00:09:16 · Anika, Maya, Priya
[00:09:16] S2: Right. Like everyone wants to do as much of it as they can. But realistically, not every company can do all of Yeah. [00:09:22] S0: Makes sense. And I wanted to talk about how this data all comes together into a model. Right now, what we see at least Figure do is they have like a dual model architecture system where one model is used for planning and one model is used for action. So the planning model basically looks around and it's like, okay, I wanna move that cup from left to right. And then the action model actually decides what the actual action is, meaning I'm gonna open up my left hand, close it, and then move it to the right. Woah. [00:09:47] S1: Sorry, I meant filming a day in the life, so I thought I would get you an action. [00:09:50] S0: Oh, okay. Well, now I'm awkward. [00:09:53] S1: Oh my god. Okay, ignore me.
00:09:56 · Maya
[00:09:56] S0: There's like one extra camera on me made me so camera shy. But yeah, so this is like one way to deploy a humanoid model or make a humanoid model. Another way is to build what they call a end to end policy learning model, which means it's just like one model that understands the entire environment. And then also decides what the action is going to be. The harder thing about making this model is that there's just not enough data to feed into making a model like this. Meaning there's not enough ecocentric data. When you have a dual model system, there's enough like third person view data. For example, like all the movies you watch, can... Or all the YouTube videos out there, that's all third person data. And that can be used to train like a planning model, which is really, really big. And just like understands contextually, like what the environment is around them. But then the smaller model, which is like the action model, you can basically make it extremely fast, low latency and specifically based on like POV, forced work, all of this like extra data that we talked about. It's just harder to now make these two models basically cohesive and work together at scale. And so then if you make it all into one model, then you have other challenges. And so we're seeing like humanoid companies in this space go after both. And I think sometimes the choices that they make is just like the lack of data. And so like what I'm really curious about is if you have all the data that you wanted in the world, what would be the better choice to go
00:11:16 · Priya, Maya
[00:11:16] S1: after? Yeah. Like what is actually the best architecture for like the end result that we're trying to get to? Yeah. So who's doing the second approach where it's like one combined model right now? [00:11:24] S0: Physical intelligence. Okay. Yeah. So their first version that they released is the combined approach. Is it [00:11:29] S1: like pie zero? Yeah. That's what it's called? [00:11:31] S0: Yeah. Pie zero. Nice. It's not... These things are not easy. Yeah. And the reason why it's not easy is because like LLMs are really good at generalizing. They're really good at taking the average. But when you think about an action, like if you wanna move a cup from left to right, there's like literally twenty hundred, maybe thousands of different ways to move... [00:11:49] S1: Think about like the average of how it would get there. It's like so funny. [00:11:53] S0: There's no there is no average. Right? There's it's actually there's like one... There's actually like an action and either works or it doesn't. And there's like a 100 actions at work and there's like a thousand actions that don't. Right? And so you need to actually produce like your output instead of being one general answer or one average answer, it should be like a 100 answers. And then you pick the one that works. And so like at the actual architecture layer, what this new model is doing is literally pushing out like a 100 options and then they're testing which option is the best one and then going and doing that option.
00:11:53 · Maya, Priya
[00:11:53] S0: There's no there is no average. Right? There's it's actually there's like one... There's actually like an action and either works or it doesn't. And there's like a 100 actions at work and there's like a thousand actions that don't. Right? And so you need to actually produce like your output instead of being one general answer or one average answer, it should be like a 100 answers. And then you pick the one that works. And so like at the actual architecture layer, what this new model is doing is literally pushing out like a 100 options and then they're testing which option is the best one and then going and doing that option. [00:12:24] S1: Yeah. Do you think like Figure... Because in my opinion, it seems like Figure is optimizing the model they're building for their hardware specifically versus a physical intelligence where it can be applied anywhere. Maybe they'll go after it themselves and have their own hardware, but it's kind of like choosing to do a hardware first approach in my opinion or like a software first approach. Do you think they'll all ultimately converge or do you guys think there's one that's like a better approach than the other to take right now? [00:12:48] S0: I think that... So I'm actually not really sure what is the better approach to take. What I do know is that the industry is not aligned on this approach. [00:12:57] S1: On what we should [00:12:57] S0: I be don't think anyone will be aligned until one player like does the thing. [00:13:02] S1: Like it actually works. And they're
00:13:02 · Priya, Maya
[00:13:02] S1: Like it actually works. And they're [00:13:03] S0: like, okay, is the right thing. It's... It was real... It to me, it's like almost kind of similar to like the Waymo Tesla debate. I was just thinking I'm like, LiDAR, [00:13:10] S1: do you need... [00:13:11] S0: Do you really need LiDAR? Do you really not need LiDAR? And I feel like this is similar in the robotics space of, like, do you need a diffusion policy model, which is this end to end learning model? Or do you do a dual system architecture? Because it's faster and it's just, like, better practically. [00:13:24] S1: So... Yeah. Because Tesla has one model. Right? And then Waymo has... Well, I guess, actually, Waymo, they just combine their model to also have one model. Yeah. So that's interesting because maybe it is. So if there's anything I [00:13:36] S0: had to guess, I would assume it's one model. Yeah. Obviously, because of, like, industry standards and, like, what we see in other industries. But it's just it's it's an easier thing like when you when you just have less moving parts in any engineering system, it is easier to reason about. So people always prefer one or like less. Yeah.
00:13:36 · Maya, Anika
[00:13:36] S0: had to guess, I would assume it's one model. Yeah. Obviously, because of, like, industry standards and, like, what we see in other industries. But it's just it's it's an easier thing like when you when you just have less moving parts in any engineering system, it is easier to reason about. So people always prefer one or like less. Yeah. [00:13:55] S2: And I feel like generally companies have good long term success with owning more of the stack rather than less. Right? Like Elon's approach also. Like, which is like different than what you guys are saying, but like another thing where like figure has that. And I think the reason why it might matter more for robotics than other things... Like, in my view, physical intelligence is making a bet that there's gonna be an open AI for robotics. Right? Mhmm. But then the thing about robotics is like the... Like we already said, the best quality data is from the robot. And so they obviously are getting that data all the time as well, but not... I don't think it can be, like, as tight of a chain if, like, you're having to integrate with a million different types of robots versus figure, like, they can actually change their hardware or their architecture based on the data that they find. [00:14:40] S0: I I... Well, I guess what I was gonna say is like, I actually do really think that physical intelligence is going to release a robot. They just haven't yet. Because if you look at their careers page, there's literally an entire set of careers just for the robots. [00:14:52] S2: Yeah. That might be the case.