Loading summary
Transcript86 lines
- [00:00]
Commercial Announcer
This is a Monday.com ad. The same Monday.com designed for every team. The same Monday.com with built in AI scaling your work from day one. The same Monday.comwith an easy and intuitive setup. Go to Monday.com and try it for free.
- [00:15]
Betterment Advertiser
When you use the trusted investing and savings app Betterment to help grow your money automatically, you have more time for new niche hobbies like collecting miniatures. The joy that brings helps you sleep better at night and even motivates you to always use your PM moisturizer. Now you've got dewy glow and a sense of balance to match. Not worrying where your money is growing. That's the Betterment effect. Get started today@betterment.com Investing involves risk. Performance not guaranteed.
- [00:45]
Dr. Jonathan Hearst
You know, if you look back at history of other technologies, like maybe you look back on airplanes, right, and some of these early machines were these things that looked like giant birds. They tried to big, big flapping machines and so on to try and fly. I think that's how we're going to look back on some of these robots like the Tesla and the figure robot that try so hard to look like a person. It's going to be looking, it's going to be silly.
- [01:08]
Podcast Host (Interviewer)
The robots are coming. In the 105 years since playwright Karel Chapek coined the word robot and told a story about worker robots rebelling and taking over the world, we have been waiting for the onslaught. And here it comes. Elon Musk says humanoid robots represent the biggest business opportunity in the history of the world. But what's the reality? What problems will humanoids solve? When and why do they have to look so much like people? Today's guest, Dr. Jonathan Hearst, says they actually don't. Jonathan is one of the pioneers of modern robotics. He's the co founder and chief robot officer at Agility, which makes a humanoid called Digit that is actually working in warehouses. He's also a professor at Oregon State. Jonathan believes robots are not a threat and in fact are the solution to many pressing problems. And they are coming, though not as fast as some of the hypesters say. Here's our conversation. Welcome Jonathan. So great to have you. Let's just start right at the top, which is I think a lot of us have seen over the past five years these videos go by of these incredibly dexterous dogs and robot people doing flips and everything else. Elon Musk is out telling us that optimists, they're going to be millions shipped. And very recently a lot of us saw an incredible video that was purportedly from China, of a troop of dancers, robot dancers, doing things that would have been just inconceivable a few years ago. So let's start right at the top, which is humanoid robots. Why are we building humanoid robots as opposed to building tax task specific robots that just do one thing and don't look cute?
- [02:51]
Dr. Jonathan Hearst
And that's so fundamentally, it is about the versatility and the generality and the ability of these robots to do many different things. And even taking a step back from that, think about the origin of robotics, right? People have wanted machines, of course, to help them and have been building machines, labor saving devices forever. Maybe the first actual robots were in the 60s and those are the bolt down robot arms that can do pick and place and painting and everything else. And then maybe in the 2000s we get kind of to the next era of robotics, robotics, which is mobile robots. And we see the autonomous mobile robots in warehouses and things like that, moving bins and totes, moving components around Amazon warehouses and all these other places. Both of these scenarios though, you have to design the environment and the space around the robot and for the piece of automation. And what we're starting to see now with these humanoid robots and these robot dogs and other things are robots that can exist in human spaces. And along with the advent of AI, which allows it to now verbally interact with us and understand its perception, what it sees in the environment, more general environments, we're really on the cusp now of this third major era of robotics, of robots on human terms, and robots in human spaces where we don't actually have to modify the environment for it. These robots can just walk into our spaces, work with us, work for us here. That's what's new and different.
- [04:14]
Podcast Host (Interviewer)
And so we've been talking about that for more than 100 years. I think this idea that we're going to build robots that are quite similar to us and can do the same things, it's taken a long time. Why is it so difficult?
- [04:26]
Dr. Jonathan Hearst
Oh, I mean, it's the generality, it's the versatility. Typically when you think about robots, certainly in the past, you know, 20 years ago, these are machines that are in a very kind of structured environment and have a very specific set of things that they do and somebody is programming the steps that they're going to take and if you just give them some variability in what they're doing, they can't really deal with that. And that's what's new, right? So AI allows robots to make decisions about what's. What are they going to do what is a doorknob, what is a computer mouse, what are those things? And then what these large models can do, these AI models is kind of give you think of the IKEA instructions of what to do, of how to assemble something or where to go or what to do in a space. But what they really don't do is handle kind of all the low level physical intelligence of how to actually grasp a screwdriver, how to grab a doorknob, or how to do all of those kinds of things. That takes a whole different set of tools, many of which different types of AI tools and hardware. So that's part of what's enabling something like a humanoid that is meant to operate in such unstructured environments to actually make sense of them and even know what to do. And now people are starting to figure out rapidly how to build the right hardware, how to do the low level reinforcement learning approaches to control them. And the real difference now is that there's billions of dollars and thousands and thousands of engineers focused on solving this problem. So it's kind of like this big snowball that's been going down a hill and it has picked up some serious steam right now and is picking up a lot more snow as it goes.
- [06:10]
Podcast Host (Interviewer)
And I gather, as I read about it, there are a lot of things that humans just do without thinking about it that turn out to be incredibly difficult from an engineering perspect. And I'd love to just talk about a couple of them. I gather your research has been focused on legs, and I think you've said that these things should move like animals. Why? And why is that so difficult?
- [06:31]
Dr. Jonathan Hearst
Sure. Well, you know, my research has, yes, is focused on some legged locomotion, but generally it's been physical interaction and it's been coming from a different direction than most roboticists. It's been coming from the physics and the hardware side of it to build hardware where that makes it even physically possible to do the kinds of physical interaction that we see more often in humans than we see in robots. The reason it's so hard is if you think about how most machines historically have been built, it's a great big electric motor and a great big gearbox, right? And so these motors are spinning thousands of rpm, you know, moving really fast to follow some trajectory. And if it comes into unexpected contact with something, you can't just stop that thing. It's spinning thousands, you know, and you have huge, huge forces and things break. So that's why these machines are typically position control. It's typically about following some trajectory and so they're good at like following a paint path for painting a car or spot welding. When you've got a thing and you just have to make touch of each individual points, what they're not good is the things that people do all day long where we use the constraints of our environment for our manipulation, for our locomotion. A simple example is putting a key into a lock. A robot to do that would have to do it with very high precision. If it's off by a few thousandths of an inch, it's gonna destroy the key as it tries to push it into a hole and it's off by a little bit. Whereas a person doing it, you just drag the key to the edge of the hole, catch it, jiggle a little bit and go in. Because it's about controlling force and it's about being compliant and having this really soft kind of interaction with the world about forces more than positions.
- [08:11]
Podcast Host (Interviewer)
And so what about legs and feet? And first question, obviously, why don't we just put wheels on them, which are very effective and st you have four legs, it's not going to fall over. So why the importance of recreating human legs effectively?
- [08:26]
Dr. Jonathan Hearst
Yeah. So let's see. One point I want to make is, at least personally and for our company, Agility Robotics, we are not trying to build a machine that looks like a person. I would go so far as to say we are trying not to build a machine that looks like a person. When you do that, if you just copy the way a person looks, often you copy all the wrong features and all the wrong things. What we're trying to do is build a human centric robot, building machines that can go in human environments, in human spaces and do useful work. And often then you look at why people are so good at something, if you can understand the first principles of why they're good at it. Now you take that principle and instead of trying to copy the morphology of a human, because humans are made of bone and muscle and nerves, you use your engineering principles with aluminum and motors and gears and things like that to try and capture the same physics. The solution may look different. So that's kind of our philosophy for it. Now, wheels are very important and there's a reason people ride bicycles and there's a reason why cars are so great. And those reasons aren't going to go away. So in a warehouse, when we have AMRs that need to really go to far distances and bring things all over the place, wheels are the right solution for that. Legs we find, though, are very Very good for times when you need to be balancing. So when you need to be picking something up off the floor and lifting it to the top of a 2 meter high shelf and doing that in a narrow aisle way while being able to walk through doorways and things like that. So even ignoring stairs, ignoring obstacles, even in just human environments that are ADA accessible. But you want to be able to reach things high, you need to be balancing. And legs are the most stable way to be dynamically stable. Much more than, much more than trying to balance on wheels.
- [10:14]
Podcast Host (Interviewer)
All right, so I want to come back to this. They don't look like people, and I'm going to be a little skeptical when we get there, but let's go back to legs. One of the things I noticed about Digit, I think that's the name of your humanoid ish machine. Why do the knees go backward?
- [10:30]
Dr. Jonathan Hearst
Yeah, that's a great question, right? It's a very long story, but I will shorten the story to say we did boil down looking at the full range of animals. We look at humans and guinea fowl and ostriches and turkeys and ghost crabs and horses and said, okay, now what is true, commonly true, among all of these animals that walk and run to try and get at what is the common truth of legged locomotion that is not tied to specific details of their morphology? And you know, we end up with a very simple math model. And then, you know, while I was a professor at Oregon State, I'm still a professor there, but I spend all of my time at agility now. But when I was full time professor, it was a science question then. And we built the machine Atreus that looks nothing like a human. It looks like a microwave on stilts, you know. But that was matching this math model. And that's the first machine that was able to reproduce the dynamics of a human walking gait. And we measured that by having Atreus walk over a force plate on the ground, measuring the ground reaction forces, measure the center of mass motion, and found that that data matched exactly a graduate student walking over the same surface. And it's not because we were trying to copy the graduate student, it's because we boiled it down, tried to capture the right physics, and then went back and tested when we capture the right physics. Now, does it look like a person walking? Now as an additional feature, all of a sudden the robot was extremely robust and stable over all kinds of terrain. And we were able to walk and run and do the continuous transitions from walk to run over grass and pavement and Gravel and you know, handle big disturbances and big step ups and step downs. And there's no perception, there's no cameras. It's a very simple machine. So we kind of got to the science foundations of that and that's the path that we took that got us to the leg that we have on digit.
- [12:21]
Podcast Host (Interviewer)
And are humans badly designed then? No, actually should our knees have gone the other way?
- [12:26]
Dr. Jonathan Hearst
So I'll say that the more bird like. It's not exact, but it looks more like a bird. Like an ostrich is very, very good for walking and running. Right. So you know, an ostrich can outrun a human for sure, 40 miles an hour, something like that, all day long. But humans are very good at manipulation. Humans are very good at picking things up off the ground and lifting heavy things. And I actually think that a humanoid configuration is better for that. So if we want great mobility, then something more like bird legs is a good way to go if we don't need to go very fast or go very far. If you do it right, if you get the function correct, if you do it physics first, it's going to end up looking a little bit more humanoid with the regular two link leg and
- [13:13]
Podcast Host (Interviewer)
sticking on locomotion and legs. I was talking to a friend of mine who's a professor who deals with machine vision and he said one of the things that is so difficult for a robot is to process changes in balance and information as quickly as a human does. Because apparently we step in a hole. And you've described this in one of your papers, we react incredibly quickly because we don't actually have to process the signal. It just immediately gets bounced back to adjust the legs. And what he was describing is that even very rapid machine vision, it's still a frame rate that is not fast enough for the robot to catch itself. So how are you doing that? What's happening there? That enables the robot to log.
- [13:52]
Dr. Jonathan Hearst
And so it is the wrong solution to try and use perception and close a control loop around very, very fast perception control to try and correct for foot placements or something like that. Right. You really have this hierarchy of control around behaviors. And Atreus was kind of the first two layers of that hierarchy where the hardware is right and then a very simple, very high rate control approach is on it. But again, there's no perception, there are no cameras at all in Atrius. If Atrius steps in a hole, it just continues forward. And you know, we did these tests on animals as well where we had them running down a Runway, a bunch of times. And then one time we move the force plate down, but there's still a piece of tissue paper over the surface. So the animal has no idea that there's this pothole in front of them. And they can handle, like, fully half their leg length in terms of a big pothole without even slowing down. And they did not see it coming. There was no perception involved. There was no planning involved. It's just that getting the dynamics right of this dynamical phenomenon that is legged locomotion means it's incredibly robust to a whole range of these kinds of disturbances. When you do need perception is for things like intentionally changing your behavior to go up and down stairs or walk across stepping stones or things like that.
- [15:07]
Podcast Host (Interviewer)
All right, so let's go back to what you said a few minutes ago, which is that agility does not design robots to look like humans. And yet, when I looked at your wonderful video this morning, I noticed immediately that Digit has a torso and arms and legs and a head with eyes, as do Tesla's Optimus and the thing from Figure AI, which is going to be in our kitchen, and the One X. Why do they all look like people? What's going on there?
- [15:35]
Dr. Jonathan Hearst
Well, I can't speak to others. I know that many of them are just copying the way a person looks. I know that many of them are making choices for their actuator designs that are very poor for actually physically interacting with the world. I know that many of them are artificially choosing to say, I need my shoulder and arm dimensions to be exactly human. Like, I need my waist width to be just like a person. And everything needs to fit. All the actuators need to fit as if this were a person in a suit. That's sort of what I think is unnecessary and actually, like, harmful, really, in trying to make a robot that's actually useful in the world. So our train of thought is this. I already talked about how if you want to be balancing and dynamically stable, to be lifting things up high and things like that, legs are the most stable way to be dynamically stable. So we started bipedal. I said that having sort of humanoid legs, if they're shaped correctly and they can have the right range of motion and everything, are probably the best way to lift heavy things off the ground and be good manipulators. Now, also, you need to put a lot of batteries and compute and things into a torso, and the best way to have it be is upright, because you need to be narrow spaces, right? And you need to be able to rotate and turn in Those narrow spaces. You also need to be able to lean the torso. If you're going to start moving forward, the first thing you can do is just lean your torso forward. You start falling that way and then you can walk in that direction. You can't start walking without starting to fall in the direction you want to go. And then the reason you want your arms, well, the reason you want it to be bimanual is so that you can pick up big things, pick up big boxes, big totes, you know, have a range of being able to pick up small things with hands and great big things. With the two arms, you want them mounted at the shoulder because you want to be able to reach as high as possible. Like at the top of your tall torso is where you want these arms because you want to reach up high while also being able to reach the ground. In addition, you want two of them because it gives you a lot of inertial actuation for controlling your yaw and also for controlling your pitch if you windmill your arms. Also you want the full 360 degree reachability to sort of catch yourself when you fall so that you can slow down that impact with your arms when you hit. So it's like a lot of reasons that come together why you need the upright torso, why you want the arms where they are. Then why do you want a head? Right, well, it's a human centric robot. And I would say early on, you know, we didn't have a head on the robot. And you can design robots. There's some recent ones I've seen out there and think of like the world of Monsters Inc. You know, there's a bunch of monsters that don't necessarily have a head. What you do need, though, is a face. What you do need is something for the people around it to not feel like it's hitting the uncanny valley, for people around it to feel comfortable with the machine. So you don't need to put a head that's shaped like a human on it. You do need to have a face, a thing that people can look at. You have an opportunity to communicate with people on human terms, with expression, with body language, with, you know, verbal language, all of those things, so that you don't have to train people to program it. You can just talk with the robot.
- [18:38]
Podcast Host (Interviewer)
And so digit, your robot, is that, is that for communication then, the head? Because again, if our, if our listeners haven't seen it, it's a very cute machine and it has a head and it has eyes. And so I know what to talk to and I certainly know how to interact with it. Is that because people are actually interacting with it?
- [18:57]
Dr. Jonathan Hearst
Yeah. I mean it's very important if we're building a machine that's coming into your warehouse or your factory and it's going to work with you to not be afraid of it. Right. To feel like this is a machine here, this is a machine here that's here to help and to do useful tasks. And you know, it's entirely non threatening and you know, you want to feel like you don't want the robot to surprise you. So having a pair of eyes that's kind of moving a little bit, blinking now and then, your brain registers this as like a thing alive rather than an inanimate object. Right. And with an inanimate object, when it moves, it surprises you and then you're uncomfortable about it. You give it some space. If it's always kind of moving a little like a person does, if the eyes are blinking, it doesn't surprise you when it moves. When it's about to move, it looks in the direction it's about to move and then it goes in that direction. So it's about sending these cues to really give this sense of projecting what it's going to do and making sure it doesn't surprise anybody.
- [19:59]
News Anchor
Zootopia 2 is coming home to Disney Plus. Let's go get ready for a new
- [20:03]
Commercial Announcer
case where the greatest partners of all time, new friends Gary the Snake and
- [20:07]
Dr. Jonathan Hearst
your last name the Snake Dream team and new habitats.
- [20:10]
Commercial Announcer
Zootopia has a secret reptile population.
- [20:12]
News Anchor
You can watch the record breaking phenomenon at home. Disney Zootopia 2 streaming March 11th on Disney Plus. Rated PG. And right now you can can get Disney plus and Hulu for just $4.99 a month for three months with a special limited time offer ends March 24th. After three months, Plan Auto renews at $12.99 a month term supply.
- [20:29]
Kara Swisher
Hey, Kara Swisher here. I want to let you know that Vox Media is returning to south by Southwest in Austin for live tapings of your favorite podcast. Join us from March 13th through the 15th for live tapings of Today Explained Teffy Talks, Prof. G Markets and of course your two favorite podcasts Pivot and On with Kara Swisher. The stage will also feature sessions from Brene Brown and Adam Grant, Marques Brownlee, Keith Lee, Vivian Tu and Robin Arzon. It's all part of the Vox Media podcast stage at south by Southwest presented by Odoo. Visit voxmedia.comsxsw to pre register and get your Special discount on your innovation badge. That's voxmedia.comsxsw to register.
- [21:17]
Commercial Announcer
Really?
- [21:18]
Kara Swisher
You should register. We sell out and we hope to see you there.
- [21:25]
News Anchor
After decapitation strikes against Iran's leadership, what can we expect next in the escalating war?
- [21:30]
Dr. Jonathan Hearst
The big question is, if there is going to be a next strongman in Iran, what kind of strongman will that person likely be? I don't think that there's going to be another powerful cleric supreme leader.
- [21:44]
News Anchor
I'm John Finer.
- [21:45]
Podcast Host (Interviewer)
And I'm Jake Sullivan and we're the hosts of the Long Game, a weekly national security podcast.
- [21:51]
News Anchor
This week we sit down with Kareem Sajapour to discuss what to expect in this next phase of the war against Iran.
- [21:56]
Podcast Host (Interviewer)
The episode's out now. Search for and follow the Long Game wherever you get your podcasts. So one of the questions that everybody is starting to struggle with, especially as we see the robots get more and more capable, is, hey, they're coming for jobs. And in this case maybe the fact that it looks like a human actually makes it scarier.
- [22:17]
Dr. Jonathan Hearst
Yeah, I mean, look, we've been building labor saving devices for hundreds of years and you know, early In American history, 90 plus percent of Americans were working in agriculture and now 2% of Americans are working in agriculture. That doesn't mean that, you know, 88% of Americans are unemployed. It means that our country is producing vastly more than it could ever produce before. And that's sort of how this always goes. So bringing in new automation and new labor saving devices it produces, or let's just talk about AI and embodied AI and humanoid robots are going to produce both capacity to do more than we could ever do before with the same number of people, rather than having fewer people do. Just, you know, we're all satisfied, we don't need anything more. Humans are obviously ambitious. We want to do more, we want to build more and we want to create more. It also builds the capability to do more things that we couldn't do before. And to illustrate that, think about the question of like, how many blacksmiths would it take to build a laptop? Doesn't matter. Blacksmiths can't build a laptop. You need to leverage your way up with new technologies. And humanoids and AI driven humanoids are going to do that for us.
- [23:30]
Podcast Host (Interviewer)
Yes, I've used that farm to non farms example many times and it actually is very reassuring. And every technology transition we've seen so far, many more jobs are created. Even though there's a disruption and often pain, if you're in the way of that.
- [23:44]
Dr. Jonathan Hearst
All right, well, I've got another. Another example. Like, I was involved early in my robotics career in the. In the DARPA Challenge, which was the first, you know, cars autonomously driving from point A to point B. And, you know, the early nav lab projects at Carnegie Mellon when I was a graduate student there, and at that time was sort of the inflection point where everybody said, oh, wow, autonomous cars are not 100 years in the future. They're actually, like, coming right away. And there's a lot of worry about putting truck drivers out of work, because driving trucks is one of the biggest careers and one of the biggest jobs in America. As it turns out, though, the rollout of autonomous driving takes some time. There are autonomous trucks on the road now, but there's not all of them. There's some of them. And it's gradually growing faster than the rollout of those robots, though. And those autonomous vehicles has been the decline of people going into the trucking business, because people have say, see that this is coming, and maybe that this isn't a lifelong career, and they choose other careers before they start. And today we have a huge shortage of truckers. It's actually a real issue. And so that's a good way that the transition happens is people finish their careers out, they retire out, new people choose different paths, and that transition happens without significant pain. That's how I hope that this can go in most scenarios.
- [25:06]
Podcast Host (Interviewer)
Yes. And despite the hysteria about AI replacing workers, even that is happening relatively gradually. I mean, we're three or four years in, so. Just a couple more questions on the anatomy of these. One is, I gather the hands are incredibly difficult. And I think one of the big detractors, who I believe, I can't remember his name, but who invented the Roomba, is saying, like, this whole thing is just never going to work.
- [25:34]
Dr. Jonathan Hearst
Yeah, you're talking about Rod books, right?
- [25:35]
Podcast Host (Interviewer)
Yes. Yeah, there we go.
- [25:37]
Dr. Jonathan Hearst
This is his job now, is to be the naysayer, be the, you know, the crotchety old man of robotics who's been seen it before and been in it forever. But he makes some very, very good points. Right. It is true, I think, that manipulation is a decade behind locomotion in terms of really being something we can do. Well, part of the challenge is for locomotion, you just have to have a model of the robot itself, and then the feet are going to be the thing that hit the ground. And all we need to do is design a system that's robust to different ground heights or changes in ground stiffness like that for manipulation though, you know, it's many more degrees of freedom in a hand. And now there's huge uncertainty about all of the things you're interacting with, about how soft or hard they are, about the frictional components of them. You know, when you pick up a cup, how heavy is it? There's water sloshing inside. You can't model those things easily. Right. It's not like a locomotion where you can just model the robot and then the world is reasonably static for what for what it is. And that makes it a lot harder. But we're well on the path to figuring it out now. You know, I'll make the point about the morphology that we did before. Lots of groups are building these five fingered hands. Why? Nobody knows. You know, I think that the reason we have five fingers has a lot to do with the first fish crawling out of the ocean having five bones in their fins. That's it, that's the reason. It's evolutionary baggage. But we see lots of dexterity from octopus, from trunks of an elephant. There's lots of ways to be super dexterous. Imagine how dexterous people are with chopsticks, who are like lifelong users of chopsticks. Right. It's a fairly simple end effector, just two sticks and they can do so much with them. So dexterity, I think has a lot more to do with the forces you're applying, with the compliance and the compliant interaction with the environment, with the sensing and perception of those forces, as well as the, the visual perception and the ability to do things intelligently that way. Now we are so far from solving it. We don't see a lot of artificial examples of dexterity that aren't kind of a reproduction of like a teleoperated thing, like a person controlling a robot remotely, basically like chopsticks, except with electricity. In between is where we see some of our most dexterous behaviors Today. We're going to see some rapid progress. It is going to be a little while.
- [27:52]
Podcast Host (Interviewer)
And where are the opportunities there? And are we going to see something suddenly like an elephant's trunk or an octopus?
- [27:57]
Dr. Jonathan Hearst
It's not going to be sudden, no, but it is going to be consistent and gradual and it's going to snowball like we talked about earlier. So, you know, agility's hands right now are fairly simple claw things. Because our use case, our first beachhead market on the path towards robots and grocery stores and hospitals and construction sites and homes. The very first beachhead is just picking up bins and totes and putting Them from like a conveyor belt to a shelf or the other way around, or picking them up off AMRs and putting them on a conveyor belt because there's a huge market for that and it's a good place for the starting point for humanoids. But we have to lift a 25 kilogram tote and you know, the 25 kilograms might be a mass rolling around in that bin. That's a really a hard thing to pick up. And there's no manipulator off the shelf, there's no, you know, five fingered hand that can do that. So we have some fairly powerful actuators in there. To be able to really grasp this thing well and lift it up over time though, it's going to have to be able to do that power grasp, but also lift up a pen off of the table. We've got lots of ideas for how to go down that path and we're working on it. And we're going to see continued improvement from us and from lots of different groups and companies trying to build dexterous manipulators. Both the hardware that makes it possible to do the things you want to do, that allows you to do the force control that means you can hit the hand and it'll flex and bend rather than just break immediately. And the AI based controls approaches that are going to make it possible to really be dexterous.
- [29:22]
Podcast Host (Interviewer)
And so last question on the fact that some of these look a lot more like people than yours. You see the videos of the robots in the kitchen unloading the groceries and unloading the dishwasher, putting away the groceries. It's very much like having a human being in the kitchen. And is that something that we are going to like? Is it necessary? And I ask it just because one could say we already have many excellent robots in the kitchen. The dishwasher does an amazing job of washing dishes. It doesn't look anything like a human. I don't feel like I'm interrupting. I don't feel like I have to be polite. I suppose a dishwasher could be made for me to talk to it and that would be fine. We could put a face on it. Why do they have to look like people?
- [30:03]
Dr. Jonathan Hearst
I mean, they obviously don't. I think we're going to look back, you know, if you look back at history of other technologies, like maybe you look back on, on airplanes, right? And some of these early machines were these things that looked like giant birds. They tried to big, big flapping machines and so on to try and fly. I think that's how we're going to look back on some of these robots like the Tesla and the figure robot that try so hard to look like a person. It's going to be looking. It's going to be silly. It's going to look like those flapping machines as we try to figure out airplanes. There's going to be similarities, though. I mean, we still have wings on airplanes, right? And I think that being bipedal and bimanual and upright torso and having some cues for humans around it is still going to be features that are important. But think about all the different stuffed animals that kids have and the variety of shapes that still have the right cues of a face to look at and things like that. I think we'll have similar kinds of variety in our multipurpose human centric robots that will be in our homes. And it's interesting and it might be a while, by the way, before this happens, right? Because there's kind of three main reasons why robots in the home, these robots that are dynamically stable, robots that look sort of like humans or something like that, number one is safety, right? These machines, if they're balancing, can fall and you can't have a robot fall on a child or a pet. It's really just unacceptable. And that's a safety risk that's very hard to overcome. We're doing it in warehouses because we can control the situation a little bit. We can make sure the robot stops moving and sits down before a person can touch it. You can't do that in a home or it's just not that useful. Right? The other reason is just the level of complexity and variability and chaos in homes. Every home is different every day. A single home is different. There's so many things to understand about it and you need such capabilities for mobility and dexterity to be useful in it. And finally, it's a cost sensitivity thing in a warehouse. You can capture half a million dollars of value in two years with a robot that's doing some really straightforward work. And boy, even if you could have a robot reliably load a dishwasher, which we can't, that's not going to be worth half a million dollars for two years of service to have a robot doing just that in somebody's homes. So we'll get there, but it's going to have to be way safer, way cheaper and way more capable. And there's all these huge markets in warehouses and commercial settings and retail and everything else on the way to being good enough to be in homes.
- [32:36]
Podcast Host (Interviewer)
So talk about that. So where are we going to see this first?
- [32:39]
Dr. Jonathan Hearst
Oh, warehouses and manufacturing.
- [32:42]
Podcast Host (Interviewer)
Because aren't those pretty automated already?
- [32:45]
Dr. Jonathan Hearst
Sure, but not fully. Right. They're not lights out environments. Like the best place for digit right now is connecting islands of automation. So there might be a whole automation system of amrs. You know what is an amring?
- [32:57]
Podcast Host (Interviewer)
I'm sorry?
- [32:58]
Dr. Jonathan Hearst
Autonomous mobile robots, like wheeled robots that are bringing bins and totes and things. Like kivas. Yeah, like a KIVA system. Right. And then there's a whole system of conveyors. They're all autonomous for sorting things and sending them where they need to be to say to be packed and sent. Right. And then there's some person that has to wait for the bin to show up. And then the robot, that light comes on says okay, pick it up. And then they just pick up the bin and put it on the conveyor. All they're doing is connecting these islands of automation and it's two different companies and you know, so they don't necessarily have a good interface. The workflow is different every time because it's slightly different in every warehouse. So there's enough variability with those that it's hard to have a single customized system or a specialized system that automates that. But it's still very workflow automated. Like it's the same action that you do over and over and over. And it's a fairly controlled environment. So it's like a really good like middle ground starting point for these more human centric, more general purpose machines to start to just get our foot in the door. Literal and figurative foot in the door for deploying these humanoids. Really learning what is hard about deploying fleets of these incredibly complex machines, servicing them, supporting them, coordinating them, fleets of them, and then gradually getting better and better and having more and more capabilities over time as it gets better manipulation. Now you can start filling your bins with things, do the each picking, then pick up the bin and take it somewhere. Eventually you can start handling cardboard boxes, which are surprisingly hard to handle, and then loading and unloading trailers and then you can stock shelves in grocery stores. We got to work our way up to those things.
- [34:39]
Podcast Host (Interviewer)
And another reason people are so scared about, in this case losing jobs to machines that look like people is you can immediately just imagine the benefits. No sick days, no vacation, no complaints, no workplace injuries, no vacation. You know, all the different things. You don't have to hire people, you don't have to fire them, so forth. Lots of things that go into managing people that are difficult. So it's Scary. Again, talking to another professor friend of mine who knows the space pretty well, I said, yeah, you got to factor in maintenance. And you talked about this before. It's like robots break and that creates a problem. And so where are we on that? Like, you talk about managing fleets. Is this a whole new series of jobs and capabilities that humans are going to have? Is managing the fleet of robots in the warehouse?
- [35:25]
Dr. Jonathan Hearst
Yes, of course. And honestly, the reason everybody is concerned about this is because they look like us. But fundamentally, these machines are just a continuum of the improvement of automation that we've been working on for hundreds of years. And there's really no, I don't believe there's a massive step function because it is hardware. It is physical hardware. We have to manufacture it and deploy it and support it. And it takes time to iterate on those things to get good at that and to deploy them. It might look more like the smartphones have evolved, some combination of smartphones and cars. And cars have been improving for 100 years and more. Smartphones have gotten to a point of real cleanness. But now all of a sudden, we have folding smartphones. You know, you see these iterations over time, but it does take time, and it will take time to really grow and deploy. Like I said, I hope that it happens at a rate that it helps us with our extreme demographic challenges of an aging workforce, with onshoring manufacturing, with all of the shortages of labor that we're currently struggling with, which is our actual problem right now. I hope that it's just in time to help us with that rather than being this surprise shock that all of a sudden puts everybody out of work. I don't really see how that could happen.
- [36:49]
Podcast Host (Interviewer)
And certainly China seems to be seeing it that way, where they're going to have a demographic issue themselves.
- [36:54]
Dr. Jonathan Hearst
The whole world.
- [36:55]
Podcast Host (Interviewer)
Yes, exactly.
- [37:00]
Commercial Announcer
This episode is brought to you by indeed. Stop waiting around for the perfect candidate. Instead, use Indeed sponsored Jobs to find the right people with the right skills fast. It's a simple way to make sure your listing is the first candidate. C According to Indeed data, Sponsored jobs have four times more applicants than non sponsored jobs. So go build your dream team today with Indeed. Get a $75 sponsored job credit at it indeed.com podcast terms and conditions apply. Starting a business can seem like a daunting task unless you have a partner like Shopify. They have the tools you need to start and grow your business. From designing a website to marketing, to selling and beyond, Shopify can help with everything you need. There's a reason millions of companies like Mattel, Heinz and Allbirds. Continue to trust and use them. With Shopify on your side, turn your big business idea into Sign up for your $1 per month trial@shopify.com specialoffer I get so many headaches every month. It could be chronic migraine, 15 or more headache days a month, each lasting four hours or more.
- [38:04]
Botox Advertiser
Botox Onobotulinum toxin a prevents headaches in adults with chronic migraine. It's not for Those who have 14 or fewer headache days a month. Prescription Botox is injected by your doctor. Effects of Botox may spread hours to weeks after injection causing serious symptoms. Alert your doctor right away as difficulty swallowing, speaking, breathing, breathing, eye problems or muscle weakness can be signs of a life threatening condition. Patients with these conditions before injection are at highest risk. Side effects may include allergic reactions, neck and injection site pain, fatigue and headache. Allergic reactions can include rash, welts, asthma symptoms and dizziness. Don't receive Botox if there's a skin infection. Tell your doctor your medical history, muscle or nerve conditions including als, Lou Gehrig's disease, Myasthenia gravis or Lambert Eaton syndrome and medications including botulinum toxins as these may increase the risk of serious side effects.
- [38:49]
Commercial Announcer
Why wait? Ask your doctor. Visit botoxchronicmigraine.com or call 1-800-44-BOTOX to learn more.
- [38:58]
Podcast Host (Interviewer)
So how long will it take? I mean, where are we in five years, 10 years, just in terms of the manufacturing and warehouse opportunity?
- [39:04]
Dr. Jonathan Hearst
Yeah, I mean, in five years we're going to have thousands of robots deployed, but we're not going to have millions of robots deployed and the market needs millions of robots and we're going to have thousands. And I hope and expect that a couple of the other companies out there are going to be up to thousands by that point as well. You know. So over the next five years, expect to see that, expect to have it be still feeling new, but starting to see it and starting to see the videos. And if you get a, you know, a tour of a warehouse, you're going to see a humanoid robot doing work in it in the next few years. Ours are deployed now. Digit is the first commercially deployed humanoid robot that's doing work and being paid for it. Right. That's a really cool historical moment, but it's still just in the vicinity of single digits, double digits of robots that are deployed and once we get the safety case handled and they step out of a work cell and there's going to be hundreds next year and then thousands coming out over the next couple of few years.
- [40:03]
Podcast Host (Interviewer)
And what about home? When are we going to have that
- [40:05]
Dr. Jonathan Hearst
vision of, I think it'll be 20 years.
- [40:07]
Podcast Host (Interviewer)
20 years.
- [40:08]
Dr. Jonathan Hearst
Okay.
- [40:08]
Podcast Host (Interviewer)
Long time.
- [40:09]
Dr. Jonathan Hearst
So by the time, you know, when I'm really old enough, I want to be aging gracefully in my home. I think we're just going to start to have some useful robots that can do some basic pickup around the house and basic manipulation tasks and maybe load a dishwasher reliably. Right. Not just a demo. The gap from doing a demo on camera to deploying something as a product is a pretty huge, vast gap. It's a 10x kind of difficulty. So I think we're gonna see a lot of really cool tech demos. People are gonna get used to the idea and I think people are gonna be very comfortable with it because, you know, a lot of people would like to have help around the house, but would rather do it from a machine that is. There's a lot more like it's impersonal that you don't have to worry about dignity. You're not worried about whether that person wants to be doing that job or not. The robot's a machine, it's fine. There's a lot of uncomfortable things around that that just vanish with a robot doing it. But people have had help around the house forever. That's the kind of thing that people have always done. It's just now a win win scenario.
- [41:18]
Podcast Host (Interviewer)
What about friends, companions? Because Elon brings it up all the time. Sexbots, other companions.
- [41:26]
Dr. Jonathan Hearst
Well, I can just say Agility Robotics does not anticipate going into that market.
- [41:31]
Podcast Host (Interviewer)
Yes, but you must have a view on it. Not the professional view. But is again, is this Elon telling a great story as he often does?
- [41:39]
Dr. Jonathan Hearst
I mean, I'm sure that some enterprising entrepreneurs, you know, this is America, are going to try anything that will sell. So whatever can sell and whatever people can try to do, we will see.
- [41:52]
Podcast Host (Interviewer)
And then what? What about China? That video was very impressive. Maybe it was AI, I have no idea.
- [41:58]
Dr. Jonathan Hearst
No, no, it was very real. Now that unitree, they're very good. That group knows what they're talking about. They're working very hard. China has identified humanoids as a national priority. They want to win that market the way they have won drones. They want to really be elite. And they know that humanoids, along with broadly AI enabled robots, are going to be foundations of the economy. Because this is the future of labor is having machines do it. Machines will we talk about when our robot's going to be in the home and I say 20 years, maybe 30. At that point, robots can physically do just about anything that a person can do, right? That's a different world right now. You know, robots can play chess and do math better than a person, but pretty soon they're going to be able to do physical things better than a person. Similarly, there'll be tools for us, but China wants to lead that well. It's not going to be a monopoly, right? We're all going to be able to do it, so long as we actually try. So long as the United States actually puts their thumb on the scale a little bit as well. Because, man, China's really leaning on the scale hard to try and support their industry. So long as the United States doesn't allow us to fall behind and kind of helps the industry do that, we're going to be fine. There's nothing in the demonstrations we've seen from the Chinese robots that we don't know how to do. It's not a surprise on the technology. What it is is a real demonstration of strength, of manufacturing something very quickly. But it's also a cultural difference. Like in the United States. Why would anybody build thousands of robots today unless they have customers who are going to buy them? We kind of like to build this foundation first, of having a return on investment, of having robots that are going to be pragmatically useful, and then that starts this flywheel of being able to build machines in high volume. And that's the path we're on. These Chinese robots don't necessarily have that flywheel. They're selling them to researchers, they're selling them to early adopters to play with them. The Chinese government is subsidizing manufacture of these things to do things like these big demos. That doesn't mean that they are actually deployed in industry and doing useful work yet. Does mean, though, that, boy, they're getting a head start on knowing how to manufacture at volume. They're getting a head start on trying out, you know, reliability and learning a lot of things that will be needed when we do get to that point of having them be deployed for pragmatic use.
- [44:29]
Podcast Host (Interviewer)
And you talked about government putting its thumb on the scale. And this is something that, particularly in Silicon Valley, for decades, we've been incredibly against. Just get out of our hair, we'll figure it out. We've got all the funding and so forth, and yet the last few years, as we have sort of really begun to realize as a country that China is way ahead of us in some of these industries of the future, suddenly you have people clamoring for it. And so what do you mean exactly by that? What should the government be doing in robotics to help US companies?
- [44:59]
Dr. Jonathan Hearst
I think it's basically so we should do it our way and not the way that China does it. We have a different culture, we have a different economy, the way we do things and there's a lot of huge, huge benefits to the way we do it. Like the way venture capital works and startups work is almost unique in Silicon Valley. It's really special and it's a very, very powerful engine for and innovation. But some of these things like Humanoids are a technology that's not mature yet. And as a professor, there's this common problem of this the Valley of death, where you get something, you research something, you get to a proof of concept idea, like this idea could be something. And then there's this valley where there's no NSF funding, there's no DARPA funding, there's no kind of research funding to get it to the point where it's product ready. And VCs don't want to start really investing in something until a lot of the technical risk is gone and you have a product something and now it's about business risk and you're trying to grow your business. And Humanoids are such a big jump, a big leap. Everybody sees the promise that venture capital has come in a little bit and funded some of these slightly higher risk, slightly earlier stage technology things. And same with AI and some others like Quantum. But if we really want to ensure that we have a lot more startups that are pushing on this, there's got to be some say incentives for the big automotives to partner with humanoids companies or automation companies that's happening in China, or some incentive that is some matching cost that bridges the gap between the high cost when you're making prototypes and the, you know, the, the value that you want to provide in order to get the return on the investment. So if you're still like three years away from having the low enough cost on the product and high enough performance to have a really good return on investment, having some way that the government can just sort of lower that bar a little bit. And they did that for electric cars, right, the subsidies on electric cars so that they weren't quite so expensive for Americans to buy while they were still fairly expensive to manufacture as a new technology is a big thing that accelerated their development for us. So something like that, that lets the inventors and the innovators and the companies lets the capitalism work for Figuring out what the right solution is, but makes it a little bit easier to do that faster.
- [47:23]
Podcast Host (Interviewer)
And is anybody talking about that? Are we close to that? Is this a priority for the Trump administration?
- [47:28]
Dr. Jonathan Hearst
I have heard lots of conversations about it, yes. I think that in the government, they're starting to recognize that this is going to be really important for our future. Starting to recognize that China is prioritizing it, feeling the sting of drones being something that we innovated on in the United States first and had some very, very great companies that then kind of got put out of business, or at least not out of business, but shrunk to a, you know, niche areas by Chinese companies like dji, just reducing the cost and manufacture, out manufacturing everybody, and then just, you know, saturating the market like that. And now America doesn't have this huge drone industry, but we should have and could have. Right. We don't want to have that happen with Humanoids. And I think people in government kind of recognize that.
- [48:23]
Podcast Host (Interviewer)
And as the industry evolves, another friend who's an investor, professional investor for a very long time, we're talking about it, listening to Elon Musk talk about biggest opportunity in the history of the world, and very bullish analysts embodied AI in 10 years. You gotta get into it. There are all these different ways. And my friend had the theory. It's like, no, this is gonna be like another hardware market. Hardware's gonna get commoditized. It'll be the software in control. You think that's true, or does this look more like an iPhone?
- [48:49]
Dr. Jonathan Hearst
It looks more like an iPhone. There's no real way to separate out, you know, the hardware and the software. There's such a holistically designed integrated system.
- [48:58]
Podcast Host (Interviewer)
All right, this is tremendously exciting. Last question, which is if you are graduating from college today, you're hearing on the one hand, oh, don't go into the white collar industries. They're all going to get destroyed by ChatGPT and everything else. Now, I'm hearing from you that the warehouse jobs are going to be under pressure. What should we focus on if we're graduating from college today? Like, where are the real opportunities going to be as opposed to our going into industries that look like they're gradually going to be phased out?
- [49:27]
Dr. Jonathan Hearst
Yeah. First I want to push back. Warehouse jobs aren't under pressure. What's under pressure is companies that have warehouses trying to provide the services and not being able to hire warehouse workers. So anyway.
- [49:40]
Podcast Host (Interviewer)
But back plenty of jobs there for a long time.
- [49:42]
Dr. Jonathan Hearst
For a long time. And. But to your question around like, what do you want to do in college? Well, look, creativity and ambition are kind of human things that are going to continue to be in demand. And, you know, learn the latest tools. Like, you know, when you're in class, don't think of using an AI as cheating. Using AI is a absolutely critically necessary skill to understand how to use those tools moving forward and how to integrate those into the new way of doing things that people haven't known how to do before. In my experience, nobody is more creative at figuring out new ways of doing things than students than new graduates because they don't know how it's been done before. All they're presented with is the current state of affairs, right? And so they don't have this inertia, this baggage. So it's really important to be using the latest tools, understand how AI can be used, and be creating and inventing the new ways to use it. I think that AI is a lot like the Internet and that it is the infrastructure of tomorrow. And so students who were in college when the Internet was just starting and started to think about how can this provide value? How can I do something with this new tool that is really valuable that other people haven't thought of yet? That's where the big businesses are created. That's where the great value is created.
- [50:55]
Podcast Host (Interviewer)
Jonathan, thank you. This has been terrific. Best of luck to you and agility and look forward to seeing, seeing where you take us in the future.
- [51:02]
Dr. Jonathan Hearst
Excellent. Thank you very much for having me and I appreciate all the great questions.