 a consumer product. So that's where you are today and 0:03:15.916 --> 0:03:18.356 where you want to get. I want to talk now 0:03:18.436 --> 0:03:23.916 about how you got here. Sure, and there's this really 0:03:24.356 --> 0:03:27.836 basic set of things you had to figure out just 0:03:28.036 --> 0:03:33.876 around how locomotion works, right, how people walk, also interestingly 0:03:33.916 --> 0:03:37.676 and sort of surprisingly, how how birds like ostriches walk. 0:03:37.996 --> 0:03:42.076 So I know there is this series of robots that 0:03:42.116 --> 0:03:45.036 you built on the way on the way to Digit, right, 0:03:45.196 --> 0:03:47.756 they were two before and then Digit and it seems 0:03:47.796 --> 0:03:50.396 like going through those and what you figured out the 0:03:50.476 --> 0:03:53.716 sort of key insight on each one is a really 0:03:53.836 --> 0:03:57.236 nice way to get a kind of deeper insight into 0:03:57.796 --> 0:03:59.716 how it works and what you had to understand to 0:03:59.756 --> 0:04:02.356 make a robot that can walk great, just being a 0:04:02.396 --> 0:04:05.556 really hard problem, right, Like there's like lots of robots 0:04:05.556 --> 0:04:08.116 that we don't even sort of think of as robots arms, 0:04:08.156 --> 0:04:11.396 and you know, self driving cars are arguably a kind 0:04:11.396 --> 0:04:13.716 of robot and whatever, but like getting a robot to 0:04:13.756 --> 0:04:16.556 walk is clearly a very hard problem. 0:04:16.716 --> 0:04:20.836 Yeah, Okay, So where I started in trying to understand 0:04:20.916 --> 0:04:24.516 how to make a machine work is on the biomechanics, 0:04:24.756 --> 0:04:27.156 try to understand how animals work. And a lot of 0:04:27.156 --> 0:04:30.596 biomechanics is about specific muscles and muscle groups and joints, 0:04:30.636 --> 0:04:33.836 and we're instead looking at holistically from a big picture, 0:04:34.196 --> 0:04:36.396 what is the center of mass of an animal doing? 0:04:36.436 --> 0:04:38.956 What are the forces happening on the ground. I did 0:04:38.956 --> 0:04:42.356 this collaboration with Monica Daily at the Royal Veterinary College 0:04:42.436 --> 0:04:45.356 and we looked at guinea fowl and ostriches and turkeys 0:04:45.396 --> 0:04:47.636 and you know, a whole bunch of different sizes. 0:04:47.876 --> 0:04:51.036 Why birds. It's really interesting to me that you did that. 0:04:51.196 --> 0:04:55.156 Why because humans are weird? How many other bipeds are there? 0:04:55.236 --> 0:04:58.436 You know, all the evolved bipeds, all of the extant 0:04:58.556 --> 0:05:00.836 existing bipeds in the world. 0:05:00.756 --> 0:05:02.836 Everything that walks on two legs. Yeah, huh. 0:05:03.236 --> 0:05:07.276 Those are all theropods. They're all kind of more like 0:05:07.316 --> 0:05:10.756 a bird than they are like or a monkey. They 0:05:10.796 --> 0:05:14.076 have evolved far longer than we have. You know, we've 0:05:14.076 --> 0:05:16.996 been out of trees for a very short period of time. 0:05:17.036 --> 0:05:19.756 Compared to all of the other existing bipeds in the world. 0:05:20.116 --> 0:05:22.636 They can all run much faster than we can for 0:05:22.956 --> 0:05:27.836 you know, very efficiently using less energy insul So. 0:05:27.756 --> 0:05:31.316 Like an ostrich is a better model for just abstract 0:05:31.396 --> 0:05:34.836 walking on two legs. Oh sure, I mean, yeah, I 0:05:34.876 --> 0:05:36.516 love that, But I want to be clear. 0:05:36.556 --> 0:05:38.636 What we're not trying to do is study how it 0:05:38.636 --> 0:05:40.636 does an Ostrich run versus how a human runs. What 0:05:40.636 --> 0:05:43.196 we're trying to do is study what is a fundamental 0:05:43.236 --> 0:05:46.476 truth between all animals in how they run, so we 0:05:46.516 --> 0:05:49.316 can try and weed out things that have to do 0:05:49.396 --> 0:05:51.596 with the size of the animal, or things that have 0:05:51.676 --> 0:05:54.036 to do with where exactly the ankle joint is or 0:05:54.036 --> 0:05:56.916 how long this joint is or that one. We want 0:05:56.956 --> 0:06:00.476 to know what is the same amongst all animals to 0:06:00.636 --> 0:06:01.156 walk and run. 0:06:01.196 --> 0:06:04.556 It's like the Platonic ideal of bipedalism. The sort of 0:06:04.556 --> 0:06:06.956 the fundrationing theory of it. Yeah, that's right. 0:06:07.076 --> 0:06:09.636 Yeah, And so biomechanists have been looking at this since 0:06:09.676 --> 0:06:11.996 the seventies and thinking about this in terms of spring 0:06:12.036 --> 0:06:14.116 mass models of locomotion. And it was only in like 0:06:14.156 --> 0:06:16.356 the two thousands sometimes that I think it was Hart 0:06:16.396 --> 0:06:19.596 McGuire and Andre Seifarth put together a paper that showed, hey, 0:06:19.956 --> 0:06:23.876 this spring mass model reproduces all of the behavior we 0:06:23.956 --> 0:06:27.116 see for walking and running and transitions between these gates. 0:06:27.756 --> 0:06:30.396 When you say, when you say spring mass model, that 0:06:30.436 --> 0:06:32.876 sounds big and exciting, But just to be clear, what 0:06:32.916 --> 0:06:35.036 do you mean when you say spring mass model. 0:06:35.036 --> 0:06:41.236 I mean a mathematical representation of a pogo stick. Go on, Okay, 0:06:41.436 --> 0:06:43.796 So a pogo stick is basically the simplest thing that 0:06:43.876 --> 0:06:46.756 can run, and a kangaroo looks a lot like a 0:06:46.756 --> 0:06:49.756 pogo sticky. Now, if you just stick a pogo stick 0:06:49.796 --> 0:06:52.276 on each leg, now you're bipedal running, you know. 0:06:52.356 --> 0:06:52.636 Okay. 0:06:53.036 --> 0:06:54.836 And then if you add a whole bunch of complexity 0:06:54.836 --> 0:06:57.516 to it, you have heel toe and you know, knee 0:06:57.556 --> 0:06:59.836 joints and all this other stuff. But if you really 0:06:59.876 --> 0:07:01.556 boil it down and try and make it as simple 0:07:01.556 --> 0:07:04.036 as possible, you get to some pretty basic math models 0:07:04.036 --> 0:07:07.356 that do represent how the progression of the center of 0:07:07.436 --> 0:07:10.196 mass of the animal moves, and how the ground reaction 0:07:10.356 --> 0:07:11.796 forces progress and so on. 0:07:11.916 --> 0:07:14.956 Right, Okay, so if I picture just like a lump 0:07:14.956 --> 0:07:17.476 of mass on top of two pogo sticks, you got it. 0:07:17.516 --> 0:07:18.276 I'm kind of in the right. 0:07:18.356 --> 0:07:21.556 Okay, you're absolutely okay. So that's a roughly a math 0:07:21.596 --> 0:07:23.716 model that at least gives you a very good concept 0:07:23.756 --> 0:07:28.036 of how do all animals run, horses, ghost, crabs, humans, ostriches, whatever. 0:07:28.356 --> 0:07:31.316 Right, So this paper comes out, this paper that says, 0:07:32.036 --> 0:07:33.836 think of a lump of mass on top of two 0:07:33.876 --> 0:07:37.436 pogo sticks. You you know about it because you're in 0:07:37.476 --> 0:07:39.876 this world? What do you What effect does it have 0:07:39.916 --> 0:07:41.516 on you? What do you do as a result of this. 0:07:42.116 --> 0:07:44.036 What I was looking at? And here's the argument at 0:07:44.036 --> 0:07:50.036 the time. Do these spring mass models simply seem to 0:07:50.076 --> 0:07:53.316 describe the things we're observing, you know? Or is it 0:07:53.396 --> 0:07:56.476 describing core physics of how it works? Like, in other words, 0:07:56.676 --> 0:07:58.956 if you build a spring mass model and a policy 0:07:58.996 --> 0:08:00.916 that works, is it going to make a robot stabilize? 0:08:01.116 --> 0:08:04.156 Or is it simply like a picture that kind of 0:08:04.196 --> 0:08:05.796 looks like what the animals are actually doing. 0:08:05.876 --> 0:08:08.316 So like if we actually do the lump of mass 0:08:08.316 --> 0:08:10.636 on top of two pot sticks as a robot, will 0:08:10.676 --> 0:08:11.036 it work? 0:08:11.436 --> 0:08:13.596 Yeah? And so look the question is, you know, how 0:08:13.636 --> 0:08:16.036 do you control these things? And then how does that 0:08:16.076 --> 0:08:18.716 translate into walking and running which had never really been 0:08:18.716 --> 0:08:21.476 done before that way, you haven't done this continuous transition 0:08:21.556 --> 0:08:23.996 between walking and running and changing the speed and everything else, 0:08:24.356 --> 0:08:26.676 and it's unknown how to stabilize that over all kinds 0:08:26.716 --> 0:08:29.436 of terrain. So that's why we built Atreus. That's why 0:08:29.476 --> 0:08:30.676 we built the robot Atreus. 0:08:30.836 --> 0:08:31.996 And so what is Atreus? 0:08:32.396 --> 0:08:35.076 So Atreus is it is a bipedal robot. 0:08:35.196 --> 0:08:37.076 Okay, what does Atrius look like? 0:08:38.636 --> 0:08:41.556 So Atrius was on the Late Night with Stephen Colbert 0:08:41.796 --> 0:08:44.036 and he described it as a microwave on stilts. 0:08:44.676 --> 0:08:51.076 Okay, that's good. So it doesn't it doesn't look humanoid 0:08:51.076 --> 0:08:53.836 at all. It looks maybe like a dancing alien or 0:08:53.916 --> 0:08:57.996 like a moon lander or something, but not humanoid. 0:08:58.436 --> 0:09:00.916 Yeah, absolutely not. So it doesn't look like an animal 0:09:00.996 --> 0:09:03.756 in any way. But it's designed entirely to be the 0:09:03.756 --> 0:09:06.276 math model of what we see an animal running. And 0:09:06.516 --> 0:09:09.436 the name Atrius is an acronym for assume the robot 0:09:09.516 --> 0:09:12.476 is a sphere, right. The whole idea is the robot 0:09:12.556 --> 0:09:13.796 is this simple math model. 0:09:14.316 --> 0:09:16.356 Where it's just some mass in the middle and then 0:09:16.396 --> 0:09:18.356 some very light springy legs. 0:09:18.676 --> 0:09:21.796 That's it. And so what we ultimately showed with Atreus 0:09:21.836 --> 0:09:25.916 it's the first robot ever to reproduce human walking gate dynamics. 0:09:25.996 --> 0:09:28.756 The robot walks across the force plate, a graduate student 0:09:28.756 --> 0:09:31.036 walks across the forest plate. Looking at the data, you 0:09:31.076 --> 0:09:32.156 can't actually tell a difference. 0:09:32.276 --> 0:09:35.596 Huh. So, like from the plate's point of view, it 0:09:35.676 --> 0:09:38.236 feels the same whether you're robot or a human is 0:09:38.276 --> 0:09:38.916 walking across. 0:09:38.916 --> 0:09:39.476 That's correct. 0:09:39.676 --> 0:09:41.956 So it's not true for earlier robots that looked like 0:09:41.996 --> 0:09:43.356 they were walking like humans. 0:09:43.076 --> 0:09:44.956 They would look the same. But if you looked at 0:09:44.996 --> 0:09:47.236 the dynamics of it, if you looked at the ground 0:09:47.276 --> 0:09:49.276 reaction forces, they differed quite a lot. 0:09:49.796 --> 0:09:52.196 Why is that important? Why is that important? 0:09:52.356 --> 0:09:55.396 It's just one symptom, right to show that, hey, we've 0:09:55.436 --> 0:09:58.316 actually captured the physics here. But the other symptom that's 0:09:58.356 --> 0:10:01.116 important is we are able to walk and run outdoors 0:10:01.196 --> 0:10:04.156 over all kinds of terrain without any sort of perception. 0:10:04.556 --> 0:10:07.476 The robot can handle amazing obstacles and it would just 0:10:07.516 --> 0:10:10.236 soak them up, you know, going over potholes, going from 0:10:10.276 --> 0:10:13.556 grass to pavement, going over big pieces of plywood we 0:10:13.596 --> 0:10:14.556 would throw in its way. 0:10:15.076 --> 0:10:16.756 And you're saying it didn't do this because it had 0:10:16.796 --> 0:10:18.956 like a clever brain. You're saying it was just the 0:10:18.956 --> 0:10:22.396 physics of the brainless machine was able to deal. 0:10:23.756 --> 0:10:26.596 No cameras on it, It had no awareness of the environment. 0:10:26.756 --> 0:10:29.756 It was a very simple spring mass model, very very 0:10:29.756 --> 0:10:32.276 simple control that did nothing but try to balance that. 0:10:32.596 --> 0:10:34.596 And it was able to just absorb all these kinds 0:10:34.596 --> 0:10:36.276 of disturbances and just keep on going. 0:10:36.716 --> 0:10:38.796 That sounds like a big deal. I agree. 0:10:39.356 --> 0:10:42.476 I'm very excited about this, right. That is the point 0:10:42.476 --> 0:10:45.036 that we decided we were going to found Agility Robotics. 0:10:45.476 --> 0:10:47.356 We said, you know, this was a mission for like 0:10:47.796 --> 0:10:50.556 years and years and years. That is why I became 0:10:50.596 --> 0:10:54.716 a professor, is to say, my goal here in academia 0:10:55.276 --> 0:10:58.916 is to show that this spring mass physics is real 0:10:59.316 --> 0:11:02.236 and really make sure we understand that well. And if 0:11:02.276 --> 0:11:04.156 we can show that and prove that and with this 0:11:04.276 --> 0:11:08.476 ATREUS project, we then can take the next steps. Right, 0:11:08.596 --> 0:11:12.636 So that's what we did. It was a scientific kind 0:11:12.636 --> 0:11:15.636 of breakthrough. But the machine could only walk and run. 0:11:15.956 --> 0:11:20.236 It couldn't stand, it couldn't turn. You know, it breaks often. 0:11:20.596 --> 0:11:23.156 If it ever fell, it would just be completely destroyed. 0:11:23.316 --> 0:11:26.236 So it's not a productive, useful machine. It's a science demonstrator. 0:11:27.596 --> 0:11:33.196 Okay, So you have a TRIOS that is academically it 0:11:33.356 --> 0:11:37.036 kind of intellectually a breakthrough, right, but nobody's going to 0:11:37.116 --> 0:11:40.356 buy it to do anything. It's not useful in a 0:11:40.516 --> 0:11:44.076 practical sense. What do you do next? 0:11:44.436 --> 0:11:46.996 Okay? So at that point we say, hey, you know, 0:11:47.316 --> 0:11:51.276 we understand a really significant portion of what does it 0:11:51.316 --> 0:11:53.516 take to make a robot that can go where people go? 0:11:54.436 --> 0:11:56.756 And what an opportunity? This is? What are all the 0:11:56.796 --> 0:11:58.316 things that robots are going to be able to do 0:11:58.756 --> 0:12:03.356 when they can be coexisting with humans? Right? And the 0:12:03.516 --> 0:12:06.436 barrier to moving forward on that now is more about 0:12:06.676 --> 0:12:11.036 execution on engineering and execution on building a use case 0:12:11.036 --> 0:12:12.516 in a business around it. 0:12:12.596 --> 0:12:15.196 Because you feel like you've solved the sort of core 0:12:15.356 --> 0:12:19.116 threshold technical problem of a robot that can walk in 0:12:19.276 --> 0:12:21.276 unfamiliar environments and not fall down. 0:12:21.476 --> 0:12:25.316 Correct, we have the foundation layer of leged locomotion. Now, 0:12:25.396 --> 0:12:27.676 sure we'd like to add on perception to it so 0:12:27.676 --> 0:12:30.036 that it can handle stairs effectively and things like that 0:12:30.116 --> 0:12:33.516 and intentionally handle obstacles. We need to do the engineering 0:12:33.556 --> 0:12:35.796 of the electrical system and the hardware system that still 0:12:35.796 --> 0:12:38.716 captures the same physics but can take a beating, can stand, 0:12:38.796 --> 0:12:41.076 can steer, can start to do things right. So we 0:12:41.156 --> 0:12:43.236 know we're going to need to build a robot with legs, 0:12:43.236 --> 0:12:46.996 with manipulators, with sensors. Okay, because so we're kind of 0:12:47.036 --> 0:12:49.956 going down a path now where we want to take 0:12:49.996 --> 0:12:54.116 the exact same first principles approach to how do we 0:12:54.156 --> 0:12:56.716 build a machine that can manipulate things in a human 0:12:56.756 --> 0:13:00.276 world and get around in it and interact with people. 0:13:00.316 --> 0:13:03.436 Build a human centric machine. So the first step to 0:13:03.476 --> 0:13:06.276 doing that was designing a robot that we could sell 0:13:06.316 --> 0:13:09.116 to other researchers to continue the work on the leg 0:13:09.156 --> 0:13:12.516 leggs as we then worked towards you know, arms and manipulators, 0:13:12.556 --> 0:13:15.996 and that was Cassie, our first robot that Agilia Robotics sold. 0:13:17.476 --> 0:13:20.676 Cassie added the ability to stand in place because it 0:13:20.716 --> 0:13:23.356 had ankles, and it had the ability to steer because 0:13:23.396 --> 0:13:26.556 you could turn the legs. But more importantly, it was 0:13:26.836 --> 0:13:30.716 much more compact and extremely robust, so this robot can 0:13:30.836 --> 0:13:32.996 fall hard on concrete and you just pick it back 0:13:33.076 --> 0:13:34.596 up and it can get going again. 0:13:34.956 --> 0:13:37.436 And so, Cassie, I'm looking at a picture of it now. 0:13:37.796 --> 0:13:42.316 It basically looks like a pair of Ostrich legs. Yeah, 0:13:42.316 --> 0:13:44.796 it looks like an Ostrich. Does an Ostrich have a wasist? 0:13:45.116 --> 0:13:47.596 I don't know an ostrich from the waist down, No. 0:13:47.596 --> 0:13:51.476 It doesn't know. Yeah, the pelvis is stationary and a 0:13:51.476 --> 0:13:54.116 bird fixed rather than a human pelvis, which is mobile. 0:13:54.476 --> 0:13:57.756 So I know it's not technically correct to say that 0:13:57.796 --> 0:14:00.756 an Ostrich legs bend backwards, but it looks that way, right. 0:14:00.836 --> 0:14:03.396 Yeah, Their thigh is very short, their knee is up 0:14:03.436 --> 0:14:05.676 next to their body, and what you perceive as their 0:14:05.756 --> 0:14:06.956 knee is actually their ankle. 0:14:07.236 --> 0:14:11.636 In designing this robot, how do you get to legs 0:14:11.716 --> 0:14:15.316 that look like Ostrich legs? Like, it's like convergent evolution, 0:14:15.516 --> 0:14:18.636 you know, convergent evolution maybe hopefully. Isn't it the case 0:14:18.676 --> 0:14:23.036 that like cephalopods that like whatever squids have eyes like 0:14:23.076 --> 0:14:26.076 our eyes, but they evolve totally independently. Is this like that? 0:14:26.516 --> 0:14:29.036 There's a lot of examples of convergent evolution. We can 0:14:29.076 --> 0:14:31.876 only guess, right because we don't necessarily know, and a 0:14:31.916 --> 0:14:35.276 scientists only hypothesize the evolutionary pressures that cause animals to 0:14:35.316 --> 0:14:36.076 be the shape that they are. 0:14:36.316 --> 0:14:38.876 But the pressure that led you you didn't say let's 0:14:38.916 --> 0:14:41.116 make legs that look like Ostrich eggs. You just did 0:14:41.156 --> 0:14:42.876 a bunch of math and you wound up with legs 0:14:42.876 --> 0:14:44.156 that look like Ostrich legs. 0:14:44.156 --> 0:14:46.316 That is correct, and there are a bunch of features 0:14:46.356 --> 0:14:48.836 on our robot that have gone down a similar path. 0:14:49.036 --> 0:14:51.476 And I actually love that because when we end up 0:14:52.636 --> 0:14:55.676 you know, blank sheet, pursue all of the different configuration 0:14:55.716 --> 0:14:57.516 options and say okay, here's what we think is the 0:14:57.516 --> 0:14:59.996 optimal choice, and we say, wow, that looks just like 0:15:00.036 --> 0:15:01.956 a person, or that just looks like a bird or something. 0:15:02.436 --> 0:15:04.556 It's actually really good. It means we're probably on the 0:15:04.596 --> 0:15:05.076 right path. 0:15:05.516 --> 0:15:09.076 Yeah, it's it's exciting in a way, right, like like 0:15:09.196 --> 0:15:11.276 you do a bunch of bath and then suddenly you 0:15:11.316 --> 0:15:12.876 look up and you see an Ostrich. 0:15:13.156 --> 0:15:15.116 But it won't always be that way, right because we're 0:15:15.156 --> 0:15:18.756 not using muscle and bone. We're using aluminum and you know, electricity, 0:15:18.796 --> 0:15:19.596 it's a whole different thing. 0:15:19.636 --> 0:15:21.956 In a way, it's surprising that it is right, like 0:15:21.996 --> 0:15:24.276 you would expect that it wouldn't look at all. 0:15:24.196 --> 0:15:26.316 Familiar, but there are clear differences. 0:15:26.596 --> 0:15:29.676 I suppose I'm projecting this. It's like, whatever is the 0:15:29.716 --> 0:15:34.356 Ostrich version of anthropomorphizing, right, I'm Ostrich pomorphizing? 0:15:34.596 --> 0:15:36.316 Yeah, you got it. And like you said, it's like 0:15:36.356 --> 0:15:38.436 a cartoon version of an Ostrich leg maybe. 0:15:38.556 --> 0:15:42.436 Yeah. Okay, So you've got this robot that is looks 0:15:42.476 --> 0:15:44.596 to my little brain like a pair of Ostrich legs. 0:15:44.636 --> 0:15:47.036 It's just a coincidence because it just turns out to 0:15:47.036 --> 0:15:49.036 be the best way to build a couple legs. And 0:15:49.076 --> 0:15:52.116 do you sell it to other academics? What do you 0:15:52.116 --> 0:15:52.516 do it? 0:15:52.756 --> 0:15:54.716 Yeah? We sold it to some of the top universities 0:15:54.716 --> 0:15:56.156 in the country and the kind in the world. 0:15:59.396 --> 0:16:02.436 So Cassie is a robot that academics can experiment with 0:16:02.636 --> 0:16:05.756 and learn from. But it doesn't have arms. It isn't 0:16:05.756 --> 0:16:09.556 built to do useful tasks. In a minute, Nathan and 0:16:09.636 --> 0:16:12.316 his colleagues build a robot that does have farms, that 0:16:12.476 --> 0:16:16.316 can do useful tasks, and that companies like Amazon are 0:16:16.356 --> 0:16:30.116 testing out in the real world right now. The latest 0:16:30.196 --> 0:16:34.036 robot from Jonathan's company is called Digit. It has two legs, 0:16:34.116 --> 0:16:37.316 two arms, It walks around, it picks stuff up, and 0:16:37.396 --> 0:16:40.436 it looks kind of like a person. But Jonathan says 0:16:40.556 --> 0:16:43.156 he and his colleagues didn't set out to build a 0:16:43.236 --> 0:16:44.516 robot that looks like a person. 0:16:44.916 --> 0:16:47.436 So yeah, so like, I'll take you down one of 0:16:47.476 --> 0:16:50.476 these thought processes that ended up looking like a person. Okay, 0:16:51.636 --> 0:16:54.796 We said, okay, we need to do some sort of 0:16:54.796 --> 0:16:58.556 inertial control of this thing because the robot can't turn 0:16:58.716 --> 0:17:01.036 very well. It's got little feet, and so when you 0:17:01.076 --> 0:17:04.196 try to turn aggressively, its skids right, okay, and if 0:17:04.196 --> 0:17:06.836 you look at any other bipads, this is one of 0:17:06.836 --> 0:17:09.356 the reasons wings evolved. It's because they're running in the stick, 0:17:09.396 --> 0:17:11.676 going out to catch the air to help them in 0:17:11.836 --> 0:17:14.636 maneuvering and turn it. You can go in a straight line, 0:17:14.876 --> 0:17:17.396 but figuring out how to maneuver quickly as hard when 0:17:17.396 --> 0:17:19.236 you've only got a little foot on the ground. You know, 0:17:19.676 --> 0:17:22.116 udropeds can really plant all four feet and twist and 0:17:22.156 --> 0:17:24.676 apply big torchs on the ground, and a biped not 0:17:24.716 --> 0:17:26.396 so much. You've got one foot at a time. How 0:17:26.396 --> 0:17:27.756 do you change your orientation? 0:17:27.916 --> 0:17:28.116 Right? 0:17:28.516 --> 0:17:31.036 So we looked at like putting a gyro on board 0:17:31.396 --> 0:17:35.076 reaction wheels or tails or things like that, because you know, 0:17:35.116 --> 0:17:37.076 we ruled out reaction wheels because that's just a big 0:17:37.116 --> 0:17:39.036 thing of brass that has to be mass you don't 0:17:39.036 --> 0:17:41.756 want in a robot like this. We thought about tails, 0:17:41.916 --> 0:17:45.596 and you look at any animals with tails, bipeds in particular. 0:17:46.116 --> 0:17:48.876 Typically that's to control the pitch. In other words, you'd 0:17:48.956 --> 0:17:51.476 leap off the ground, and then you want to reorient 0:17:51.556 --> 0:17:53.956 your body and your feet so that your feet land 0:17:54.196 --> 0:17:57.116 forward rather than you know, just tumbling in the air. Okay, 0:17:57.636 --> 0:17:59.276 but we don't want to control pitch. We want to 0:17:59.316 --> 0:18:02.116 control y'all. We want to steer the robot. So what 0:18:02.156 --> 0:18:04.476 we kind of came up with is that the best 0:18:04.516 --> 0:18:07.796 way to do that would be a pair of tails 0:18:07.836 --> 0:18:10.556 that are symmetrical on the front of the back or 0:18:10.596 --> 0:18:12.276 the side of the side of the robot, so that 0:18:12.316 --> 0:18:14.916 when you swing those tails, you're controlling exactly down the 0:18:14.916 --> 0:18:18.996 center line of your yaw, and so that just happens 0:18:19.036 --> 0:18:21.116 to be where our arms and shoulders are. You know, 0:18:21.676 --> 0:18:24.996 this bilaterally symmetrical pair of tails that can inertially actuate 0:18:25.076 --> 0:18:27.076 you around the center and allow you to steer. 0:18:27.836 --> 0:18:31.756 So that's the sort of intellectually elegant version of how 0:18:31.796 --> 0:18:35.796 you get to a robot that looks like it has arms. 0:18:35.956 --> 0:18:37.596 You're saying, in fact, in terms of the way you 0:18:37.636 --> 0:18:40.156 thought of it, it's a pair of tails that happened 0:18:40.156 --> 0:18:43.796 to sit where our arms sit. I mean, presumably there's 0:18:43.796 --> 0:18:45.676 a simpler version, which is, you want to build a 0:18:45.756 --> 0:18:47.796 robot that can do stuff in a world that is 0:18:47.836 --> 0:18:50.756 built for humans, and having arms would be useful in 0:18:50.796 --> 0:18:52.996 that context as well. Or no, was it true? 0:18:53.036 --> 0:18:55.516 Member? Yeah, what we're not trying to do is make 0:18:55.556 --> 0:18:57.476 a humanoid robot that looks like a person. What we're 0:18:57.476 --> 0:19:00.516 trying to do is on first principles, understand exactly why 0:19:00.516 --> 0:19:01.196 we do each thing. 0:19:01.516 --> 0:19:03.676 But are you really just adding the arm so that 0:19:03.756 --> 0:19:05.876 it can turn when it's walking, Like I feel like 0:19:05.956 --> 0:19:07.636 that's what you're saying, and I'm skeptical. 0:19:08.156 --> 0:19:11.196 Yeah. It's also that there were three other reasons why 0:19:11.236 --> 0:19:13.916 the arms should be there that were all aligned. They 0:19:13.916 --> 0:19:16.716 were not compromises where you know, putting the arms here 0:19:16.796 --> 0:19:18.396 is better for one thing or another. 0:19:18.676 --> 0:19:18.836 Huh. 0:19:19.076 --> 0:19:21.356 It's that. Hey, if you go on just the just 0:19:21.476 --> 0:19:24.716 the path of I want to improve my locomotion capability, 0:19:25.076 --> 0:19:26.996 you land at the solution of where the arms are. 0:19:27.356 --> 0:19:29.476 If you go down the path of this robot's going 0:19:29.516 --> 0:19:32.236 to fall, and we know that it can't just fall 0:19:32.276 --> 0:19:34.796 and land on its torso it'll break things. How do 0:19:34.876 --> 0:19:37.516 we put manipulator's arms something on it so that it's 0:19:37.516 --> 0:19:38.956 going to be able to catch itself when. 0:19:38.876 --> 0:19:41.196 It falls oriented? Okay. 0:19:41.676 --> 0:19:43.836 And then the third one, of course, is picking up 0:19:43.876 --> 0:19:46.636 things right manipulation in the world and being by manual 0:19:46.676 --> 0:19:49.796 in your manipulation so that you can basically a giant 0:19:49.836 --> 0:19:51.916 pincher grass. That's how you pick up boxes and tots 0:19:51.916 --> 0:19:53.916