00:38:45 · Jonathan Stephens, Steve Xie
[00:38:45] S1: Yeah. And that makes sense. But as people adopt it and get more use, then we'll have a good leaderboard because I think that's something people like to know. Like, I have another I have a new frontier model. Where does it land amongst others? There's something beyond that one benchmark to use one or the other, but it it it even it's a playing field of where everyone lines up. [00:39:07] S0: Totally. Right. Yeah. It it is extremely important to let everyone understand where they are, not only to help them benchmark against themselves, say, maybe a day ago, a month ago, but also help them to, benchmark against, you know, the industry, especially the top players in the world. Right? So that's, in my opinion, the most important thing. And, you know, it's it's a lot of work. We are working on, I would say, scaling the benchmark because, basically, we are working on a lot... We are working with a lot of industrial partners. Right? And, also, we are working, we are getting a lot of data in order to build the most realistic, diverse, physically accurate, scalable benchmark in the world, but also at the same time, leveraging that benchmark to provide a little bit to help everyone understand where they are.
00:39:07 · Steve Xie, Jonathan Stephens
[00:39:07] S0: Totally. Right. Yeah. It it is extremely important to let everyone understand where they are, not only to help them benchmark against themselves, say, maybe a day ago, a month ago, but also help them to, benchmark against, you know, the industry, especially the top players in the world. Right? So that's, in my opinion, the most important thing. And, you know, it's it's a lot of work. We are working on, I would say, scaling the benchmark because, basically, we are working on a lot... We are working with a lot of industrial partners. Right? And, also, we are working, we are getting a lot of data in order to build the most realistic, diverse, physically accurate, scalable benchmark in the world, but also at the same time, leveraging that benchmark to provide a little bit to help everyone understand where they are. [00:39:59] S1: Yeah. Makes sense. Okay. So... Okay. So we have the the kind of the three different portions of infrastructure. We have a world, we have behavior of the evaluation. You kind of need all three. You can't you can't run simulation without doing all three. Right? You need to build a world, do some actions, train a model and evaluate how well it does. So you know where to improve it, or to create, get more assets, create more assets, try new behaviors. It's just almost like a flywheel or, you know, an ecosystem there. One thing that I've been diving into is world models and how how... Where does that fit in that? Is that something that we're gonna be using for multiple parts of that ecosystem or is that more for just world generation?
00:40:47 · Steve Xie, Jonathan Stephens
[00:40:47] S0: Yeah. It's basically everywhere. We are actually leveraging world model everywhere. And I really see that, you know, world model and simulation, they go hand in hand. Right? So it's fully integrated into, you know, our pipeline. For example, in the word generation. Right? In the word generation, you know, you are getting a lot of data from EgoSuite. So EgoSuite is Lightwheel's egocentric data product. Right? And we have already delivered more than one million hours of EgoSuite to the industry already. Right? And our goal this year is to hit more than 10,000,000. So for EgoSuite, it is bringing a lot of different type of real data. Right? The real scene, the real trajectory from human data collector, and also the real task. Right? And we are leveraging world model, for example, Marble, to actually digitize and transfer those real things into USDs and put into RoboFinals. Right? So that's basically how we scale RoboFinals, leveraging our EgoSuite, our real data collection, from the real world, and also through our real two sim pipeline. [00:42:06] S1: That's been a great unlock. Right? Like, I I Yes. Our team can be spending more time creating SimReady Assets and improving those aspects and less time creating the part of the scene we're not necessarily interacting with us, but Marble cannot put the backdrop that you still need to have.
00:42:06 · Jonathan Stephens, Steve Xie
[00:42:06] S1: That's been a great unlock. Right? Like, I I Yes. Our team can be spending more time creating SimReady Assets and improving those aspects and less time creating the part of the scene we're not necessarily interacting with us, but Marble cannot put the backdrop that you still need to have. [00:42:21] S0: Exactly. You don't want time creating. Exactly. Yeah. So so I would say that that's extremely important. And also in, for example, the behavior layer, meaning the data layer, right, we have, you know, the auto data gen, which is, you know, like, leveraging our sim words, sim things, leveraging our, like, automated data collection algorithms in simulation, and also leveraging word models. For example, COSMOS to augment them, right, to help us, to provide, more data for our customers. So that's also where we would use, you know, the world models.
00:43:02 · Jonathan Stephens, Steve Xie
[00:43:02] S1: Mhmm. Yeah. I feel like I feel like there's a little bit of a misunderstanding of what all world models can and will do. I think some people are really focused on the real world real time frame generation and, think like Genie three and you're walking around, but then there's Cosmos who's saying, Hey, I gotta have a robot doing interactions in this kitchen. I want the kitchen to be completely different, but I want the exact same interaction. So we can easily redo that without having to actually go there or even setting up a new sim environment. Or we have world mar... Marble, world lives as marble where you can just say, want a scene. We don't need exact replicas most of time. We just say, I need a kitchen of this theme, or I need a warehouse that looks like this specific kind of scene, but you don't, you don't need a one for one replica of a, of a warehouse because we're trying to build models where building models are generalizing across styles and scenes, not one. I don't... You know, I I need it working more than just one kitchen. So that's the great thing. Right? Can adapt out a 100 kitchens in a day where a three d artist would take a month or more to create a 100 kitchens in in in Isaac Sim. [00:44:11] S0: Totally. Alright.
00:44:11 · Steve Xie, Jonathan Stephens
[00:44:11] S0: Totally. Alright. [00:44:12] S1: Totally. Then moving forward, just kinda wrap this up. What what part of that stack do you think will look most different in, let's say, two years from now of those three, world or the world, have the behavior, we have eval, which one do you think is gonna really like change the most? What we're gonna see a big advance in? Or or is it gonna just have to be all three?