Dream Machines: Physical AI, Robotics, Being Human
00:35:48 · Evan Helda, Aravind Kandiah
[00:35:48] S1: Got it. Got it. [00:35:49] S0: And just double clicking into Charles' point because I think you brought up a really good point, which is real to him is sure it's a technological problem, but I think more importantly, it's actually a trust building exercise, which is how do you establish a set of experiments that both people can have, like, very high trust in? And, like, how do you incrementally grow that set of experiments? So it's like, okay. You know, we trained in sim. Let's have these 10 scenarios where we demonstrate it. And, like, getting all the stakeholders to actually align for these actual robot deployments in, like, mission critical areas is actually really tricky because there's multiple stakeholders that we're trying to bring to the table. Yeah. [00:36:31] S1: So I can can only imagine that that sales process, especially if that last guy who's responsible for the safety element in the funnel, putting the machine in the field, especially in these defense areas that you guys seem to focus on. [00:36:45] S0: But I think that's the... We, you know, we actually turned it around and it became our unique selling of value proposition. Okay. Right. So it wasn't actually a problem. It was actually the... That was the reason why people are coming to us because it's like, oh, actually, this makes certification much easier. [00:37:02] S1: That's compelling. Yeah. His anxiety was way worse before.
00:37:02 · Evan Helda, Aravind Kandiah, Charles Wong
[00:37:02] S1: That's compelling. Yeah. His anxiety was way worse before. [00:37:06] S0: Way worse before because, like, so many unknowns. Right? But now it's like, okay. Now I know what's tractable. I know how to approach the problem. Yep. And then you're getting all the stakeholders to sit around the same thing, which is like Yeah. [00:37:17] S1: Makes a lot sense. [00:37:17] S0: Okay. These are the set of scenarios. Right? [00:37:19] S1: He he now has a story to tell to his boss. Why does he gonna be safe? [00:37:23] S2: Exactly. [00:37:23] S1: He didn't really have have that justification before. [00:37:26] S0: And in my perspective, I feel like this is one of the biggest problems in physical AI, which is like, you have this system that has a promise of being able to do anything and everything, but no one is defining the bounds of the system. It's like, what can't it do? What can it do? And it says expectation mismanagement. [00:37:48] S2: Yeah. [00:37:48] S0: Yeah. And that's like the the big thing that we always see between, like, a vendor who's, like, building and selling the robotic system and the actual end user. And I said, oh, but you told me it should be able to handle this thing. Why can't it handle this thing? Right? Like, self driving, they got away with this by, you know, having levels. It's like level one, level two, level three, level four. [00:38:08] S2: Yeah. [00:38:09] S0: But we're still... Like, there's still some gray areas and, like, oh, actually, you know, it's Tesla's thing, this or that.
00:38:09 · Aravind Kandiah, Evan Helda, Charles Wong
[00:38:09] S0: But we're still... Like, there's still some gray areas and, like, oh, actually, you know, it's Tesla's thing, this or that. [00:38:15] S1: Seems like you guys might have to agree with a customer upfront on upper... Like, your own custom leveling for what they're doing and sort of have the same semantic understanding of that ladder coming together. Right? [00:38:28] S0: Exactly. L system is just a way for them to define that ladder and system. Interesting. [00:38:34] S1: Yeah. You need your own l one through l four. This is a way to define that and get there [00:38:39] S2: and benchmark the Like on Bifrost scenario library, here are the suite of tests you have to pass to qualify for so and so regulatory authorities level two. [00:38:47] S1: Yeah. Yeah. I wanna get into how you guys differentiate for other players out there. Before we do that, what's the biggest reason you guys lose a deal or customer? And I'm sure, eventually, we can talk about how you're feeling it, but like there's gotta be edges you bump into where the customer just doesn't go forward. Like, what... What's the most common failure mode there in your engagements today? [00:39:12] S0: Mhmm. I would say the most common one is actually the... They haven't visibly... They think they can get away with enough real data. [00:39:23] S1: Got it.
00:39:23 · Evan Helda, Aravind Kandiah, Charles Wong
[00:39:23] S1: Got it. [00:39:23] S0: Right. So they... They'll come to us and they're like, oh, no syntax data, then our our price point is a little high. So they'd be like, oh, it's... You know, we don't need this now. We could just, like... We can use real data. And this is typically when they have they have a good enough large amount of data, but they haven't faced the problem of actually trying to deploy the robot. Right? So they haven't faced the visceral, it's failing in production, pains yet. Right? They can deploy in a one constrained environment, and it's, like, working reasonably well. Mhmm. [00:39:56] S2: The irony is that, you know, synthetic data is incredibly useful, useful, you know, all the way at the beginning of the AI development process. When you don't have enough data, you need to bootstrap, you need to get something out for your first demo, your first POC. But the problem is that, you know, from a customer's POV or user's POV, the... Unless they've put in the cycles they've actually tried to deploy something, they realize it doesn't work, they need a better data approach, that that feeling of like, synthetic data is the answer here, doesn't doesn't sit as viscerally with them. Right? Mhmm. Yeah. You know, the best customers that we've worked with actually have tried to build simulation in house. Right? [00:40:33] S1: They've I know the pain.
00:40:33 · Evan Helda, Charles Wong, Aravind Kandiah
[00:40:33] S1: They've I know the pain. [00:40:34] S2: They've experimented with the tools. They've tried to roll it out, and they realized that, oh, this this is not something my team can just whip up overnight. I think this is something either... Like, whole company has to focus on. Right? And then they come to us and they realize then that the price point is very is very... Makes a lot of sense. [00:40:53] S0: Yeah. There's this weird graph, that you have is, like, essentially, there's like a life cycle of these, like, physical AI, robotics teams. And the very, very start when they have almost, like, little to no data, synthetic data, they love it. They're like, oh my god. It's great. Bootstrap. And then there's, like, a... It moves slightly further along, and they have, like, significant amount of data, like real world data. Then they're like, oh, I have enough data. I don't need synthetic data. And then they accumulate more and more, and then they try to deploy the system. And it's like, oh my god. Okay. I need synthetic data. So it's like, at the very start and at the end, it's like, okay. I need synthetic. But in the middle, you think you don't need data because, ah, I have enough. I'm getting, like, reasonable performance. So let's keep going. [00:41:34] S1: I feel like you guys need to create some kind of meme. Yes. Some kind meme. [00:41:39] S2: The meme is real. So... Yeah. [00:41:41] S1: It's like a synthetic data epiphany meme. [00:41:43] S0: Exactly. And like, you know, if you just look at the world's top, like, systems that have deployed autonomy. Right?
00:41:43 · Aravind Kandiah, Evan Helda
[00:41:43] S0: Exactly. And like, you know, if you just look at the world's top, like, systems that have deployed autonomy. Right? [00:41:51] S1: Mhmm. [00:41:51] S0: It is Tesla. It is Waymo. They're actually out driving the cars. All of them have huge internal simulation teams. Yep. Right? Because there's no way you can do testing at scale at the speed of software. Right? You wanna roll out a model update, you need to test a whole bunch of stuff, but you can't expect to just, like, test it on the road itself. Right? [00:42:16] S1: I might have to go [00:42:17] S0: quick to take a hard lesson. [00:42:18] S1: After this. I'm writing a piece right now about the future of simulation and and, like, robot ops in general. I'll probably have to put it in there. And I'll I'll I'll give you guys credit for it. Don't worry. Okay. So let's turn to the strengths where you're just like a hot knife through butter and you're winning super easily. What what what vertical markets is that occurring in and for what use cases and what kind of autonomous systems today? [00:42:44] S0: Yeah. Yeah. Charles, can you take this?