The argument for humanoid AI robots with Brett Adcock from Figure

0:05Hi, listeners. Welcome to another episode of No Priors. Today, we're here with Brett Adcock, the founder and CEO of Figure AI, which is developing and delivering humanoid general purpose robots that can do unsafe and undesirable jobs. They recently announced a monster round of funding,$675 million, from Microsoft, OpenAI, NVIDIA, Intel, and Jeff Bezos. Brett, thanks so much for doing this. Yes, sir. You have this wild company doing humanoid robots. You just raised almost$700 million. Can you talk a little bit about how you get from a farm to software to vertical takeoff and landing to rope humanoids?

0:47Pretty normal path. Yeah, that's what I did too. Yeah, yeah. So my story started, I grew up in Illinois on a third generation farm. And it ended up basically at a pretty early age, started coding and getting the software and building things. and that basically has been now about 20 years of building companies. A little over 10 in software and a little under 10 in hardware. At one point I started a software company and sold it and then I started a company called Archer Aviation. We build electric vertical takeoff and landing aircraft and then about 21 months ago I started Figure. Can we pause there for a second because most people aren't like, oh, I'll just start an aircraft company.

1:33Like, how do you go from a software business feels less exotic to me to that? I grew up on a lot of hardware. And so I looked at hardware as like I really wanted for a long time to build hardware, build like kind of areas of deep tech. Only way to really do that was like self-fund your own venture and get it really moving. So after I sold, I sold Vetteri in 2017. And right away, I knew I wanted to build electric aircraft. And I actually went back down. Do you fly planes or something? I'm super passionate about, A, fixing traffic problems. We have like, you know, half the world living in cities and traffic's just getting worse and worse.

2:11There's just been no, there's been no solution there. And two is I'm a big believer in sustainable transport. I think, you know, all transport besides rockets will move electric, hopefully in our lifetime. So what we do at Archer is we build vertical takeoff landing aircraft. So aircraft that are kind of like helicopters, but fully electric. They can take off from a helicopter landing pad, so like inside of a city. and they can take you from here back to San Francisco in under 20 minutes. And sold. That would be a dream, right? Instead of driving for two hours. So that was a really hard business.

2:40I basically started the company out of University of Florida. I started in engineering at University of Florida and basically built a lab there for the first two years and built aircraft. And then moved the company out to California about three or four years ago. That's now a public company. It's on the clearance path. How do you go from that to humanoid robots? I would say less of a leap from software into Archer than it was from Archer to figure. Yeah. But what's the thesis for it? The thesis here is that if you assume that technologies are possible to build a humanoid robot, and just for listeners, humanoid means human form, so legs, arms, hands, and you can do basically human-like work with a humanoid robot, A, it's going to be the biggest business in the world by probably order of magnitude.

3:28Half of GDP is human labor, like an order of magnitude bigger than like all of transportation market. It's just an enormous industry. Two is we think we can have like significant economical value if we can basically have robots doing real work every single day. I think it would be a age of abundance as it relates to goods and services prices. And three is I think we'll have over enough time, I think we'll have an impact on the AGI timeline here. It distills down to perhaps the most important business I think I could be working on in my life. So I left 21 months ago to start Figure. And what we're trying to do here is commercialize humanoid robots.

4:00We're trying to get them into market in a significant way, build a fleet of robots, and then build an AI data engine and train those robots on how to do useful work. Why does being humanoid matter? I think there are plenty of people who work in robotics that say like, bipedal is really unstable. There are lots of reasons that these humanoid forms are not optimal for doing lots of different work. We can have 10 arms and be stronger and whatever else. I think there's like really two schools of thought of how to go about this problem. We can either rebuild like thousands or millions of special types of robots that do special use cases all over the world, or we can build a humanoid robot.

4:35And the reason we believe humanoid robots are the right solution is that the whole world was built for humans. It was built around humans, meaning the way we look biologically had a significant impact on the way our environment looks. If we were 10 feet taller and we had like, you know, six arms, a lot stronger, the world would look a lot different. When people ask, like, is the humanoid the right form factor? It's the wrong question. Just the wrong question to be answered. The humans are the humans like the ideal form. It's the wrong way to look at it. We're like a weird biological species. We're a weird form.

5:07But that's the world that exists. We built it so we can interact with it. If you go to Mars today, you're going to want to go grab coffee. You're going to want to walk around. You're going to want to like live in a habitat. You're going to want to do things. And you're going to build a human world around it. And then, so if we want to automate work, you want to build a general interface to that. You want to build like the equivalent of like the keyboard and mouse to the internet in the physical world. That equivalent is a human form. You can do everything a human can. And the world was optimized specifically for us.

5:37It wasn't optimized for us to have two more arms or nine feet tall. It was optimized just for the average human, the average like non-expert human. so it's really clear to us here at the the right form factor is the humanoid you can build one hardware system that can do everything i mean you can spend all this money and time on this robot that can be amortized over like you know millions of different tasks and applications and then conversely let's say you don't believe that you're going to go out in the world and look at every single use gate like every single task and build a special robot for it that special robot needs a company needs a brand it needs a team needs a culture it needs to go raise capital it needs the team needs probably an order of like 15 different skill sets that one single person can't do like in a full stack software role it needs firmware and embedded software and operating system and rotor and stator and electromagnetics and battery systems and bms systems and power distribution systems and thermals and the list goes on controls and ai and everything else and And so then you need to raise a lot of capital to sustain it.

6:44And then you're going to go build millions of those. How is that even tractable? How do you do that across the whole world and all these like special use cases? And then you're going to make a role. It is how the robotics industry works today. Specialized robots, right? Yeah. I'm not saying that's efficient, but I think that's where the question comes from. I don't know if we should be reasoning by that analogy. Okay. I think it's the wrong way to look at it. One thing you and I have talked about before is like, is this the right decade to build the company like what made you say like we have it now we can do it we need to prove it and i think we've started to prove it on like an mvp level like you can see the robot doing pretty useful stuff now and it's been 21 months took the first year to hire enough employees to come here to do stuff so we're kind of been around for kind of a year maybe a year and a half of like really useful like run rate in terms of like um what we've been able to accomplish can i actually um just describe the two public demos so far and like why they're important?

7:41Yeah, we've been doing kind of like two divergent set of demonstrations to the world. The first is we do plan as a business to start launching into more kind of industrial solutions, like more like the corporate labor market, you know, manufacturing, supply chain, logistics, those type of areas. We think that allow us to build the AID to engine quicker because we're shipping robots faster and it'll help us build manufacturing volumes quicker, which will help cost. Those are like the reasons why we're doing it. there's another market that we're extremely excited about, which is in the home. And the home is a really messy place.

8:12It's very unstructured. Everything's different. There's like a higher variance of failures where we're like, you know, if we drop like the number one dad cup at home or the number one mom cup, like not great. We drop a bin in like a warehouse, like who cares? You know what I mean? It's a little bit different scenario. Also safety is impacted. There's pricing compression as we move into the consumer world. There's just a bunch of stuff that happens. So we've done a few demos so far. First is we've done like bin moving, very traditional industrial solutions roles where we're taking bins from pallets into conveyor systems.

8:41We're doing that fully autonomous end to end on our robots now, all bipedal. And the second is we're doing kind of full consumer level manipulation and like speech to speech reasoning. So we're able to talk with the robot. It's able to understand what we're saying. It's able to visualize over the scene. It's able to do useful works like go and grab things like an apple or a certain place or make a coffee on a Keurig. and it's able to do all of that end to end, not only autonomously, but only from neural nets. So it's taking in speech, it's taking in video feeds in the cameras, we're processing that on a model and then doing inference and then we're outputting trajectories across the robot.

9:19I would say like, you know, as an entrepreneur, the one thing that we're always afraid of is hitting some like technical wall. It's like, we go out and do this and like everything just slows. We hit this like, you know, this upper bound and we just can't push through. We don't even know where the upper bound is right now. And that's what's really exciting for us and why we're really excited about the next 24 months is we don't even see it. We don't know where it's at. And we're still searching for that and moving as hard as we can to try to find it. Why is now the right time? Yeah, there's a few reasons.

9:51I don't think this was really possible five years ago. Looking back on the last, like, you know, even few decades, the, like, the powertrain or the system that we use is all electromechanical here. So it's batteries and motors. If you look at the amount of energy we have in the batteries or amount of like, say, power, torque density in the motors, those have improved significantly the last couple of decades. We just, we didn't have the runtime 10, 20 years ago in the robot with like lithium ion batteries to make this really work. We didn't have the power out of the actuators, much of motors to make this happen.

10:21So like a best analogy is like, you know, 10 years ago, a Tesla went 100 miles and now my Tesla goes 300 plus miles. It's because specific energy in the battery cell on a watt-hour per kilogram has improved, you know, maybe a 7 % KG since, you know, the last two decades. That makes sense. Like, if it, you know, can't carry more than 20 miles or it can't carry fast enough to go to the highway, it's just not a useful car. Run, like, yeah. If it was really heavy and it runs, like, you know, 10 minutes and it can't carry anything and the speeds in the motors are not very great, like, you just can't get anything useful.

10:51I think the second thing is in locomotion controllers. Like, 10 years ago, humanoid robots, bipedal was a huge risk. like they were clumsy they were slow they're falling over um the dark but robotics challenge is a good example of that like about you know about 10 years ago now uh you couldn't like look at it and envision that being like really useful or at least in your home that's completely changed here we actually started walking our humanoid uh from the time i filed the c corp in the business and we uh walked the robot it was under a year from when we started which was a pretty impressive feet um and i think we probably we're probably doing some of the like some of the best work in the world on bipedal locomotion here on it from a controller controller standpoint and the third is like basically ai systems like the the computation the algorithms um were just not feasible to do i would say 10 years ago um and i can say arguably the fourth would which would be you know we spend a lot of time now working with open ai as we announced and um we believe the default user interface to the robot is speech um you're going to want to talk to the robot we've even an industrial setting when you're unboxing the robot for the first time we think the initialization process is speech there's no um there's no like you know configuration demonstration you're talking to it and i think um i mean humans that's what we're doing today right we we do gestures and we talk, either through, you know, written text or speaking, whether it's transcribed or not, we're like using language.

12:22And we think that's primarily the main user interface. We think by default, the robot's going to be using. And, you know, five years ago, that wasn't possible either. You said you, the team wasn't sure where the bounds were, like where the walls you can imagine hitting or what were you afraid of when you started? I mean, this is just a hard problem. There's just, if you're afraid of walls, this is like the nightmare scenario for anybody. It's just a fun house of walls. Yeah. We have this like one of our five corporate values is like aggressively optimistic. We looked at some slides together where you were looking at one of the trades and some risk on a component and you had these mitigating factors.

13:05And the second mitigating factor was sprint harder. Like we will figure it out. Yes. Like work harder. Love that. That's the only solution to the risk at this point. The risks are profound. Nobody has been able to deliver a commercial humanoid into a market in human history. We have to not only... The threshold is we have to do a certain amount of an equivalent of human-like work performance, which is extremely hard. Humans are very productive. and we have to do that reliably and continuously over the course of months and years. And then to add on to all that, our robot has like, you know, over 30 degrees of freedom, like, you know, joints that can move on the robot.

13:57And, you know, the amount of, like the amount of the action space or the amount of orientations the robot could possibly be in is extremely high. Either you have to code your way of saying, this is, you know, I have to write, C++ or have to write a script for everything the revolution do. Yeah. Everywhere in the world, you have to solve that through robot learning is the short answer. So, and we haven't seen that work extremely well in human history either. I mean, we're watching that unfold right now with self-driving cars. We have to take all of those challenges on hardware, never having been done before, reliability and safety and performance, and then we have to learn everything.

14:37So yeah, to get back to your question, it's like, they're all there. They're all in the shadows, all these problems. There's all these other shadow of unknown unknowns that you hit at every single time. The reason why we don't feel like we see an upper bound is because now we kind of see the roadmap for the next 12 to 24 months and we're kind of just optimizing. We're just making the robot more reliable and faster. We're not trying to get the robot to do an end-to-end application for the first time. We've already done that. And we've done that for our first clients' use cases. It just has to be more robust.

15:11Zero to 100 or zero to, you know, walking bipedal robot in a year since you incorporate. How do you assemble a team that's so multi-domain so quickly? So it hasn't been my first rodeo. Having built the, like, you know, built all my teams previously at Vetteri and Archer. when I started to figure, there was a few things that no matter like what company it was, I would do and I would do again. The first is we set the mission, vision, values. I then wrote a master plan, which is basically like a 10-year journey over what we're trying to do. I wrote a culture document, which also lives online, about like what we stand for here.

15:49Like we like to move fast. We, you know, we do this, we don't do this. Or aggressively optimistic. Yeah, exactly. We like, we work hard. Things like that. we have to say um we work in the office every day five to seven days a week i spend i spent basically the first year like hyper focus on building the team how do we build like the world's best engineering team and who are those people and i built an org chart you know i'm at the top and we basically built out the teams with in detail of like who this all these groups should look like, whether it's, you know, controls, AI, actuation, battery systems, kinematics, integration and tests, industrial design, all of this.

16:32And then the skill sets underneath there. So it could be motor. We have like a rotor, stator, transmission, sensors, thermals, motor controls. That all makes sense as like a picture of it. But then like the reality of somebody who spends a lot of time recruiting for early stage companies is like, I can't just go like pick up that guy up from Boston Dynamics because I decided he's the right guy. Yeah. So I then went out and found everybody online that I thought was the best in the world. And then I did 300 phone calls over six months and I cold emailed all of them. So Jerry picks up and he's like, sure, Brett.

17:08They don't say sure on the first call, which they did that. So a few calls later, yeah. A few meetings later, they do. Yes. Or a certain percentage will. Yes. This is no different the one I did at Vetteri. I built, you know, the first few hundred like this at Archer. Um, and yeah, the first, you know, 30 to 50 over the first year, I, um, I identified the role in the org chart, what skills were necessary. I went out and found the right ideal, uh, person. I cold emailed, I did phone calls. I closed them. I gave them offer letters. Um, I wrote their 30, 60, 90s. I brought them in. I worked on them with a shared vision of what to do.

17:45And I worked with them day to day next door to them. I literally, I literally sit right there on the floor with everybody else. And I attend every engineering meeting and I work with them on designing the product and making those trades locally on speed and timeline and what to do and help them help people ship. Entrepreneurs, you heard it here. Just, just follow the plan. Yeah. Just follow the plan. It was just a lot of hard work, but we have, um, we have, you know, you really want to build an incredible team and you want to build a really great product. And then you want to get that product to market, that's really great.

18:16And you want people to really like it. And then you want them to keep using it over time. Like that's the secret. And so everything starts axiomatically with like, where's that team and how do you go get them? So, um, it was a brute force effort. Great. My favorite, my favorite type of strategy. Aggressively optimistic. How far away are we from like, you know, companies buying robots en masse, deploying them? So companies are, you know, paying us now for robots as we're like delivering them this year. We hope over the next 12 to 18 months, our robots are in our clients' test, like real facilities, doing real work and real work sales, getting paid for it.

18:55And I feel decently confident. For our audience, what's a work sell? Just, you know, like I'm at BMW and I'm supposed to be, you know, moving a bin or box. This is your work sell. We're moving bins over here to a conveyor system. That would define a work sell. Like I need to be in this location. I need to be doing this job. It's basically a job or a task. So I feel pretty confident in the next 12 to 18 months, we'll start doing those. And then we got to make it reliable, extremely reliable, extremely safe as we like, you know, branch out to hundreds and then thousands on, you know, inside of, say, a factory floor.

19:30And then we got to manage a fleet of it. And then we have to do AI training at scale. And then updates that to that fleet at scale. And then we have to manufacture a scale. No problem. Yeah, I think if we solve the robot humanoid doing these full stuff, the other things are extremely doable. It's extremely doable to take a robot that is reliable and that can do the performance and make a lot of them. A lot of these other problems have been solved before. They're just very hard versus new problem. Yeah. I don't want to wake up in like 10 years and say, we had the humanoid really work well. We just can't make enough of them.

20:04That just seems like, you know, we should be able to manufacture millions. I certainly don't. I hope we don't hit that roadblock. And then, yeah, I feel like in a mass way, I think because, you know, like the manufacturing volumes will like. The dependency here is getting the robot to do fully reliable, useful, continuous work. And once that as soon as that's happening, we're parallel pathing volume manufacturing. Figure operates as a really unique company, I think, or I've just not encountered many orgs that work this way. You are as vertically integrated as I've ever seen. You have your own actuators, you wrote your own operating system.

20:48Like, why do that when the project itself is so huge to begin with? The trade on like build versus buy, like should we use ourselves or should we go buy them is something we do at every single like component level. So it's not a philosophy of like we're going to build everything from scratch to begin with. No, no. I think our philosophy is we would rather, yeah, our philosophy is that we would rather not do it and we'd rather buy. But you ended up building a huge amount. Yeah, we ended up building everything. Yeah. We started buying by, like we bought, yeah, the mix. Basically everything but the GPU.

21:19It feels like that, yeah. Yeah, everything maybe besides like the battery cell and the GPU and CPU at this point feels like we've done what we're doing. To be clear, our default is to buy. Like building is extremely difficult. Any part that we have to go build, we have to have like, we have to put job listings out. We have to go hire people. We have to manage humans and make sure they're happy. We have to make sure their performance is good. We have to push, you know, push products. We have to check it. We have to integrate it in. It has to not break anything. We have to maintain it. When it gets broken, they have a supply chain to go manage.

21:50like um it's like it's a mess so eyes wide open like how do you end up there's no mature supply chain for what we're doing and there is no other option we have this philosophy that we have at the we do like a nine o 'clock stand up every day on the um like every morning like rain or shine and during like bring up processes we're bringing up like new robots and stuff it's a mess like the robots like never really work well in the first time everything breaks because we're like getting all the systems to start working together there are things on software and hardware that haven't communicated before.

22:21There's just nothing available that we could go buy that would satisfy our needs. So we've been forced to go build. And like design, and then, you know, in a lot of cases, we have a manufacturer. Yeah, I saw the machining. We did buy a decent amount of things. We found that we're like an incredible dependency for us. Last year, we had to go build teams around them to go fix it on both the sensor side and software side. Yeah, I don't think you can build a humanoid robots company without kind of going all in on all of it. Big question, but can you describe, like, if you want to run a hardware project, a hardware and software project like this, with this complexity at velocity, like, how do you manage product development?

23:00From, like, a thesis perspective, I strongly believe in, like, an iterative design approach. We really don't believe on spending a lot of time, like, just doing research and analyzing. We spend a lot of time on just testing, building and testing here. And that helps us really shake out all the problems. It helps us learn, helps us recursively add it into a continuum of product that's coming down, coming out. And so first, that's our strategy. We want to be continuously updating the hardware and software forever. I don't think it will ever be good enough for us. So we have a whole process built around building a robot from a basically hardware and software design that we run here.

23:46We first set out with understanding who are the customers, like what does a robot need to do? From there, we basically set requirements like, okay, we need the robot to lift this much pounds. It needs to run this long. It needs to charge here. The safety requirements so that it can't, battery can't burn down the building. There's a bunch of stuff we have to, the environment on IP rating has to be done on other actuators. There's just a bunch of requirements that come from there. From there, we look at those requirements and we do engineering design. And we have basically like three big phases.

24:20We have a conceptual and preliminary and critical design review that we do here throughout the year. The whole company is involved. So we have these like design gates that we work through. Similar practice that I instituted. Exactly similar. Well, similar practice. I instituted Archer from an engineering design perspective or philosophy. And yeah, we work through it in a very methodical way, like all the way through that serially. And how does integration and testing work in a way that's different from a software company since you've also done that? I'd imagine really differently. Yeah, we try to test and we try to prototype and test as fast as we can to see if we're right.

24:55Same with software. It just happens on a longer timeline. Okay. Well, software, you'll come in one day and I'll say, okay, we talked to the client. We believe we have all these things on the product backlog list we want to do. you'll somehow have some heuristics where you'll score those and you'll basically comb the backlog and you'll say, I'm going to go, we're going to add these like six things to the sprint. You'll do story points and you'll basically, you'll, you'll assign those out and you'll basically manage that whole process. And then you'll launch it and you'll get feedback, right? You'll try to either A-B test things, you'll watch the analytics and you'll say, did that work?

25:27Did that work? You really want to do that. And you want to have that kind of a scientific method around it. Say like, okay, was that, did that actually help, you know, fix this problem? uh same here we have the client we have requirements that we set like they need to do this we are designing things like we are designing hardware from scratch like and um so we take our we're designing an actuator we're going to take our CAD system and we're going to from scratch design it we're going to make uh assumptions on and trade studies on like what the different trade-offs are of how we can do it up front so we don't spend a lot of time designing something that just didn't work okay uh so we're gonna be pretty methodical about it like much more methodical than your software because the timelines are, you know, order of magnitude plus longer.