00:57:21 · Jordi Hays, Bernt Øivind Børnich, John Coogan
[00:57:21] S0: Okay. Take us through the full journey of developing hands for Neo. Lot of people in tech love to, you know, follow an an Elon style playbook, just make the most simple version of something, simplify, simplify, simplify. This looks incredibly beautiful, but also incredibly complex. And so I wanna understand, like, how you how you got here basically, like, version by version. [00:57:48] S5: So it is a very complicated hand, but in my opinion, it is the simplest version of it that exists that is good enough to do what needs to happen. So first of all, Onex is fully vertically integrated, so we do absolutely everything in house and in our factory here in California where I'm sitting right now. We do everything from designing the production processes to producing our own motors, our own tendons, the entire system, right, sensors, electronics, everything in house. And that allows us to iterate very, very fast Because I do really believe in, like, you have to have this first principles approach. Right? So start with, like, what what am I trying to solve? We're quite lucky in that we have, humans to look at with respect to how do you solve this problem. Nature did a pretty good job. [00:58:33] S1: So when you say you're you're you're looking at humans, what type of training data are you using? What's useful? What are you discarding from just there's a lot of hand videos on the Internet, I'm sure. People doing all sorts of things. You can do teleoperation. You can have people wear gloves and do motion capture. You can do simulation. [00:58:51] S0: X-ray.
00:58:51 · Jordi Hays, John Coogan
[00:58:51] S0: X-ray. [00:58:52] S1: Yeah. X-ray. I mean, there's so many different ways. Like, you're using everything. Is there one, one path that you found specifically, valuable?
00:58:52 · John Coogan, Bernt Øivind Børnich
[00:58:52] S1: Yeah. X-ray. I mean, there's so many different ways. Like, you're using everything. Is there one, one path that you found specifically, valuable? [00:59:02] S5: So I think, actually, it starts from, like, starting with first principles. Right? So, Womx has always been about we want to design robots that are safe so they can live and learn among people. Mhmm. But also because safety is what allows us to learn. So we probe the world for truth and of course if our fingers break while doing that or we break whatever we're trying to touch, it doesn't work. So you need to design this beautiful kind of like compliant soft systems that force can flow both ways because you're both seeing with your hands and acting with your hands. And that's really what this is all about. So like, how do you create that? And you kind of have to look at how nature works, like our muscle, nothing really moves fast. There's no gears, no nothing like this. So we've designed this from these first principles. But we go way deeper than that. I think what we haven't talked enough about yet, and we'll share more about this later, is like how incredibly seriously we take closing the gap towards a human. And it's not to look like a human. It's because we want the work system to work like a human. So even like if you look at Neil's hand behind me here, we worked so deeply on how do you make these fingers nonlinearly just like be compliant exactly like a human finger? Because if you get all these details right, you can take all of the video that's out there on the Internet. You can train
00:59:02 · Bernt Øivind Børnich, John Coogan
[00:59:02] S5: So I think, actually, it starts from, like, starting with first principles. Right? So, Womx has always been about we want to design robots that are safe so they can live and learn among people. Mhmm. But also because safety is what allows us to learn. So we probe the world for truth and of course if our fingers break while doing that or we break whatever we're trying to touch, it doesn't work. So you need to design this beautiful kind of like compliant soft systems that force can flow both ways because you're both seeing with your hands and acting with your hands. And that's really what this is all about. So like, how do you create that? And you kind of have to look at how nature works, like our muscle, nothing really moves fast. There's no gears, no nothing like this. So we've designed this from these first principles. But we go way deeper than that. I think what we haven't talked enough about yet, and we'll share more about this later, is like how incredibly seriously we take closing the gap towards a human. And it's not to look like a human. It's because we want the work system to work like a human. So even like if you look at Neil's hand behind me here, we worked so deeply on how do you make these fingers nonlinearly just like be compliant exactly like a human finger? Because if you get all these details right, you can take all of the video that's out there on the Internet. You can train [01:00:18] S1: Sure.
01:00:18 · John Coogan, Bernt Øivind Børnich, Jordi Hays
[01:00:18] S1: Sure. [01:00:18] S5: Huge role models based on this, and it just works on a robot. And that's what the VODEX role model lab is about. So to enable that general intelligence for robotics, you need to design the robot so that it interacts with the world exactly like a human. And then you wanna do that with the least amount of complexity possible, and that's essentially what we have here. But of course, the complexity of that is pretty high because you're you're mimicking a human head. [01:00:37] S1: Let's talk about grip strength. We got grip strength testers here in the studio. Is this an important benchmark? Clicking these together, how strong is the hand currently? Where do you wanna go? Because it feels like there's a trade off there where if the hand like, the stronger you make the hand, the heavier, the more, the more dangerous it could potentially be. At the same time, there's certain tasks that you expect a certain level of grip strength. [01:01:02] S0: Also yeah. Yeah. Yeah. It's it'd be interesting to understand how often tasks come up in your daily life where you need, like, insane grip strength. [01:01:11] S1: Yeah. Right? It's pretty rare, I think. But certainly in, you know, industrial capacity or even around the home, picking things up, moving a chair, you need to be able to grab it without dropping it. It's a safety issue at the end of the day.
01:01:11 · John Coogan, Bernt Øivind Børnich
[01:01:11] S1: Yeah. Right? It's pretty rare, I think. But certainly in, you know, industrial capacity or even around the home, picking things up, moving a chair, you need to be able to grab it without dropping it. It's a safety issue at the end of the day. [01:01:24] S5: Yeah. I think it it actually appears quite often, but you don't think that much about it because you don't do it for a long period. Like Yeah. Yeah. You're not grasping that hard, but then something starts slipping and you tighten your grip more like it, you're actually using quite a bit of force. So the hand is roughly the same. Yeah. That's a good one. The hand is roughly the same strength as an average human. Really? So and that's so, I mean, it needs to be able to do the full capabilities of a robot, right? So the robot can deadlift 150 pounds. [01:01:51] S1: Yeah. So the [01:01:51] S5: hands need to hold the bar of 150 pounds. Yeah. Not because deadlifting is useful in everyday life, but because it's a good metric for like how capable we are. [01:01:59] S1: Yeah.