Going Direct Conversations
00:02:07 · Chris Paxton, Michael Cho
[00:02:07] S0: That's a good question. That's a better answer. Yeah, that's right. I think Sarah must be. [00:02:17] S1: So yeah, I'm the odd one because I'm actually not a researcher. Chris is a really good researcher. I'm not even a researcher. But I've been really fascinated by this kind of like AI research for like a decade now. Literally, my hobby is for the last 12 years, let's say, or maybe even longer, I don't know, something like 10 years, sorry, something like 10 years, is to read papers before I sleep. I've been doing that for something like 10 years. What happened was, I think back in 2016, I was in the Bay Area and I crashed this course at Stanford called CS231N. which then was taught by this one PhD student called Andrej Karpathy. So I was there in the front row. I have no idea what's going on, but then I thought that he was just a phenomenal teacher. And since then I've been reading like computer vision, mostly CV papers. in the initial years and of course like in more recent years like robotics paper but that's that's how and then of course bump into Chris and one of those robotic conference and I think Chris is just one of those rare researchers who us I don't know how Chris find the time but he finds the time to share right and which I think is very rare and I thought Yeah, it's been a real privilege for me to do this with him. Anyway, I should stop yapping. Yeah, it's been a lot of fun. [00:03:34] S0: I'm pretty sure I was thinking about it where we met. We met at ICRA in Yokohama, right? So last year in Japan. [00:03:39] S1: Is that right, in person?
00:03:39 · Michael Cho, Chris Paxton, Jordan Wolfe, Jordan Wolfe
[00:03:39] S1: Is that right, in person? [00:03:41] S0: Yeah, I think that's right. I think so, yeah. Yeah, so a year and a half, right? That was... And ironically, also, I learned deep learning from Andrew Karpathy's lectures, too. So I think that's true of basically everyone of my generation in AI. So it's, anyway. [00:03:57] S2: Cool. And so, Chris, I'll start with you. So kind of your journey in the robotics space, you were at NVIDIA and Meta for a little while and now at Agility. Is that right, Agility? [00:04:09] S3: Yeah. So just kind of walk through kind of your kind of experience in the embodied AI [00:04:15] S2: space, getting kind of up to date. So I actually got my start in my PhD at Johns Hopkins University in Maryland.
00:04:15 · Jordan Wolfe, Chris Paxton
[00:04:15] S2: space, getting kind of up to date. So I actually got my start in my PhD at Johns Hopkins University in Maryland. [00:04:26] S0: Initially, we were doing surgical industrial robotics. We did a bunch of work on how to teach robots to do different tasks and try to make them easier to use and all this. And this was all before deep learning really worked. So it's all stuff that it's ironic looking back at it now, because a lot of it doesn't like It ended up being quite different, I think, than the things that actually we use these days. But then after that, I went to... Actually, during that, I missed one, which is that I did a co-op at Zoox, so I went to Zoox for like a year and a half. So Zoox is a self-driving car company. So that was... But then I did post-doc. And was there for a while, moved to Pittsburgh. And then I moved to Pittsburgh, then I moved to Meta. And since then, I've been at startups after Meta, which is, yeah, so I've done, I've worked in a lot of different parts of robotics over the years, actually. I mean, yeah, many different robotics fields and industries. [00:05:24] S2: I don't know. Is there kind of an underlying theme with kind of your research or what you work on that you enjoy doing across the different robotic systems? Or is it kind of pretty wide ranging?
00:05:24 · Jordan Wolfe, Chris Paxton
[00:05:24] S2: I don't know. Is there kind of an underlying theme with kind of your research or what you work on that you enjoy doing across the different robotic systems? Or is it kind of pretty wide ranging? [00:05:35] S0: I think it's been wide-ranging, but my research area, the area that I've always been excited about is long horizon reasoning and planning for robots. How do we get robots to do multi-step tasks? How do we get them to reason and be able to interact with the world? One of the things I was most excited about last year was working on the work on Meta and the work that I did last year was around home robots and reasoning about how can they reason about and represent their environments to be able to make long horizon decisions. What that means in practice is me being able to tell my robot, hey, can you go find me my drink and have it go and search the house and look for it and pick it up and bring it to you, stuff like that. Because obviously this is actually still fairly challenging for robotics, that kind of stuff. And I think home robotics is something that I've always been very excited about, passionate about, which is not quick. [00:06:29] S2: And Mike, so Mike, over to you. So obviously, outside of RoboPapers, what's the day job or what do you do every day?
00:06:29 · Jordan Wolfe, Michael Cho
[00:06:29] S2: And Mike, so Mike, over to you. So obviously, outside of RoboPapers, what's the day job or what do you do every day? [00:06:40] S1: Yeah. So, yeah, I'm the co-founder of this project called BitRobot, where we're trying to use crypto incentives to incentivize, well, gather resources to push forward like embodied AI research. I don't know whether I lost you at the moment I say crypto, it just made me like 10 times less serious. But I genuinely believe you. [00:07:02] S2: Not at all, but go ahead.
00:07:04 · Michael Cho
[00:07:04] S1: Yeah. So how I started in this space was I literally was just with my younger brother, Sam, who's my co-founder. So during the pandemic, we were just like... On the weekend, we'll... So he's in China, I'm in Singapore at the time. And during the pandemic, I want to say it. So it's the end of the world. We don't know what we should do, but we want to have a reason to keep in contact. So every weekend, we'll come together, build our own sidewall robot. It honestly just started out as a weekend hacking project to keep in touch with my brother. And then we decided to start a company because our observation was that a lot of sidewalk robots that were built by these bigger startups and the Amazons of the world, their robots are pretty expensive, like $10,000 or more. Because at the time, we were assembling our own. I have a pretty good handle on the build of materials. And so we thought that if you take the business worldview, maybe we already have the recipe, all this offline RL invitation learning stuff, which I think four years ago is starting to show some signs of life. And if you look at these robots as real-world robotic data collection devices, like $10,000 robot will just not scale, like won't get you the data. So we thought, okay, very naively, to be honest, at that time, me and my brother have never done robotics as a real, we are not roboticists to begin with. We thought that maybe someone should build a toy that's like $500, that's like a science. sidewall robot toy basically and make we thought it can be built at that price by just making a dumb robot so yank out the big gpu very cheap cameras basically and just have it 100 like human control so no ai so just a dumb robot this is during the pandemic right so the other of the fact that we made was that There was this huge investment that was brought forward, let's say, a decade so that we can have video calls like this. Right. So because before that, the Internet wasn't really geared towards like real time communication, but because of that. So we thought, OK, maybe it's possible for someone in New York to control a robot toy somewhere else. And that could be a really interesting gamification. So honestly, it all started off like that. Everything along the way, you know, it's a stumble into it after the other.