Latest / What's Your Problem? / Building a Robot That Can Walk the Walk

Transcript

Building a Robot That Can Walk the Walk · What's Your Problem?

0:15
Speaker 1: Pushkin. So how long have you been trying to make
0:23
Speaker 1: a robot walk?
0:25
Speaker 2: It's been my entire career, starting from why I went
0:27
Speaker 2: to college in the first place.
0:29
Speaker 1: Why why that particular problem? Why is that your life's work?
0:35
Speaker 2: You know, there's few things more interesting and more dynamically
0:38
Speaker 2: complex and more elegant than the way animals move in
0:41
Speaker 2: the world. And to be able to get machines that
0:43
Speaker 2: can move that way, they can interact physically with the
0:46
Speaker 2: world the way humans and animals do. What a fun
0:50
Speaker 2: and interesting thing to work on for a career.
0:57
Speaker 1: I'm Jacob Goldstein and this is What's Your Problem? The
1:00
Speaker 1: show where I talk to people who are trying to
1:02
Speaker 1: make technological progress. My guest today is Jonathan Hurst. He's
1:06
Speaker 1: a professor at Oregon State University and founder and chief
1:10
Speaker 1: robot officer at Agility Robotics. Agility Robotics has built a
1:15
Speaker 1: robot called Digit. Digit looks kind of like a person.
1:19
Speaker 1: It walks around on two legs. It's got this flat,
1:21
Speaker 1: rectangular head, and it has two arms that it can
1:24
Speaker 1: use to pick stuff up. Jonathan's problem is this, how
1:28
Speaker 1: can you make a walking robot that can do useful
1:31
Speaker 1: work and that companies will actually pay for. Jonathan says
1:36
Speaker 1: that robot Digit is already being tested out in warehouses
1:40
Speaker 1: in the real world.
1:43
Speaker 2: We are deploying robots with cost We have two announced.
1:46
Speaker 2: We've announced a couple with Amazon and with GXO. You know,
1:50
Speaker 2: you place an order and Digit handles that order as
1:53
Speaker 2: part of the workflow that has happened is happening right now, and.
1:56
Speaker 1: So specifically, what are your robots doing well? First at Amazon, Yeah,
2:02
Speaker 1: the first use case or the first class of use
2:05
Speaker 1: cases that works for us is picking up these plastic
2:08
Speaker 1: tote plastic bins and putting them somewhere else. And then
2:12
Speaker 1: in warehouses, Yeah, but anywhere warehouses, in logistics, in manufacturing,
2:18
Speaker 1: you know, the whole environments like that that are a
2:20
Speaker 1: bit structured. They're kind of like there's these islands of
2:24
Speaker 1: automation they call them, you know, where one machine is
2:27
Speaker 1: putting things in a bin, another machine sorts bins and
2:30
Speaker 1: sends them different parts of the warehouse. Right now, sometimes
2:33
Speaker 1: a person will stand there, the robot will tell them
2:36
Speaker 1: which bin to pick up, and then all they are
2:38
Speaker 1: basically is a manipulator for the robots system. They pick
2:41
Speaker 1: up the bin and put it on the conveyor belt
2:42
Speaker 1: and wait for the robot to tell them the next
2:44
Speaker 1: thing to do. And it's really hard to hire people
2:46
Speaker 1: for that. There are a lot of open jobs in that,
2:50
Speaker 1: and so it's kind of a perfect place for Digit
2:52
Speaker 1: to walk in in this relatively structured first use case. Now,
2:57
Speaker 1: digits of course going to evolve towards picking up boxes, depalletizing,
3:01
Speaker 1: loading and unloading you know, tractor trailers and eventually getting
3:04
Speaker 1: out to things like retail and stocking shelves and you
3:08
Speaker 1: know sticks in hospitals carrying things around and eventually become
3:12
Speaker 1: a consumer product. So that's where you are today and
3:15
Speaker 1: where you want to get. I want to talk now
3:18
Speaker 1: about how you got here. Sure, and there's this really
3:24
Speaker 1: basic set of things you had to figure out just
3:28
Speaker 1: around how locomotion works, right, how people walk, also interestingly
3:33
Speaker 1: and sort of surprisingly, how how birds like ostriches walk.
3:37
Speaker 1: So I know there is this series of robots that
3:42
Speaker 1: you built on the way on the way to Digit, right,
3:45
Speaker 1: they were two before and then Digit and it seems
3:47
Speaker 1: like going through those and what you figured out the
3:50
Speaker 1: sort of key insight on each one is a really
3:53
Speaker 1: nice way to get a kind of deeper insight into
3:57
Speaker 1: how it works and what you had to understand to
3:59
Speaker 1: make a robot that can walk great, just being a
4:02
Speaker 1: really hard problem, right, Like there's like lots of robots
4:05
Speaker 1: that we don't even sort of think of as robots arms,
4:08
Speaker 1: and you know, self driving cars are arguably a kind
4:11
Speaker 1: of robot and whatever, but like getting a robot to
4:13
Speaker 1: walk is clearly a very hard problem.
4:16
Speaker 2: Yeah, Okay, So where I started in trying to understand
4:20
Speaker 2: how to make a machine work is on the biomechanics,
4:24
Speaker 2: try to understand how animals work. And a lot of
4:27
Speaker 2: biomechanics is about specific muscles and muscle groups and joints,
4:30
Speaker 2: and we're instead looking at holistically from a big picture,
4:34
Speaker 2: what is the center of mass of an animal doing?
4:36
Speaker 2: What are the forces happening on the ground. I did
4:38
Speaker 2: this collaboration with Monica Daily at the Royal Veterinary College
4:42
Speaker 2: and we looked at guinea fowl and ostriches and turkeys
4:45
Speaker 2: and you know, a whole bunch of different sizes.
4:47
Speaker 1: Why birds. It's really interesting to me that you did that.
4:51
Speaker 2: Why because humans are weird? How many other bipeds are there?
4:55
Speaker 2: You know, all the evolved bipeds, all of the extant
4:58
Speaker 2: existing bipeds in the world.
5:00
Speaker 1: Everything that walks on two legs. Yeah, huh.
5:03
Speaker 2: Those are all theropods. They're all kind of more like
5:07
Speaker 2: a bird than they are like or a monkey. They
5:10
Speaker 2: have evolved far longer than we have. You know, we've
5:14
Speaker 2: been out of trees for a very short period of time.
5:17
Speaker 2: Compared to all of the other existing bipeds in the world.
5:20
Speaker 2: They can all run much faster than we can for
5:22
Speaker 2: you know, very efficiently using less energy insul So.
5:27
Speaker 1: Like an ostrich is a better model for just abstract
5:31
Speaker 1: walking on two legs. Oh sure, I mean, yeah, I
5:34
Speaker 1: love that, But I want to be clear.
5:36
Speaker 2: What we're not trying to do is study how it
5:38
Speaker 2: does an Ostrich run versus how a human runs. What
5:40
Speaker 2: we're trying to do is study what is a fundamental
5:43
Speaker 2: truth between all animals in how they run, so we
5:46
Speaker 2: can try and weed out things that have to do
5:49
Speaker 2: with the size of the animal, or things that have
5:51
Speaker 2: to do with where exactly the ankle joint is or
5:54
Speaker 2: how long this joint is or that one. We want
5:56
Speaker 2: to know what is the same amongst all animals to
6:00
Speaker 2: walk and run.
6:01
Speaker 1: It's like the Platonic ideal of bipedalism. The sort of
6:04
Speaker 1: the fundrationing theory of it. Yeah, that's right.
6:07
Speaker 2: Yeah, And so biomechanists have been looking at this since
6:09
Speaker 2: the seventies and thinking about this in terms of spring
6:12
Speaker 2: mass models of locomotion. And it was only in like
6:14
Speaker 2: the two thousands sometimes that I think it was Hart
6:16
Speaker 2: McGuire and Andre Seifarth put together a paper that showed, hey,
6:19
Speaker 2: this spring mass model reproduces all of the behavior we
6:23
Speaker 2: see for walking and running and transitions between these gates.
6:27
Speaker 1: When you say, when you say spring mass model, that
6:30
Speaker 1: sounds big and exciting, But just to be clear, what
6:32
Speaker 1: do you mean when you say spring mass model.
6:35
Speaker 2: I mean a mathematical representation of a pogo stick. Go on, Okay,
6:41
Speaker 2: So a pogo stick is basically the simplest thing that
6:43
Speaker 2: can run, and a kangaroo looks a lot like a
6:46
Speaker 2: pogo sticky. Now, if you just stick a pogo stick
6:49
Speaker 2: on each leg, now you're bipedal running, you know.
6:52
Speaker 1: Okay.
6:53
Speaker 2: And then if you add a whole bunch of complexity
6:54
Speaker 2: to it, you have heel toe and you know, knee
6:57
Speaker 2: joints and all this other stuff. But if you really
6:59
Speaker 2: boil it down and try and make it as simple
7:01
Speaker 2: as possible, you get to some pretty basic math models
7:04
Speaker 2: that do represent how the progression of the center of
7:07
Speaker 2: mass of the animal moves, and how the ground reaction
7:10
Speaker 2: forces progress and so on.
7:11
Speaker 1: Right, Okay, so if I picture just like a lump
7:14
Speaker 1: of mass on top of two pogo sticks, you got it.
7:17
Speaker 1: I'm kind of in the right.
7:18
Speaker 2: Okay, you're absolutely okay. So that's a roughly a math
7:21
Speaker 2: model that at least gives you a very good concept
7:23
Speaker 2: of how do all animals run, horses, ghost, crabs, humans, ostriches, whatever.
7:28
Speaker 1: Right, So this paper comes out, this paper that says,
7:32
Speaker 1: think of a lump of mass on top of two
7:33
Speaker 1: pogo sticks. You you know about it because you're in
7:37
Speaker 1: this world? What do you What effect does it have
7:39
Speaker 1: on you? What do you do as a result of this.
7:42
Speaker 2: What I was looking at? And here's the argument at
7:44
Speaker 2: the time. Do these spring mass models simply seem to
7:50
Speaker 2: describe the things we're observing, you know? Or is it
7:53
Speaker 2: describing core physics of how it works? Like, in other words,
7:56
Speaker 2: if you build a spring mass model and a policy
7:58
Speaker 2: that works, is it going to make a robot stabilize?
8:01
Speaker 2: Or is it simply like a picture that kind of
8:04
Speaker 2: looks like what the animals are actually doing.
8:05
Speaker 1: So like if we actually do the lump of mass
8:08
Speaker 1: on top of two pot sticks as a robot, will
8:10
Speaker 1: it work?
8:11
Speaker 2: Yeah? And so look the question is, you know, how
8:13
Speaker 2: do you control these things? And then how does that
8:16
Speaker 2: translate into walking and running which had never really been
8:18
Speaker 2: done before that way, you haven't done this continuous transition
8:21
Speaker 2: between walking and running and changing the speed and everything else,
8:24
Speaker 2: and it's unknown how to stabilize that over all kinds
8:26
Speaker 2: of terrain. So that's why we built Atreus. That's why
8:29
Speaker 2: we built the robot Atreus.
8:30
Speaker 1: And so what is Atreus?
8:32
Speaker 2: So Atreus is it is a bipedal robot.
8:35
Speaker 1: Okay, what does Atrius look like?
8:38
Speaker 2: So Atrius was on the Late Night with Stephen Colbert
8:41
Speaker 2: and he described it as a microwave on stilts.
8:44
Speaker 1: Okay, that's good. So it doesn't it doesn't look humanoid
8:51
Speaker 1: at all. It looks maybe like a dancing alien or
8:53
Speaker 1: like a moon lander or something, but not humanoid.
8:58
Speaker 2: Yeah, absolutely not. So it doesn't look like an animal
9:00
Speaker 2: in any way. But it's designed entirely to be the
9:03
Speaker 2: math model of what we see an animal running. And
9:06
Speaker 2: the name Atrius is an acronym for assume the robot
9:09
Speaker 2: is a sphere, right. The whole idea is the robot
9:12
Speaker 2: is this simple math model.
9:14
Speaker 1: Where it's just some mass in the middle and then
9:16
Speaker 1: some very light springy legs.
9:18
Speaker 2: That's it. And so what we ultimately showed with Atreus
9:21
Speaker 2: it's the first robot ever to reproduce human walking gate dynamics.
9:25
Speaker 2: The robot walks across the force plate, a graduate student
9:28
Speaker 2: walks across the forest plate. Looking at the data, you
9:31
Speaker 2: can't actually tell a difference.
9:32
Speaker 1: Huh. So, like from the plate's point of view, it
9:35
Speaker 1: feels the same whether you're robot or a human is
9:38
Speaker 1: walking across.
9:38
Speaker 2: That's correct.
9:39
Speaker 1: So it's not true for earlier robots that looked like
9:41
Speaker 1: they were walking like humans.
9:43
Speaker 2: They would look the same. But if you looked at
9:44
Speaker 2: the dynamics of it, if you looked at the ground
9:47
Speaker 2: reaction forces, they differed quite a lot.
9:49
Speaker 1: Why is that important? Why is that important?
9:52
Speaker 2: It's just one symptom, right to show that, hey, we've
9:55
Speaker 2: actually captured the physics here. But the other symptom that's
9:58
Speaker 2: important is we are able to walk and run outdoors
10:01
Speaker 2: over all kinds of terrain without any sort of perception.
10:04
Speaker 2: The robot can handle amazing obstacles and it would just
10:07
Speaker 2: soak them up, you know, going over potholes, going from
10:10
Speaker 2: grass to pavement, going over big pieces of plywood we
10:13
Speaker 2: would throw in its way.
10:15
Speaker 1: And you're saying it didn't do this because it had
10:16
Speaker 1: like a clever brain. You're saying it was just the
10:18
Speaker 1: physics of the brainless machine was able to deal.
10:23
Speaker 2: No cameras on it, It had no awareness of the environment.
10:26
Speaker 2: It was a very simple spring mass model, very very
10:29
Speaker 2: simple control that did nothing but try to balance that.
10:32
Speaker 2: And it was able to just absorb all these kinds
10:34
Speaker 2: of disturbances and just keep on going.
10:36
Speaker 1: That sounds like a big deal. I agree.
10:39
Speaker 2: I'm very excited about this, right. That is the point
10:42
Speaker 2: that we decided we were going to found Agility Robotics.
10:45
Speaker 2: We said, you know, this was a mission for like
10:47
Speaker 2: years and years and years. That is why I became
10:50
Speaker 2: a professor, is to say, my goal here in academia
10:55
Speaker 2: is to show that this spring mass physics is real
10:59
Speaker 2: and really make sure we understand that well. And if
11:02
Speaker 2: we can show that and prove that and with this
11:04
Speaker 2: ATREUS project, we then can take the next steps. Right,
11:08
Speaker 2: So that's what we did. It was a scientific kind
11:12
Speaker 2: of breakthrough. But the machine could only walk and run.
11:15
Speaker 2: It couldn't stand, it couldn't turn. You know, it breaks often.
11:20
Speaker 2: If it ever fell, it would just be completely destroyed.
11:23
Speaker 2: So it's not a productive, useful machine. It's a science demonstrator.
11:27
Speaker 1: Okay, So you have a TRIOS that is academically it
11:33
Speaker 1: kind of intellectually a breakthrough, right, but nobody's going to
11:37
Speaker 1: buy it to do anything. It's not useful in a
11:40
Speaker 1: practical sense. What do you do next?
11:44
Speaker 2: Okay? So at that point we say, hey, you know,
11:47
Speaker 2: we understand a really significant portion of what does it
11:51
Speaker 2: take to make a robot that can go where people go?
11:54
Speaker 2: And what an opportunity? This is? What are all the
11:56
Speaker 2: things that robots are going to be able to do
11:58
Speaker 2: when they can be coexisting with humans? Right? And the
12:03
Speaker 2: barrier to moving forward on that now is more about
12:06
Speaker 2: execution on engineering and execution on building a use case
12:11
Speaker 2: in a business around it.
12:12
Speaker 1: Because you feel like you've solved the sort of core
12:15
Speaker 1: threshold technical problem of a robot that can walk in
12:19
Speaker 1: unfamiliar environments and not fall down.
12:21
Speaker 2: Correct, we have the foundation layer of leged locomotion. Now,
12:25
Speaker 2: sure we'd like to add on perception to it so
12:27
Speaker 2: that it can handle stairs effectively and things like that
12:30
Speaker 2: and intentionally handle obstacles. We need to do the engineering
12:33
Speaker 2: of the electrical system and the hardware system that still
12:35
Speaker 2: captures the same physics but can take a beating, can stand,
12:38
Speaker 2: can steer, can start to do things right. So we
12:41
Speaker 2: know we're going to need to build a robot with legs,
12:43
Speaker 2: with manipulators, with sensors. Okay, because so we're kind of
12:47
Speaker 2: going down a path now where we want to take
12:49
Speaker 2: the exact same first principles approach to how do we
12:54
Speaker 2: build a machine that can manipulate things in a human
12:56
Speaker 2: world and get around in it and interact with people.
13:00
Speaker 2: Build a human centric machine. So the first step to
13:03
Speaker 2: doing that was designing a robot that we could sell
13:06
Speaker 2: to other researchers to continue the work on the leg
13:09
Speaker 2: leggs as we then worked towards you know, arms and manipulators,
13:12
Speaker 2: and that was Cassie, our first robot that Agilia Robotics sold.
13:17
Speaker 2: Cassie added the ability to stand in place because it
13:20
Speaker 2: had ankles, and it had the ability to steer because
13:23
Speaker 2: you could turn the legs. But more importantly, it was
13:26
Speaker 2: much more compact and extremely robust, so this robot can
13:30
Speaker 2: fall hard on concrete and you just pick it back
13:33
Speaker 2: up and it can get going again.
13:34
Speaker 1: And so, Cassie, I'm looking at a picture of it now.
13:37
Speaker 1: It basically looks like a pair of Ostrich legs. Yeah,
13:42
Speaker 1: it looks like an Ostrich. Does an Ostrich have a wasist?
13:45
Speaker 1: I don't know an ostrich from the waist down, No.
13:47
Speaker 2: It doesn't know. Yeah, the pelvis is stationary and a
13:51
Speaker 2: bird fixed rather than a human pelvis, which is mobile.
13:54
Speaker 1: So I know it's not technically correct to say that
13:57
Speaker 1: an Ostrich legs bend backwards, but it looks that way, right.
14:00
Speaker 2: Yeah, Their thigh is very short, their knee is up
14:03
Speaker 2: next to their body, and what you perceive as their
14:05
Speaker 2: knee is actually their ankle.
14:07
Speaker 1: In designing this robot, how do you get to legs
14:11
Speaker 1: that look like Ostrich legs? Like, it's like convergent evolution,
14:15
Speaker 1: you know, convergent evolution maybe hopefully. Isn't it the case
14:18
Speaker 1: that like cephalopods that like whatever squids have eyes like
14:23
Speaker 1: our eyes, but they evolve totally independently. Is this like that?
14:26
Speaker 2: There's a lot of examples of convergent evolution. We can
14:29
Speaker 2: only guess, right because we don't necessarily know, and a
14:31
Speaker 2: scientists only hypothesize the evolutionary pressures that cause animals to
14:35
Speaker 2: be the shape that they are.
14:36
Speaker 1: But the pressure that led you you didn't say let's
14:38
Speaker 1: make legs that look like Ostrich eggs. You just did
14:41
Speaker 1: a bunch of math and you wound up with legs
14:42
Speaker 1: that look like Ostrich legs.
14:44
Speaker 2: That is correct, and there are a bunch of features
14:46
Speaker 2: on our robot that have gone down a similar path.
14:49
Speaker 2: And I actually love that because when we end up
14:52
Speaker 2: you know, blank sheet, pursue all of the different configuration
14:55
Speaker 2: options and say okay, here's what we think is the
14:57
Speaker 2: optimal choice, and we say, wow, that looks just like
15:00
Speaker 2: a person, or that just looks like a bird or something.
15:02
Speaker 2: It's actually really good. It means we're probably on the
15:04
Speaker 2: right path.
15:05
Speaker 1: Yeah, it's it's exciting in a way, right, like like
15:09
Speaker 1: you do a bunch of bath and then suddenly you
15:11
Speaker 1: look up and you see an Ostrich.
15:13
Speaker 2: But it won't always be that way, right because we're
15:15
Speaker 2: not using muscle and bone. We're using aluminum and you know, electricity,
15:18
Speaker 2: it's a whole different thing.
15:19
Speaker 1: In a way, it's surprising that it is right, like
15:21
Speaker 1: you would expect that it wouldn't look at all.
15:24
Speaker 2: Familiar, but there are clear differences.
15:26
Speaker 1: I suppose I'm projecting this. It's like, whatever is the
15:29
Speaker 1: Ostrich version of anthropomorphizing, right, I'm Ostrich pomorphizing?
15:34
Speaker 2: Yeah, you got it. And like you said, it's like
15:36
Speaker 2: a cartoon version of an Ostrich leg maybe.
15:38
Speaker 1: Yeah. Okay, So you've got this robot that is looks
15:42
Speaker 1: to my little brain like a pair of Ostrich legs.
15:44
Speaker 1: It's just a coincidence because it just turns out to
15:47
Speaker 1: be the best way to build a couple legs. And
15:49
Speaker 1: do you sell it to other academics? What do you
15:52
Speaker 1: do it?
15:52
Speaker 2: Yeah? We sold it to some of the top universities
15:54
Speaker 2: in the country and the kind in the world.
15:59
Speaker 1: So Cassie is a robot that academics can experiment with
16:02
Speaker 1: and learn from. But it doesn't have arms. It isn't
16:05
Speaker 1: built to do useful tasks. In a minute, Nathan and
16:09
Speaker 1: his colleagues build a robot that does have farms, that
16:12
Speaker 1: can do useful tasks, and that companies like Amazon are
16:16
Speaker 1: testing out in the real world right now. The latest
16:30
Speaker 1: robot from Jonathan's company is called Digit. It has two legs,
16:34
Speaker 1: two arms, It walks around, it picks stuff up, and
16:37
Speaker 1: it looks kind of like a person. But Jonathan says
16:40
Speaker 1: he and his colleagues didn't set out to build a
16:43
Speaker 1: robot that looks like a person.
16:44
Speaker 2: So yeah, so like, I'll take you down one of
16:47
Speaker 2: these thought processes that ended up looking like a person. Okay,
16:51
Speaker 2: We said, okay, we need to do some sort of
16:54
Speaker 2: inertial control of this thing because the robot can't turn
16:58
Speaker 2: very well. It's got little feet, and so when you
17:01
Speaker 2: try to turn aggressively, its skids right, okay, and if
17:04
Speaker 2: you look at any other bipads, this is one of
17:06
Speaker 2: the reasons wings evolved. It's because they're running in the stick,
17:09
Speaker 2: going out to catch the air to help them in
17:11
Speaker 2: maneuvering and turn it. You can go in a straight line,
17:14
Speaker 2: but figuring out how to maneuver quickly as hard when
17:17
Speaker 2: you've only got a little foot on the ground. You know,
17:19
Speaker 2: udropeds can really plant all four feet and twist and
17:22
Speaker 2: apply big torchs on the ground, and a biped not
17:24
Speaker 2: so much. You've got one foot at a time. How
17:26
Speaker 2: do you change your orientation?
17:27
Speaker 1: Right?
17:28
Speaker 2: So we looked at like putting a gyro on board
17:31
Speaker 2: reaction wheels or tails or things like that, because you know,
17:35
Speaker 2: we ruled out reaction wheels because that's just a big
17:37
Speaker 2: thing of brass that has to be mass you don't
17:39
Speaker 2: want in a robot like this. We thought about tails,
17:41
Speaker 2: and you look at any animals with tails, bipeds in particular.
17:46
Speaker 2: Typically that's to control the pitch. In other words, you'd
17:48
Speaker 2: leap off the ground, and then you want to reorient
17:51
Speaker 2: your body and your feet so that your feet land
17:54
Speaker 2: forward rather than you know, just tumbling in the air. Okay,
17:57
Speaker 2: but we don't want to control pitch. We want to
17:59
Speaker 2: control y'all. We want to steer the robot. So what
18:02
Speaker 2: we kind of came up with is that the best
18:04
Speaker 2: way to do that would be a pair of tails
18:07
Speaker 2: that are symmetrical on the front of the back or
18:10
Speaker 2: the side of the side of the robot, so that
18:12
Speaker 2: when you swing those tails, you're controlling exactly down the
18:14
Speaker 2: center line of your yaw, and so that just happens
18:19
Speaker 2: to be where our arms and shoulders are. You know,
18:21
Speaker 2: this bilaterally symmetrical pair of tails that can inertially actuate
18:25
Speaker 2: you around the center and allow you to steer.
18:27
Speaker 1: So that's the sort of intellectually elegant version of how
18:31
Speaker 1: you get to a robot that looks like it has arms.
18:35
Speaker 1: You're saying, in fact, in terms of the way you
18:37
Speaker 1: thought of it, it's a pair of tails that happened
18:40
Speaker 1: to sit where our arms sit. I mean, presumably there's
18:43
Speaker 1: a simpler version, which is, you want to build a
18:45
Speaker 1: robot that can do stuff in a world that is
18:47
Speaker 1: built for humans, and having arms would be useful in
18:50
Speaker 1: that context as well. Or no, was it true?
18:53
Speaker 2: Member? Yeah, what we're not trying to do is make
18:55
Speaker 2: a humanoid robot that looks like a person. What we're
18:57
Speaker 2: trying to do is on first principles, understand exactly why
19:00
Speaker 2: we do each thing.
19:01
Speaker 1: But are you really just adding the arm so that
19:03
Speaker 1: it can turn when it's walking, Like I feel like
19:05
Speaker 1: that's what you're saying, and I'm skeptical.
19:08
Speaker 2: Yeah. It's also that there were three other reasons why
19:11
Speaker 2: the arms should be there that were all aligned. They
19:13
Speaker 2: were not compromises where you know, putting the arms here
19:16
Speaker 2: is better for one thing or another.
19:18
Speaker 1: Huh.
19:19
Speaker 2: It's that. Hey, if you go on just the just
19:21
Speaker 2: the path of I want to improve my locomotion capability,
19:25
Speaker 2: you land at the solution of where the arms are.
19:27
Speaker 2: If you go down the path of this robot's going
19:29
Speaker 2: to fall, and we know that it can't just fall
19:32
Speaker 2: and land on its torso it'll break things. How do
19:34
Speaker 2: we put manipulator's arms something on it so that it's
19:37
Speaker 2: going to be able to catch itself when.
19:38
Speaker 1: It falls oriented? Okay.
19:41
Speaker 2: And then the third one, of course, is picking up
19:43
Speaker 2: things right manipulation in the world and being by manual
19:46
Speaker 2: in your manipulation so that you can basically a giant
19:49
Speaker 2: pincher grass. That's how you pick up boxes and tots
19:51
Speaker 2: and all these things you want to move. That's also
19:53
Speaker 2: the best place for them. So basically, you just set
19:57
Speaker 2: out to build a machine that could go where humans
20:00
Speaker 2: go and pick up things that are the size that
20:03
Speaker 2: humans pick up, and from first principles, with your eyes closed,
20:07
Speaker 2: you wound up with a thing that looks like a guy.
20:10
Speaker 1: Absolutely so okay, so this is how you get to
20:13
Speaker 1: digit the robot that you're now building and selling to people.
20:17
Speaker 2: That's right. So now we're taking this transition right now
20:20
Speaker 2: as a company from that very intellectual and first principles
20:23
Speaker 2: approach that I shared with you to now working with
20:26
Speaker 2: the customer understanding what their use case is, writing down
20:29
Speaker 2: the sets of requirements, like you know, the temperature ranges,
20:33
Speaker 2: the you know, weights of all the things that you're
20:35
Speaker 2: going to be able to pick up, the safety requirements,
20:39
Speaker 2: you know. And it's a massive list, hundreds of things
20:41
Speaker 2: in a list to write down the requirements documents so
20:44
Speaker 2: that we can engineer a system that is a product. Yeah,
20:47
Speaker 2: very different from designing a robot that can do cool things.
20:50
Speaker 2: We're engineering a product, and that's the pivot that our
20:53
Speaker 2: company is in right the second.
20:54
Speaker 1: And like I imagine for you personally, that must be
20:56
Speaker 1: a significant shift, right if you spent whatever twenty some
20:59
Speaker 1: years in the kind of abstract academic world of like
21:06
Speaker 1: let's build the thing that works and know think deep
21:09
Speaker 1: thoughts to like let's mass produce a product that people
21:14
Speaker 1: will pay us for. That's quite different.
21:15
Speaker 2: Oh, it's fundamentally different. It's a whole different way of thinking.
21:19
Speaker 2: In fact, I changed my title to chief robot Officer, right,
21:23
Speaker 2: and we hired Melanie Wise as our new chief technology officer.
21:27
Speaker 2: She comes out of FET Robotics. She was the founder
21:30
Speaker 2: there and recently sold that company and they were deploying
21:33
Speaker 2: thousands of robots in logistics squarehouses. And she is an
21:36
Speaker 2: absolute expert on understanding customers and product and creating a product.
21:40
Speaker 2: And what we've done is we've shifted our organization. So
21:43
Speaker 2: you know, Melanie is in charge span of that whole
21:46
Speaker 2: product side of the organization and the engineering to make
21:49
Speaker 2: a product. I'm now leading the innovation team.
21:52
Speaker 1: So you get to keep doing kind of the stuff
21:55
Speaker 1: you've been doing.
21:56
Speaker 2: The things I'm good at.
21:57
Speaker 1: Yeah, so what is the frontier on the innovation side.
22:00
Speaker 1: What are you trying to figure out next?
22:02
Speaker 2: It is fundamentally hardware that enables the kinds of physics
22:07
Speaker 2: that we want to achieve, powered by some of these
22:10
Speaker 2: new AI tools. You know, we're getting to a point
22:13
Speaker 2: now where some of these tools will allow us to
22:16
Speaker 2: create behaviors and create things that as an engineer we
22:20
Speaker 2: don't know how to model. Huh, And that's super interesting.
22:24
Speaker 2: So instead you're describing the symptoms of it, and then
22:27
Speaker 2: the system, the learning system, figures out how to make
22:30
Speaker 2: that happen.
22:31
Speaker 1: Amazingly different than what you've been describing. You've been describing
22:34
Speaker 1: of like, let's think of you know, first principles, just
22:37
Speaker 1: the physics of the universe, and from that build a machine.
22:41
Speaker 1: And now you're talking about you know, an era when
22:45
Speaker 1: possibly you'll be able to ignore all of that and
22:47
Speaker 1: say to the AI, you figure it out, here's what
22:49
Speaker 1: I want to do.
22:50
Speaker 2: Well, let me put some caveats on that.
22:52
Speaker 1: Yeah, that sounds ridiculous when I say it that way.
22:55
Speaker 2: Well, remember that the AI has to operate on a
22:56
Speaker 2: piece of hardware. Yeah right, And so that piece of
22:59
Speaker 2: hardware we still have to engineer and design to be
23:02
Speaker 2: able to achieve the physics that we want to achieve.
23:05
Speaker 1: Though you could say to an AI, here's what we
23:07
Speaker 1: want to do. What should the hardware look like? Maybe
23:10
Speaker 1: in one hundred years you think one hundred, one hundred,
23:13
Speaker 1: who knows one hundred.
23:14
Speaker 2: It's fine, fine, three years, you know, future, Not today.
23:18
Speaker 1: Not certainly not. So what specifically are you doing today?
23:21
Speaker 1: Are you taking this robot that you have digit and
23:23
Speaker 1: seeing if you sort of put an AI layer in it,
23:26
Speaker 1: on it near it? What can you do?
23:27
Speaker 2: Is that what's happening all of the above, So, you know,
23:30
Speaker 2: on the hardware side, So there's a lot on the
23:32
Speaker 2: hardware you do just to make it even possible for
23:34
Speaker 2: the AI to learn. We're building then a whole architecture
23:38
Speaker 2: of a digital twin so that you can learn things
23:40
Speaker 2: in simulation first, and then you know, transfer from sim
23:42
Speaker 2: to reel.
23:43
Speaker 1: A digital twin is basically making a version of the
23:46
Speaker 1: robot that exists as software that exists virtually.
23:49
Speaker 2: Version of the environment as well that the robot operates in,
23:52
Speaker 2: so that everything can be done, you know, decades of
23:55
Speaker 2: experience on the robot can be done in hours of
23:58
Speaker 2: time through the you know, et cetera.
24:00
Speaker 1: So the digital twin is allowing you to try and
24:03
Speaker 1: generate data to train the AI. Is that that's what's the.
24:06
Speaker 2: Source of the data, right, A lot of language models
24:08
Speaker 2: and things like that are based on data from the internet. Well,
24:11
Speaker 2: I don't think that that's feasible for robot control because
24:14
Speaker 2: the physics of the hardware is so unique. So even
24:16
Speaker 2: if you're trying to teleoperate this thing, you've got this
24:19
Speaker 2: weird translation between what a person would do to then
24:22
Speaker 2: the robot trying to mimic that, which is probably not
24:25
Speaker 2: the dime of the right.
24:25
Speaker 1: Dynamic robot doesn't actually walk like a person, even if
24:28
Speaker 1: it looks like it's.
24:29
Speaker 2: Right, that's right, everything's sort of different internally about it.
24:32
Speaker 2: How it would control itself.
24:34
Speaker 1: What what what are you worried about? Like you what
24:40
Speaker 1: could go wrong and how are you trying to get
24:42
Speaker 1: it not to go wrong?
24:43
Speaker 2: So before I talk about what I'm worried about, let
24:45
Speaker 2: me tell you what I'm excited about. Fair We had
24:47
Speaker 2: ten of these Cassie robots out in the world, and
24:49
Speaker 2: so researchers all over the place for for years and
24:51
Speaker 2: years are working on various kinds of controllers. When we
24:54
Speaker 2: were able to successfully get a learned policy working on Cassie,
24:58
Speaker 2: were to run a five k across campus, we were
25:01
Speaker 2: able to the world record in the one hundred meter dash.
25:04
Speaker 2: We were able to.
25:05
Speaker 1: Do for a robot. Yes, and when you say learned,
25:08
Speaker 1: you mean developed with machine learning as opposed to in
25:11
Speaker 1: the old way. Is that what you mean?
25:12
Speaker 2: Correct? It's an entirely machine learned policy that was learned
25:15
Speaker 2: in simulation and then put on the thing.
25:18
Speaker 1: What's a definition of a policy? As you're using the word, I.
25:22
Speaker 2: Mean, a policy is just a bunch of math that
25:27
Speaker 2: takes as input all of the sensors and then spits
25:30
Speaker 2: out numbers that describe the torques you should apply to
25:33
Speaker 2: the motors.
25:34
Speaker 1: Okay, so it's sort of if this happens in the world, robot,
25:37
Speaker 1: you should do this set of things.
25:40
Speaker 2: Yeah, you're based on you know, but you know it's
25:42
Speaker 2: like the input from thirty encoders or one hundred different sensors,
25:46
Speaker 2: all of that complex input.
25:47
Speaker 1: But if this is complicated and the then that are complicated, Yeah.
25:51
Speaker 2: No, you're rreat you're right about that. Yeah, it's an equation.
25:53
Speaker 1: Good. So the machine learning basically made the robot work
25:57
Speaker 1: way better.
25:58
Speaker 2: It made the robot work much better, and even more importantly,
26:02
Speaker 2: once the pipeline was in place, we can learn new
26:06
Speaker 2: skills and learn new policies much faster with much less
26:09
Speaker 2: engineer time. How we can get there faster and have
26:13
Speaker 2: higher performance using learning approaches to control.
26:16
Speaker 1: It's like a productivity like supercharger. It just makes everything
26:19
Speaker 1: go much faster and more efficiently.
26:21
Speaker 2: Absolutely, it's a new tool. It's amazing, that's right. Okay,
26:25
Speaker 2: So what am I worried about? Right? What am I
26:29
Speaker 2: worried about? I'm worried that this is one of those
26:32
Speaker 2: kind of black Swan events. Right, this is one of
26:34
Speaker 2: those things that changes everything, and everybody doesn't exactly know
26:38
Speaker 2: what all the implications are yet and what are the
26:41
Speaker 2: you know, the right paths forward, and so everybody's trying everything.
26:45
Speaker 1: And this being basically a useful bipedal robot like it,
26:50
Speaker 1: it could be hugely important in ways that we don't understand,
26:53
Speaker 1: and there could be unintended consequences.
26:55
Speaker 2: What I actually mean is that this new realization that
26:59
Speaker 2: we can use learning policies to control dynamic robots and machines, yeah,
27:06
Speaker 2: means that the entire way it all robotics controls people
27:11
Speaker 2: have been doing robot controls before is not as relevant.
27:17
Speaker 2: And this new tool that nobody really understands that well
27:20
Speaker 2: yet is clearly the future of how it's going to work.
27:25
Speaker 1: So what are I mean? I understand that if it's
27:29
Speaker 1: really a black swan, you don't know what's going to happen,
27:31
Speaker 1: because if you knew, it wouldn't be a black swan,
27:33
Speaker 1: But like, what are you thinking of? What could it mean?
27:37
Speaker 1: You know, plainly, bipedal robots are a very powerful tool,
27:41
Speaker 1: and you could imagine malevolent uses of them, right, I guess, So,
27:46
Speaker 1: I guess we've already got drones. Right, It doesn't matter
27:48
Speaker 1: whether the robot that kills you looks like a dude, right,
27:51
Speaker 1: those things already exist.
27:52
Speaker 2: In fact, a humanoid robot is probably the least effective
27:54
Speaker 2: way to do that I think, honestly, my biggest worry
27:58
Speaker 2: about AI in general has a lot more to do
28:00
Speaker 2: with its ability to influence people, its ability to model
28:03
Speaker 2: people's feelings and that kind of thing.
28:06
Speaker 1: So that's more like using large language model for kind
28:09
Speaker 1: of personalized misinformation or that.
28:11
Speaker 2: Stat it that's the biggest threat building robots that are
28:15
Speaker 2: going to like take people. I don't know, I just
28:17
Speaker 2: don't see it fair.
28:19
Speaker 1: I mean, I'm kind of tired of talking about technological
28:22
Speaker 1: unemployment because the robot looks like a person. It makes
28:25
Speaker 1: me feel like we should touch on it. Do you
28:26
Speaker 1: want to just speak for a moment to the prospect
28:28
Speaker 1: of technological unemployment?
28:30
Speaker 2: Sure, I'll say our entire business model is centered around
28:33
Speaker 2: the number of unfilled roles that exist in the logistics environment.
28:37
Speaker 2: It's not centered around how it's going to be less
28:39
Speaker 2: expensive than human Labor's centered around how they actually in
28:44
Speaker 2: geographic locations do not have enough people to provide the
28:48
Speaker 2: service that they're providing, you know, the way they're doing
28:51
Speaker 2: it now, there's no way forward for improved logistics and
28:55
Speaker 2: getting your things in one day and you know, all
28:58
Speaker 2: the stuff that people really really want. There's no path
29:00
Speaker 2: forward to do that with more human labor doing it,
29:03
Speaker 2: it must be automated in a significant way. So you know,
29:07
Speaker 2: that's our whole business model is based on unfilled roles.
29:11
Speaker 1: What is your like happy version of the future in
29:14
Speaker 1: whatever number of years seems like the right number of.
29:17
Speaker 2: Man These robots are actually safe and smart enough in
29:22
Speaker 2: order to do a lot of useful things in the world.
29:23
Speaker 2: And the relationship with people is kind of like a
29:25
Speaker 2: service animal. And you know, these robots are everywhere and
29:29
Speaker 2: delivering all the packages to your door, and you know,
29:32
Speaker 2: being a telepresence device that you can easily log into
29:34
Speaker 2: and keep in touch with people, and you know, in
29:38
Speaker 2: a lot of warehousing and stocking and you know, doing
29:41
Speaker 2: the dull, dirty, dangerous to classic three d's of robotics.
29:45
Speaker 2: It's we've always wanted that. It's all about improving the
29:49
Speaker 2: quality of life. It's all about letting enabling humans to
29:52
Speaker 2: be more human, letting people do the things that are
29:55
Speaker 2: that they want to do that involve the social interaction
29:58
Speaker 2: and the creativity and the variety that people are so
30:00
Speaker 2: good at, and having robots that can pick up all
30:02
Speaker 2: of the tasks that we'd rather not do, and having
30:06
Speaker 2: robots that can be in environments that are designed around us.
30:09
Speaker 2: Is a really important step to being able to achieve
30:12
Speaker 2: that in a really great way.
30:14
Speaker 1: Is there anything else you want to talk about?
30:17
Speaker 2: MM. One thing I want to make sure that we
30:22
Speaker 2: get across is kind of the clear argument that the
30:26
Speaker 2: fastest path to general purpose machine that does use full
30:30
Speaker 2: work in human spaces is to do one thing first
30:33
Speaker 2: and then the second thing, and to do it for
30:35
Speaker 2: customers and to get it deployed and to figure out
30:38
Speaker 2: the reliability and the safety and so on. That's the path.
30:41
Speaker 2: There is no good way to just like jump to
30:43
Speaker 2: the answer.
30:44
Speaker 1: So basically you're saying, you can't just build a robot
30:46
Speaker 1: that does everything. You have to build a robot that
30:48
Speaker 1: does one thing and then figure out how to make
30:51
Speaker 1: it to a second thing.
30:52
Speaker 2: Yeah, but you want to stay on that vision. You
30:53
Speaker 2: want to have your guess of what the everything robot
30:55
Speaker 2: looks like, but you know you're going to be wrong,
30:58
Speaker 2: and so then you start on what's the match to
31:00
Speaker 2: the first thing that it should be able to do,
31:03
Speaker 2: and then you keep on revising and iterating on that path.
31:06
Speaker 2: So I'll give an example. This This is through our
31:08
Speaker 2: entire est then of building the function first and the
31:11
Speaker 2: physics first and the first principles approach to figuring out
31:13
Speaker 2: things like the legs. Right, yeah, I want to point out,
31:16
Speaker 2: like hands, how do you produce a dextrous manipulator? By dexterous,
31:20
Speaker 2: I mean something that can pick up pens and pick
31:22
Speaker 2: up objects and do useful things with the hand, open doorknobs,
31:25
Speaker 2: all the stuff. Right, I don't care how it looks,
31:28
Speaker 2: I care what it does. So what we're doing for
31:32
Speaker 2: our first use case is picking up totes that can
31:35
Speaker 2: have a twenty kilogram bowling ball rolling around in it.
31:38
Speaker 2: That's a very hard thing to pick up. And so
31:41
Speaker 2: now our grippers are these big graspers that can grasp
31:43
Speaker 2: the side of this tote and pick it up. A
31:46
Speaker 2: lot of groups are sort of have these five figured
31:48
Speaker 2: hands on their robots which look like a human's hand,
31:53
Speaker 2: but certainly couldn't pick up those twenty kilogram totes.
31:55
Speaker 1: What do you have just done? Pincer sort of basically yeah.
31:58
Speaker 2: You know, effectively it's a big sort of four fingered
32:01
Speaker 2: pincher thing that can have leverage on it and grasp
32:04
Speaker 2: the sides of the tote.
32:05
Speaker 1: I mean does that mean that it can't do sort
32:07
Speaker 1: of fine motor things. The trade off that you're making this.
32:10
Speaker 2: I would say the current pincher design, Yeah, it's not
32:13
Speaker 2: designed to pick up small objects. But you know, we
32:15
Speaker 2: see a vision where that's actually a tool, not a hand.
32:19
Speaker 2: And just like people use tools, robots are gonna use tools.
32:22
Speaker 2: But there's such an opportunity to have the tool be
32:26
Speaker 2: attached at the wrist or the fingers or the elbow
32:29
Speaker 2: or the forearm or wherever. And we don't know exactly
32:31
Speaker 2: how that should be, and nobody does. But one thing
32:34
Speaker 2: I can say with confidence is that these five fingered
32:38
Speaker 2: hands cannot do dexterous manipulation. It's not just a controls problem.
32:42
Speaker 2: They're making the same mistake of creating something that has
32:47
Speaker 2: the same morphology. It looks like a person. It looks
32:50
Speaker 2: like a person, but that doesn't mean it can apply
32:52
Speaker 2: the right forces, or have the right kinematics, or have
32:55
Speaker 2: the right dynamics or the right compliance or anything that.
32:58
Speaker 2: We don't know how to do that. It's one of
32:59
Speaker 2: the grand challenges in robotics. So the fastest path to
33:02
Speaker 2: get there is to start manipulating things and do stuff
33:05
Speaker 2: that has metrics right measures, a.
33:09
Speaker 1: Job that someone is actually willing to pay for that's right.
33:11
Speaker 1: So right now it's moving tots around as basically, what's
33:15
Speaker 1: the second job?
33:16
Speaker 2: Boxes? Cardboard boxes, huh? And then the third job is
33:19
Speaker 2: starting we want to start doing each picking, you know,
33:22
Speaker 2: picking up things and putting them in the boxes and
33:24
Speaker 2: in the So those are much.
33:25
Speaker 1: Smaller things, different than picking up a big punt.
33:30
Speaker 2: That's right. And so the question is, and nobody knows
33:32
Speaker 2: the answer to this, is that two different tools or
33:35
Speaker 2: is that one general purpose manipulator that can do both
33:38
Speaker 2: of those things.
33:38
Speaker 1: They need to put a different hand on the robot
33:40
Speaker 1: to pick up a little thing versus pick up a
33:42
Speaker 1: big sure, or is it optimal?
33:44
Speaker 2: And you know this will be the this gets into
33:46
Speaker 2: customer requirements question rather than fundamental science question. Yeah, that's
33:51
Speaker 2: your that's for your product person to figure out. Yeah,
33:53
Speaker 2: but that's how this is going to evolve. That's how
33:55
Speaker 2: we're going to get to dexterous manipulation in the world
33:58
Speaker 2: in a way that really is meaningful.
33:59
Speaker 1: Investors, one job at a ton, Yeah.
34:01
Speaker 2: Figure out if you've got the first like you know,
34:05
Speaker 2: the first four or five jobs, and now you've got
34:07
Speaker 2: your requirements, yours of requirements and your measurements and your
34:11
Speaker 2: metrics and now engineers are going to be able to
34:13
Speaker 2: iterate on that really make something work. But just trying
34:16
Speaker 2: to copy a five figure in hand and say now
34:17
Speaker 2: now it's an AI problem completely false, that's just not
34:20
Speaker 2: going to happen.
34:24
Speaker 1: We'll be back in a minute with the lightning round.
34:38
Speaker 1: Let's finish with the lightning round. I've read that you
34:42
Speaker 1: like to jog at night for work, but like, are
34:45
Speaker 1: there specific things that have happened to you or that
34:47
Speaker 1: you've done to try and sort of you know, put
34:50
Speaker 1: yourself in hard walking or running settings and where you've
34:54
Speaker 1: actually had a thing happened and thought, oh, we need
34:57
Speaker 1: to make sure the robot doesn't fall over when X happens.
35:00
Speaker 2: Honestly, it's more like when you're taking a hike, you know,
35:03
Speaker 2: and you kind of get into the mode of you're
35:05
Speaker 2: just your long day and a long hike and you know,
35:08
Speaker 2: watching and being mindful, I guess, and thinking about what
35:12
Speaker 2: are my feet doing and how is the contact progressing
35:16
Speaker 2: with the ground and how does that feel and why
35:18
Speaker 2: is that happening? And sort of daydreaming while you're thinking
35:21
Speaker 2: that through is a really good way then to start
35:24
Speaker 2: to recognize connections to research papers and connections to things
35:29
Speaker 2: that people have found scientifically, and then start to pull
35:32
Speaker 2: together hypotheses about how you would implement something like this
35:35
Speaker 2: on a robot, or what is necessary because we're human,
35:39
Speaker 2: and what is necessary because it's fundamental to look emotion
35:43
Speaker 2: like why do we have feet? That's very different from
35:45
Speaker 2: bird feet atreus didn't have feet, you know, And then
35:49
Speaker 2: what exactly do we get anyway? Thinking through all that
35:51
Speaker 2: kind of stuff.
35:56
Speaker 1: C three po or R two D two.
36:01
Speaker 2: Why that is a tough one. I'm going to say
36:04
Speaker 2: R two D two.
36:05
Speaker 1: Not the bipedal one, not because that walks on two legs.
36:10
Speaker 2: C three, Well, no, I'm going to change my mind.
36:13
Speaker 1: I buyased it. Wait, why were you going to say
36:15
Speaker 1: or two I was going to say to.
36:16
Speaker 2: You because C three pos a protocol droid And if
36:19
Speaker 2: it's just about language, why do you have legs? Was
36:22
Speaker 2: On the other hand, you.
36:24
Speaker 1: Just you don't need a robot at all, right, you
36:25
Speaker 1: just need a little phone or whatever.
36:27
Speaker 2: On the other hand, right, it's a human centric robot.
36:30
Speaker 2: It's a robot that's meant to be a translator and
36:33
Speaker 2: be existing with people in the room and so on,
36:35
Speaker 2: and so actually making something that's more of a humanoid
36:38
Speaker 2: makes more sense if it's basically a social robot.
36:41
Speaker 1: Yeah, okay, Uh, what's something you wish that a robot
36:45
Speaker 1: could do for you, you know, outside of work.
36:50
Speaker 2: Honestly, what I want is for robots to make everything cheaper.
36:55
Speaker 1: Uh huh. I love it when technology makes things cheaper. Well,
36:58
Speaker 1: and that's what's been happening for the past couple hundred years,
37:00
Speaker 1: and all of automation has effectively made everybody richer effectively
37:04
Speaker 1: and improved everybody's quality of life because everything is cheaper.
37:09
Speaker 1: And I want to be able to book my vacation
37:12
Speaker 1: to the moon. I want to be able to not
37:15
Speaker 1: really worry about you know, as it is, I don't
37:17
Speaker 1: really worry about how much my phone costs or whatever.
37:20
Speaker 1: If it breaks, it's not an expensive phone, it's fine,
37:23
Speaker 1: it works, and all my software to transfers over. I
37:25
Speaker 1: want everything to be like that. I want my lifestyle
37:28
Speaker 1: to be supported by things that don't really cost a lot. Yes, well,
37:32
Speaker 1: and I mean in the very long run, the sort
37:36
Speaker 1: of technology driven productivity gains lift people out of poverty. Right,
37:40
Speaker 1: everybody cares about our phones, but there are a lot
37:43
Speaker 1: of people who have three dollars a day right now,
37:45
Speaker 1: and it would be great if they could get to
37:47
Speaker 1: three hundred dollars a day.
37:48
Speaker 2: Or beyond that. It's just that, you know, all the
37:50
Speaker 2: things that we need become very easy to acquire. Food
37:54
Speaker 2: is no longer, labor on farms is done in an
37:57
Speaker 2: automated way, Labor in terms of logistics and transporting things
38:01
Speaker 2: has done in an automated way, and so all of
38:03
Speaker 2: this stuff becomes so affordable that it's easy to uplift
38:05
Speaker 2: the quality of life of everybody on earth. That is
38:08
Speaker 2: what a lot from Robotics.
38:15
Speaker 1: Jonathan Hurst is the co founder and chief robot officer
38:19
Speaker 1: at Agility Robotics. Today's show was produced by Gabriel Hunter Chang.
38:24
Speaker 1: It was edited by Lydia Jane Kott and engineered by
38:28
Speaker 1: Sarah Bruguer. You can email us at problem at Pushkin
38:32
Speaker 1: dot FM. I'm Jacob Goldstein, and we'll be back next
38:35
Speaker 1: week with another episode of What's Your Problem