Dream Machines: Physical AI, Robotics, Being Human
00:00:00 · Aravind Kandiah
[00:00:00] S0: The most interesting thing, I think I've seen so far is what I call the sim to sim to sim to real. You train on one type of physics engine, you show that it generalizes to another physics engine, and then you show that it can jump to another physics engine, and therefore, it improves their generalizability when you actually jump to reality as well. Because reality is just another type of physics engine. So you're training a policy that can generalize across these types of physics engine.
00:00:26 · Evan Helda
[00:00:26] S1: Dream Machines. Hello, and welcome to Dream Machines, the show about physical AI, robotics, and the future of the human experience. I'm Evan Helda, I'm here to learn with you about how to utilize these tools and navigate this transition. Because one thing is certain, as intelligence escapes the screen and enters the physical world, it's going to challenge our sense of self, of reality, and what it means to be human. So this show is not just about where this tech is going, it's about where it's taking us. If you enjoy this episode, please be sure to share with a friend and leave us some stars on your favorite podcast player. And for deeper dives on these topics and more, check out our newsletter over at dreammachines.ai. Love to have you join our growing community of thinkers, creators, and doers. Alright, folks. Welcome back to Dream Machines. If you wanna understand what it actually takes to get robots out of the lab and into production, this one's for you. Today, I'm joined by the cofounders of Bifrost, CEO Charles Wong and CTO Aravind Kandiah. Bifrost is one of the leaders in synthetic data generation and evaluation with simulation, and they have one of the most impressive sets of tools that I've seen out there to date for helping robotics teams train, test, deploy autonomy into the real world. So in this episode, we get into what most teams still get wrong with simulation and synthetic data, why real world data alone is not enough, why valuation is becoming the real bottleneck in robotics, and how the best teams are closing the infamous SIM to Real Gap. This one is... I can't miss in my opinion if you're building on the front lines. These guys are some of the sharpest founders building this space today, and they've been at it long before the physical AI hype train left the station. I found their insights super valuable and rich with signals from the front lines. So with that, enjoy this episode with Charles and Aravind. First off, how are you guys doing? What's new? How are things?
00:02:09 · Aravind Kandiah, Evan Helda, Charles Wong
[00:02:09] S0: Thing that never has been a better year for robotics. [00:02:13] S1: Hell, yeah. Amen to that. [00:02:16] S0: It's crazy. Right? Because we saw the whole arc. It was, like, from having to educate people on what, like, AI was to what generative AI was and then to what robots are. And just, like, living through that whole arc has been, like, a a nice journey. [00:02:31] S1: You guys timed it pretty perfectly, I think. I mean, you you started way before physical AI was a word. You started before robotics was as hyped up as it is. If I recall, think I saw you guys pop up on my radar, like, three years ago. Is that when you guys started roughly? [00:02:48] S2: About a little more than that. But, like, you know, the the... You know, when we started the company, it didn't feel like we timed it just right. Right? Like, it felt like we were too early. I remember we're pitching our first, like, round of VCs. A, what... They were asking about is AI. B, oh, oh, mean robotics? Robotics sounds very uninvestible. You're not even building robots. You're a dev tool for robots. And you're like, oh god. Okay. We'll get there. We'll get there. Right? And now it's paying off. So...
00:02:48 · Charles Wong, Evan Helda
[00:02:48] S2: About a little more than that. But, like, you know, the the... You know, when we started the company, it didn't feel like we timed it just right. Right? Like, it felt like we were too early. I remember we're pitching our first, like, round of VCs. A, what... They were asking about is AI. B, oh, oh, mean robotics? Robotics sounds very uninvestible. You're not even building robots. You're a dev tool for robots. And you're like, oh god. Okay. We'll get there. We'll get there. Right? And now it's paying off. So... [00:03:18] S1: I think that's part of every major success story is that they... When they first launched the business, it does feel a little bit too early. But the founders had the vision. They had the instincts that they were just early enough that there's gonna be the ability to... Opportunity to build the foundational capabilities to be paddling for the wave when it when it rises versus having to catch up to it. So kudos to you for that. Before we jump in here, what would make this a home run for you guys? I'm excited to share this out with your friends, family, colleagues, investors. What story? What points? What ideas do wanna land?
00:03:18 · Evan Helda, Aravind Kandiah
[00:03:18] S1: I think that's part of every major success story is that they... When they first launched the business, it does feel a little bit too early. But the founders had the vision. They had the instincts that they were just early enough that there's gonna be the ability to... Opportunity to build the foundational capabilities to be paddling for the wave when it when it rises versus having to catch up to it. So kudos to you for that. Before we jump in here, what would make this a home run for you guys? I'm excited to share this out with your friends, family, colleagues, investors. What story? What points? What ideas do wanna land? [00:03:57] S0: I think maybe one of the key ones is, like, what does it actually take to deploy robots? I think that's, like, the unique insight that we have. Like, there's, like, two groups of people. Right? There's the the folks, you know, in the labs. You know, now this year, they're starting to, like, slowly deploy. But deploying robots into the wild, what does it, like, actually take to deploy those robots? And I think, you know, being able to share that perspective with a lot of the customers they're working with and also, like, a lot of the investors and, like, hey, these are the... That unique perspective and, like, the edge and insight that we're seeing from all our customers who are actually deploying robots into the wild. I think that could be an interesting take. Charles, anything on your end?
00:03:57 · Aravind Kandiah, Charles Wong, Evan Helda
[00:03:57] S0: I think maybe one of the key ones is, like, what does it actually take to deploy robots? I think that's, like, the unique insight that we have. Like, there's, like, two groups of people. Right? There's the the folks, you know, in the labs. You know, now this year, they're starting to, like, slowly deploy. But deploying robots into the wild, what does it, like, actually take to deploy those robots? And I think, you know, being able to share that perspective with a lot of the customers they're working with and also, like, a lot of the investors and, like, hey, these are the... That unique perspective and, like, the edge and insight that we're seeing from all our customers who are actually deploying robots into the wild. I think that could be an interesting take. Charles, anything on your end? [00:04:40] S2: No. I I think one one is that, like, Evan, to your first point, like, like, we've been here, we've been building this and thinking about this for a while. Right? We're not just Gen AI groupies who jumped on when, you know, foundational models became a thing. Very much very much [00:04:58] S1: Didn't say groupies. I'm gonna steal that term.