No Priors: Artificial Intelligence | Technology | Startups
00:15:58 · Sarah Guo, Tony Zhao, Cheng Chi
[00:15:58] S1: Do you feel like you know something now that you didn't when you started the company? [00:16:02] S0: Absolutely. So I I think at the beginning, I'll describe it as, like, we see light at the end of the tunnel of... There are two axes. There's dexterity. There's generalization. When we add more data, things works better. And what this company about is the cross product of these two, how can we scale and have both dexterity and generalization. And this is something we're able to show in our generalization demo, which is that we can pick up these, like, very precise, like, forks, like, actual metallic forks, only ceramic plates with very high success rates. And, honestly, this is not something that we thought that would work so easily just by having so much more data. [00:16:42] S2: Yeah. So, actually, just wanna expand a little bit. You know, it's actually the process was long and painful. So there are so many issues, like just scaling up a robotic system is very, very hard. There are mechanical issues, reliability issues, there's data quality issues that come out of it. In the beginning, I actually thought it's gonna be much easier than this. But, really, just... It takes time and effort to grind out all the little details for this to work. I also think that if we compare TidyOp, it's much harder to get this system scaled up. But once it's scaled up, it's very powerful and very repeatable. [00:17:19] S1: So it is both harder than you thought it would be to get to here, and you are further than you thought you would be. Yes. Yeah.
00:17:19 · Sarah Guo, Tony Zhao, Cheng Chi
[00:17:19] S1: So it is both harder than you thought it would be to get to here, and you are further than you thought you would be. Yes. Yeah. [00:17:24] S0: And I remember in the beginning, we're having this, like, funny conversation of we're like, if we build this, someone can just, like, take our glove, and they'll build the same thing. Like, what mode do we have? Are we worried about that? And at the beginning, actually, we are a little bit worried because we thought, like, oh, you know, they can probably just replicate it. But as we go along the path, it turns out things are so much harder than we thought it was. [00:17:48] S2: There's so so many small demands. No. It's... Yeah. Yes. [00:17:51] S1: And when you say scaling up the robotic system, you mean the data collection to training pipeline and the hardware itself? [00:17:57] S2: Yeah. So actually, like, for for this... Yeah. For this to work at all, you need the data collection system. You need the robotic and control system to be able to deliver the hand to where we wanna go. You also need the data filtering pipeline and data cleaning pipeline and the training pipeline. And all these things need to be iterated together. So I actually gone through several loop of these. It's kinda hard to imagine without having a full stack team in house how this can even be done. [00:18:21] S0: Yeah. The glove we're using right now is... We call it, like, v five. [00:18:25] S2: Mhmm. [00:18:26] S0: And for v zero to v five, each version has, like, around 20 iterations. [00:18:30] S2: Okay. [00:18:31] S1: And So a 100.
00:18:31 · Sarah Guo, Tony Zhao
[00:18:31] S1: And So a 100. [00:18:33] S0: Yes. Yes. And, also, like, just when you make these at scale, right now, we have more than 500 people using these gloves in the wild. Like, all the things that could go wrong will go wrong. For example They they did. They did. Yes. For example, like, how things are assembled. If you don't specify exactly how it should be done, people will assemble it in creative ways. And the creativity doesn't help us here because, like, we really want the data collection device to be extremely precise. [00:19:02] S1: So you guys can't obviously know everything that's happening in every company in in academia and industry. But from what you know, how would you compare the scale of training data that you have today relative to the industry? [00:19:15] S0: At this point, we are almost 10,000,000 trajectories being collected in the wild. And those trajectories are not just like, oh, pick up a cup. It's these long trajectories with, like, walking, with navigation, and then, like, doing these long horizon tasks. [00:19:34] S1: Tony, as you mentioned, like, it's an open question, actually, what the right way to scale data up is. And so there are strong theories around teleop, around, like, pure RL, around video and world models. Like, how did you think about all of these?
00:19:34 · Sarah Guo, Cheng Chi, Tony Zhao
[00:19:34] S1: Tony, as you mentioned, like, it's an open question, actually, what the right way to scale data up is. And so there are strong theories around teleop, around, like, pure RL, around video and world models. Like, how did you think about all of these? [00:19:50] S2: Yeah. So from our perspective, actually, it's kinda somewhat surprising. So in the beginning, we worried that the data from GloVe or UMI like data that has higher quality but lower quality compared to TallyUp. Because for TallyUp, you're using exactly the same hardware and software stack between training and testing. It's perfectly distribution matched. But what we realized is actually this glow form factor encouraged people to do more text version, more natural movement, and those actually result in a more intelligent behavior on the modeling side. And in terms of, you know, data quality, we don't really see a difference in terms of, you know, how much... Like like, there's a gap between TidyOp and GloV data. [00:20:28] S0: After we did the 20 engineering, like, series method? Yeah. Yeah. Like... Because, like, apparently, there is a mismatch, right, that's in the camera frame, there's a human inside of the robot, and there are just a lot of things that we need to do to kind of convert a human data one to one to, like, as if it is robot data and have the model not being able to tell the difference. [00:20:52] S2: Yeah. And that kinda relies on, again, the whole... The full cycle iteration between hardware and software. [00:20:57] S1: What about RL?
00:20:57 · Sarah Guo, Cheng Chi, Tony Zhao
[00:20:57] S1: What about RL? [00:20:59] S2: We see a lot of great promise for RL in locomotion, and we think that will continue to be true for locomotion. So what we see really... RL as a method is very powerful, but it is much less sample efficient compared to imitation learning, and we see that to work great in environments where it's easy to simulate. For the case of locomotion, you only need worry about rigid body dynamics and rigid body contact between the robot and the ground. And, you know, because you engineer a robot, you know everything. But for manipulation, it's kinda hard for [00:21:31] S0: us to [00:21:31] S2: imagine, like, have... There's actually the same amount of diversity and the distribution of real object in terms of matching both appearance... Its its appearance and physical properties. And we think that it's gonna be challenging compared to global data collection and tally up.
00:21:31 · Cheng Chi, Tony Zhao, Sarah Guo
[00:21:31] S2: imagine, like, have... There's actually the same amount of diversity and the distribution of real object in terms of matching both appearance... Its its appearance and physical properties. And we think that it's gonna be challenging compared to global data collection and tally up. [00:21:49] S0: Yeah. I think it's really about which method can get us there faster. There might be different methods that will eventually get there. For example, like, you know, simulation world model. Right? And, like, it's almost a tautology to say that if I have a perfect world simulator, anything can be done there. Like, as long as you can do it in the real world, you can do it in a simulation, and you can, like, you know, cure cancer in a simulator. Right? But what it turns out for robotics is that just some some things are harder than others, and it really depends on the problem itself. So in the case of locomotion, as I mentioned, all we need to model in a simulator are point contacts with a somewhat flat ground. [00:22:33] S1: Mhmm. Like feet. [00:22:34] S0: Yes. [00:22:34] S2: Yeah.