# A humanoid’s possible poses outnumber atoms in the universe

Markdown copy of https://realalpha.app/s/a-humanoids-possible-poses-outnumber-atoms-in-the-universe/ on Real Alpha, the best minutes from long interviews with the people building AI. Cite and link the page, not this copy.

- Speaker: [Brett Adcock](https://realalpha.app/builders/brett-adcock/) · Brett Adcock is founder and CEO of Figure. · Founded Figure AI to build general-purpose humanoid robots; previously co-founded Archer Aviation and Vettery.
- Host: [Shawn Ryan](https://realalpha.app/people/shawn-ryan/) · Host, The Shawn Ryan Show
- Show: The Shawn Ryan Show
- Episode: [Brett Adcock - Shawn Ryan’s First Interview with a Robot | SRS #292](https://realalpha.app/v/99pOdGEGu6s/), Mar 30, 2026
- Moment: 52:34 to 58:46 of the episode (6 min), playing on https://realalpha.app/s/a-humanoids-possible-poses-outnumber-atoms-in-the-universe/; the episode from there: https://realalpha.app/v/99pOdGEGu6s/?t=3154

## Quote

> “there are **more states in the robot** than atoms in the universe.”
> — [Brett Adcock](https://realalpha.app/builders/brett-adcock/) at [55:54](https://realalpha.app/v/99pOdGEGu6s/?t=3354) · The Shawn Ryan Show

## Takeaways

1. Figure’s humanoid has about 40 degrees of freedom.
2. The huge range of joint configurations makes hand-coded control impractical.
3. Whole-body commands update over 200 times per second.

## About this segment

Brett Adcock uses a humanoid’s 40 degrees of freedom to explain why hand-coded motion cannot cover its possible states. Figure updates whole-body control over 200 times per second.

From “Brett Adcock - Shawn Ryan’s First Interview with a Robot | SRS #292”, Mar 30, 2026. [Watch the full episode](https://realalpha.app/v/99pOdGEGu6s/)

## Transcript

Transcribed by machine (Whisper) from the episode’s audio, so a word may be off here and there.

[52:41](https://realalpha.app/v/99pOdGEGu6s/?t=3161) Humanoids? Let's... Yes. Let's do it. Yeah. How did this idea start? Yeah. So, um, you know, it's been like five or six years working on like a pretty crazy robotics work at Archer and, um, like the ultimate like meta problem in robotics space is can you, can you build like a general purpose machine to do everything in the world? like much of what, say, humans can in the world. And I have this big belief that, you know, we have a weird biological species.

[53:19](https://realalpha.app/v/99pOdGEGu6s/?t=3199) Like we look, we're like, you know, we have these weird hands and arms and legs and certain height and sensors. And then we ended up building this world around us so we can interact with it. I mean, if we get dropped into Mars today, we're going to build like coffee cups that we can hold and stairs and doors. And we're going to build this stuff again. And it's like the human operating system. We're building things we can use and operate in that makes it easy for our lives.

[53:44](https://realalpha.app/v/99pOdGEGu6s/?t=3224) And we built it around the form factor that we are. Meaning if we look differently, the world would look different. Our Expresso machine would be different looking. We might not even like Expresso or caffeine in this case. So we built this whole world around us. The holy grail for robotics is can you basically build a general purpose machine that can do what humans can, which for me is like a humanoid robot.

[54:07](https://realalpha.app/v/99pOdGEGu6s/?t=3247) And a humanoid robot is just a robot that has like a human form. So it has legs so it can walk upstairs and walk over like, you know, uneven terrain or say things on the ground and bend down, which legs are important for, or reach up. It has like arms. in hand so we can manipulate objects and do things like, you know, grab the stuff, open these gummies and, you know, fold laundry and do real work.

[54:36](https://realalpha.app/v/99pOdGEGu6s/?t=3276) And then we have the right sensors so we can, like, see the world and understand what to go do and use, you know, our biological neural net to kind of figure out how to reason from. Um, and, um, you know, having worked on like, you know, kind of like aircraft or, you know, now five or six years, I, I thought it was like pretty possible to go build, um, an electric humanoid robot and electric is important for cost and it's important for safety and it's important, uh, because the performance will be much greater.

[55:04](https://realalpha.app/v/99pOdGEGu6s/?t=3304) And at the time, I mean, one of the best humanoid robots then was probably like the Boston Dynamics Atlas, it had like a hydraulic system. It was like really heavy and big and high torque and very leaky, like the oil is everywhere. And also didn't run very, like maybe ran for 20 minutes on a single charge. So you needed to kind of radically transform the hardware. And then you needed to figure out a way to build like an AI brain.

[55:28](https://realalpha.app/v/99pOdGEGu6s/?t=3328) The humanoid is so complex. It has like, let's call it like 40 degrees of freedom. And degrees of freedom is like a joint. So like an elbow is a degrees of freedom. You know, shoulder's got three. A ball and socket has three, like a pitch-on roll. And our robot has about, let's call it 40 degrees of freedom in it. Each degree of freedom is a motor that can spin 360 degrees. So if you only look at how many positions the body could be in at any given time, like this is a position, this is a position, and keep moving, the amount of states.

[55:57](https://realalpha.app/v/99pOdGEGu6s/?t=3357) It's, mathematically, it's 360 degrees to the power of 40 actuators. So there are more states in the robot than atoms in the universe. There's more positions the body can be in. No shit. By far. It's a much greater number. Done the math a few times. Very confident in this, even though it sounds ridiculous. So you just can't code your way out of this problem. Like, how are you supposed to write code?

[56:22](https://realalpha.app/v/99pOdGEGu6s/?t=3382) Like, how's a human supposed to write C++ or code to tell the robot at any given timestamp what to go do? Like, if I want to grab this, I need to move my whole upper body and maybe lean over, and I'm moving my fingertips and my hand, like, my wrist and hand get in position to grab this. Like, it's an intractable problem for code. So, um... I mean, you were saying earlier, I'm going to butcher this, but it's updating the foot 200 times a second.

[56:53](https://realalpha.app/v/99pOdGEGu6s/?t=3413) Yeah, our controller is running... For balance. Our whole controller, so we have a main computer, is processing what to tell all the joints to do, 200, maybe more than 200 times a second, to make sure we can just balance and then we can like do the task. It could be like reaching over and grabbing this or balancing. If we run that too slow, we just like don't have enough feedback. Then we just fall over just like, yeah, we have to fully balance.

[57:16](https://realalpha.app/v/99pOdGEGu6s/?t=3436) You know, it's dynamic. So it's if you, generally if you power it off mid run, it's going to just fall down. It's not like a four-legged dog or quadruped robot where like any given point is usually statically stable. So it makes it very difficult because you have to be able to even move your hand. I'm moving my pelvis. And my whole body, my torso's moving, my head's moving, like all of it becomes very complicated now.

[57:39](https://realalpha.app/v/99pOdGEGu6s/?t=3459) It's not just like move my hand, it's like move my whole body to get my hand in the right spot. So every joint, all those 40 joints have basically position encoders. We know exactly what position the motor's at, or even the case of the knee or this. And we have force sensing, torque sensing on board. We have the ability to detect all the forces that that knee is seeing. It could be really high when it's walking, or it could be like you know it could be like it could be powered off and have no forces on the leg all of that feedback and is as being sent in the main computer and then we're telling uh all the joints what to do uh over 200 times a second some of the other feedback is happening at like five or six kilohertz so the force feedback is happening six five six thousand times a second to the motor

[58:25](https://realalpha.app/v/99pOdGEGu6s/?t=3505) control on board and we do the motor control the brain for all the motors has done it locally at the motor level because it needs to happen so fast that's being fed back to a main computer that runs a control software that tells the rest of the whole body what to go do at every time stamp to keep balance