#336 Byron Boots - He Turned a Polaris RZR Into a Self-Driving Military Vehicle
The Shawn Ryan Show
In this episode of The Shawn Ryan Show, host Shawn Ryan interviews Byron Boots, co-founder and CEO of Overland AI, a company developing auto
Key takeaways
- Autonomous vehicles reduce risk to soldiers by handling dangerous logistics tasks on the battlefield.
- Overland AI's success stems from integrating hardware and software into a production-ready platform for military use.
Main topics
- Autonomous ground vehicles for military logistics
- Overland AI's Polaris RZR-based self-driving platform
Notable quotes
"Autonomous systems are like any other tool. It's how you use them that matters."
Conclusion
Byron Boots presents a compelling vision of how autonomous AI systems can enhance
Transcript preview
Speaker 1 (0:00) If you spend any time off-road like I do, you already know how easy it is to end up second-guessing where you're at, whether you're still on a legal trail or if the route ahead is even worth taking. That's why I've been using OnX Off-Road. It's an off-road navigation app that shows trails, public and private land boundaries, places to camp, and detailed trail info all in one place. And what makes it really useful is the amount of actual trail data. Speaker 1 (0:27) You can check difficulty ratings, terrain details, trail photos, even recent reports from other riders before you head out there. The other big thing is you can download maps ahead of time. Because once you lose service out there, your phone is pretty much useless. But with OnX Off-Road, everything still works offline. And if you're riding with a group, their location sharing feature lets everybody stay on the same map so nobody gets lost or separated. What I like most about it is it gives me a lot more confidence when I'm exploring somewhere new. I'm not wasting time backtracking, accidentally ending up on private land, or trying to figure things out once I'm already deep into the trail system. I can plan ahead, know what I'm getting into, and spend more time actually enjoying the ride instead of worrying about navigation the whole time. Speaker 1 (1:17) Search Onyx Offroad in the App Store or Google Play. Again, that's Onyx Offroad in the App Store or Google Play. Most guys will spend money on gear, training, supplements, everything else, and then still wear the same uncomfortable underwear all day. If you're moving, traveling, working out, or sitting for hours, you feel it. That's why I like sheath. What makes sheath different is the dual pouch design. It keeps everything separated, supported, and comfortable so you're not constantly adjusting or thinking about it. And I like that sheath was originally built by an active duty operator who needed something that could handle long, hot, miserable days. So it was built to solve a real problem, not just look good in packaging. They're soft, breathable, and built for daily wear, travel, training, or long days on set. Feel it to believe it. Speaker 1 (2:14) Go to sheath.com slash SRS and use code SRS for 20% off. And sheath offers a first pair guarantee. So if it's not for you, you get your money back. That's S-H-E-A-T-H dot com slash SRS and use code SRS for 20% off. Sheath, the underwear of legends. Thanks to sheath for sponsoring the episode. Speaker 1 (2:44) Fire in Boots, welcome to the show. Thanks for having me. It's awesome to be here. It's awesome to have you. So I can't even remember. I think I actually found you guys on LinkedIn, which I'm never on. And I think I saw like a video on LinkedIn or something of Overland AI and then started found you guys on YouTube and started following you and then found out APC is an investor. And so. or ledger round and um so yeah i wanted to get in touch and love what you guys are doing looks uh you know i've had lots of tech giants on here a lot of drone stuff ceramic with the water stuff i guess they i didn't even know your guy just told me out back that they just hit the uh they hit a iranian port Speaker 1 (3:32) yeah pretty pretty amazing um so you know first of all it's it's an honor to be on the show and in that company i mean that's that's incredible but um yeah just in the last day or so there was news of uh ceronix boats being used um in an offensive operation in the state of hormuz wow yeah and so you're the ground guy that's right you're bringing up autonomous wheels vehicles to uh to land warfare so uh Really excited to dig into this. But let me kick it off with an introduction here. Byron Boots, you're the co-founder and CEO of Overland AI, a company building autonomous ground vehicles for the U.S. military. Before founding Overland AI, you earned your Ph.D. in machine learning from Carnegie Mellon, became a professor at the University of Washington, and led the winning team in DARPA's Racer Off-Road Autonomy Program. Speaker 1 (4:28) Overland AI has raised more than $140 million and become the first autonomous ground vehicle company to win a production contract with a fully integrated hardware and software platform. Congratulations. Your vehicles are already being used by military units around the world to move supplies, support operations, and reduce risk for soldiers in the field. Welcome to the show. Well, thank you so much. Got a lot to talk about here. It's been a minute since I've talked to somebody that's doing the kind of stuff that you're doing. But before we get going, got a couple things to crank out here. Everybody gets a gift. Oh, wow. Those are autonomous gummy bears. I'm excited. I love gummy bears. Right on. Right on. Speaker 1 (5:23) And I got a question for you. I got a Patreon account. It's a subscription account. And so they're the reason I get to sit down here with you today. And so they get the opportunity to ask every single guest a question. So this is from Thomas W. You've dedicated your career to advancing robotics and AI, including technologies with defense applications. From your perspective, What responsibility do scientists and engineers have to ensure these innovations ultimately reduce human suffering rather than prolong conflict? And do you believe AI and autonomous systems could one day become tools that prevent wars through deterrence and de-escalation? Or do they risk making armed conflict more frequent and easier to justify? That's a great, great question. You know, I think autonomous systems, they're like any other tool. The way that I think about them is really... Speaker 1 (6:24) a tool, a technology, and in the context of defense, it is something which allows a warfighter to be safer, right? So it reduces exposure, pulls them away from the point of contact, and then also potentially provides force multiplication on the battlefield. I'm sure we'll get into some of these things, but it is a tool which is used by humans. So, you know, how you use them is really, I think, a human question. So, you know, in the context of saving lives, I think that they will save lives for our war fighters on the battlefield. It's very clear how they do that. Speaker 1 (7:14) you know they also can serve as a deterrence like any like any technology might um you know if if we have significantly stronger um robotic systems at our disposal then you know we're going to be a little bit you know tougher to defeat in the battlefield and our adversaries will see that and so um in in that way you know they can certainly serve as a deterrence as well right on i mean yeah watching some of the models you showed me outside and then um you know, the the the videos and in and how they're going to be integrated in with the warfighters and in combat. I mean, it's, you know, is a is a former SEAL. It's like seeing what seeing the, you know, war, what war has developed into is just I mean, it's it's fascinating. And I mean, just. It's been over 20 years since I've been on the ground in a war and well, I guess not but still been over 10 years, but I mean I already have Speaker 1 (8:16) tons of questions and I can see so many different applications where this would be useful in just so many different scenarios. Yeah, we should definitely get into it. Before we do that, though, I do want to give you a gift as well, if you don't mind. So let me come over here and we got you a chair. Now, this isn't just any chair. Right on. This is... It looks kind of like an office chair, but this is actually a seat from one of the vehicles. So as I was explaining earlier, we pull the seats out of the Polaris Rangers and we turn them into those autonomous vehicles that we saw outside. Well, what do you do with the seats once you've pulled them from the vehicle? We make them into chairs. So we made one for you. Oh, man. Thank you. You can check it out. This is awesome. Yeah, yeah. Speaker 1 (9:16) There you go. Put this in the office. Thank you. That's right. Yeah, of course. That's awesome. All right, Byron. So before we get into everything Overland AI, let's do a little backstory on you. Where did you grow up? How did you get into this stuff? I mean, what's the backstory here? Yeah, I had a whole career before moving into defense tech. So I grew up outside of New York in Connecticut. I was into computers and was in the Boy Scouts and played sports and had, I think, a pretty typical upbringing. So, you know, that's maybe where things got started, just love of the outdoors and taking apart computers and playing video games and doing all the sorts of things that kids often do. Did you watch The Terminator growing up? I sure did, yeah. Did you? So... Speaker 1 (10:14) Terminator 2 was an unbelievable movie. And happy to talk about that a little bit more in the context of what we're building. But obviously robotics and science fiction were something that I really enjoyed. Were you a gamer? Yeah, I used to play, so back when I was in high school, I used to play StarCraft quite a bit. StarCraft 1, that was before StarCraft 2 came out. So real-time strategy games, I did a lot of, played a lot of games like that. We talked about Warcraft before. I used to play that too. So that was really what I was most drawn to. But yeah, I mean, love computer games. How, I mean... Speaker 1 (11:01) Well, we'll get into it later. I was gonna ask how similar, you know, is what what's happening today is controlling one of those games. We'll get into that in a little bit. So yes, where did you where did you go to school? What did you go to school for? So I went to college at a small liberal arts school called Bowdoin College. It's in Maine. And, you know, I spent four years there. I was a computer science and philosophy double major. as an undergrad. I started out really thinking about, you know, I love computer science, like I said, you know, all through high school. I also really liked history, read a lot of history. And so when I went to college, I was thinking about maybe double majoring in computer science and history. I thought it would be cool to have a more technical degree and something which is, you know, more humanities oriented with history. But I quickly Speaker 1 (11:56) you know, sort of figured out my first year that, well, I love history. It involved tons of reading and things like this that I really like to do, but also involved foreign languages. And you need to actually read about, you know, about history through contemporary sources and, you know, in the language that folks wrote in. And so it was something I was not great at. I was not particularly good at languages. I didn't really enjoy them. And so, you know, I started to, I took a couple courses in philosophy. I started out with a course called Logic and Formal Systems, which was really delving into formal frameworks for understanding arguments and analyzing arguments. And I really started to fall in love with that. And so, you know, I. Speaker 1 (12:50) brought that together with computer science and majored in both of those areas as an undergrad. Interesting. You were fascinated with the brain too, correct? Yeah, yeah, yeah. So I mean, in undergrad, I was taking computer science and took courses in artificial intelligence and started to do research in robotics. And this is back over 20 years ago. They had artificial intelligence courses 20 years ago? Yeah, it was pretty interesting. So my university is a small school. There was only four faculty in the computer science department there. And at the time, computer science was seen as like an offshoot of mathematics. And so a lot of these smaller schools had combined departments with computer science and mathematics. And you take a lot of courses in both areas. But at Bowdoin, two of the four professors were actually Speaker 1 (13:46) folks who studied artificial intelligence. So it was something, you know, it's been around for a long time. I mean, people were working on aspects of AI, you know, back in the 70s and 80s, but it was just starting to kind of come to the forefront and be an area that was really starting to accelerate around, you know, 2000 when I was an undergrad. So I started to study AI there and then in philosophy I was thinking about things like philosophy of mind and philosophy of science and you're just kind of trying to understand how the the human mind worked. Well, yeah. Speaker 1 (14:27) I think that's an interesting discussion itself. Yeah. I mean, how deep did you get into that before you kind of switched gears? Yeah. So, again, as an undergrad, I double majored in computer science and philosophy. And when I graduated, I was thinking about what I wanted to do next. I initially took a job at a robotics company. I was an engineer. Speaker 1 (14:55) working on problems related to perception and mapping and robotic systems. So these were mobile robots, much smaller than the ones that we just saw outside. So robots that are about, you know, about this big that moved around inside of buildings and you have to determine where they are and how they get from one place to another and things like that. So I worked as an engineer working on those sorts of problems. But I was really thinking about Speaker 1 (15:23) you know, kind of like, what do I want to do next? I knew I didn't want to just be, you know, kind of working as a software engineer. I wanted to go back to school. And the question was, like, what area should I study? So I really liked computer science. I really liked philosophy. And one of the things that I started to think about was cognitive science, you know, just sort of how the mind works. So with AI, you know, you're trying to program a computer that... can almost think like a human, that can perceive the world, that can understand it somehow. And then in philosophy, you're really thinking through language and by writing arguments, thinking about how does the human mind work, how does it contend with reality, things like this. But the piece that I was missing was actual neurobiology, right? Like the human nervous system, the substrate of... Speaker 1 (16:17) of the mind. And so I decided that before I went back and entered into a graduate program, I needed to learn more about neuroscience. And so I managed to get a job at Duke University in a neurobiology lab studying human perception. So I worked as an engineer for about a year, and then I went to Duke, and I worked there for two years. And this was really, it wasn't a graduate program. It was just working in a neurobiology lab. And I was auditing courses on neuroscience and neurobiology while I was there, trying to learn, you know, how does the mind work? I mean, wow. How the human mind perceives the world. Yeah. Speaker 1 (17:08) How, I mean, did you, did that, is that helpful in what you do today? It is. It's pretty interesting. So the lab that I was working in was really focused on trying to understand how humans perceive the world. And so let me just give you an example of why this is difficult and interesting. So the, when you look out like at an environment like, you know, this room, there's light which bounces off of surfaces, it comes back and it moves through your eye and essentially is projected on your retina. So for each of your eyes, there's a 2D projection of light from the room. And the question is, how do you go from that 2D projection, that 2D image, to understanding what's actually out in the world, like the 3D environment, the surface reflectance? Speaker 1 (18:05) you know, properties of things like the wall or, you know, the carpet or whatever. How do you sort of solve that problem? And it's called the inverse optics problem. So it's the notion that you have a 2D image and you're trying to kind of understand this complex 3D world. And the challenge is that there's actually not an easy solution to this. Because you're moving from essentially, like, three dimensions to two dimensions, information is lost. And so, another way to think about this is that an infinite number of different worlds could have produced the same visual image on your retina. And this manifests itself through illusions. So, there are certain types of illusions. There's something, for example, called an Ames room. Speaker 1 (18:54) where when you look at the room, it looks like a rectangular room, but in fact, it has this kind of crazy shape. It's something you can look up maybe later. But the interesting thing about that is just the fact that something that appears to you to be like a normal rectangular room is actually something completely different. That is just an example of one of these optical illusions. Now, the interesting thing about illusions is that they basically... Everything you see in some ways is an illusion, right? So they're not outliers. It's not like every once in a while your mind makes a mistake and you kind of see the world incorrectly. You are always inferring some world that is not quite what is actually out there. Speaker 1 (19:49) It's the rule, not the exception. And so this is, it forms almost a philosophical problem. It's like, if you are looking at the world, but you can't actually infer what generated the images that you see, how do you even interact with it? Like, how do you continue to exist if you're not seeing things properly? And so... you know, the conclusion, one of the conclusions that we came to was that really the way that you see the world is whatever way is necessary to allow you to continue to persist. So we kind of think about this as like you see the world in an evolutionarily sort of appropriate way, in a way which informs your actions so that... Speaker 1 (20:43) You kind of do the right things, you continue to exist, you can ultimately reproduce and continue on. And so, it's just one of the problems that we wrestled with. Now, what does that mean? It means that your perception of the world is really shaped through experience. You're just perceiving the world in the best possible way for you to take actions. And some of these... fundamental ideas actually carried through into the work I did in graduate school and even some of the things that we do today with the systems that we built, the robotic systems that we built. Very interesting. Wow. Wow. Speaker 1 (21:28) When we first started building the Sean Ryan Show storefront, we didn't have everything figured out. We had products, merch, ideas, and a growing audience. But turning that into something people could actually shop from and keep them coming back, that takes the right setup. And that's why we built it on Shopify. Shopify gives you what you need to start selling without having to piece a bunch of different systems together. Speaker 1 (21:53) You can build the storefront, manage products, take payments, track orders, and keep the business moving from one place. And if you're just getting started, you don't need to know everything on day one. Shopify makes it simple to launch your store and start selling in a few steps. They also have design templates and AI site building tools to help you get the storefront looking right without starting from scratch. And once people are ready to buy, ShopPay makes checkout fast and easy. which is huge. You don't want somebody ready to support your business and then lose them because you got a clunky checkout. I like Shopify's product so much. It's so simple. I had the CEO and founder, Toby, on my show. So if you've been sitting on an idea, waiting until you have it figured out, stop waiting. You can start simple. Learn as you go and build it from there. If you're ready to hear the... Speaker 1 (22:47) Of your first sale today, head over to shopify.com slash SRS to start your free trial today. That's right. Start your free trial at shopify.com slash SRS. That's shopify.com slash SRS. Speaker 1 (23:05) So where do we go from here? So, you know, I think, like, one thing, one of the, you know, when you think about building a robotic system, one of the ways that we kind of think about this is that when you're perceiving the environment, you're not just measuring it. You're taking the context and your prior experience, and you're using that to... predict what you think the world actually is. So for example, if you look out at a set of trees, you're not just measuring that there's some obstacles out there in front of you. You're also predicting that there's free space behind them that you can potentially move through. And you get to that point by essentially seeing Speaker 1 (23:53) lots of trees like in your past, right? Like you use that prior experience and then you can understand when you see a pattern like this, that actually means that there's, you know, sort of space out there behind those trees that you can then leverage in order to make decisions faster, to move more aggressively. That's really a machine learning way of thinking about things. You're using lots of data, lots of experience to understand what you're seeing in a functional way. and then use that in order to move a robot more quickly or more aggressively. And those notions sort of led me from neurobiology at Duke, like really kind of thinking about data and machine learning as fundamentally interesting things towards my graduate education, which I pursued after that at Carnegie Mellon University. Wow. Speaker 1 (24:51) But I've not heard anybody talk about that, that it's gotten into the machine learning AI stuff. That's pretty fascinating. Yeah. And I think, like, another thing which is pretty interesting here that I'll also highlight, you know, I was originally went to Duke to try to understand the brain and how it works. And my hope was that by understanding that, that would help me to maybe better understand you know, artificial intelligence or how to build machines and things like this. What I pretty quickly realized was that, you know, neuroscience is really hard, right? People have been studying the human brain for almost 300 years, and progress is slow. It's very difficult to understand how the brain works. We don't have a great grasp of it, even now. Speaker 1 (25:44) um we can describe a lot about it but not really understand it functionally and so um one of the one of the lessons from that was that um you know i almost came away with with the opposite conclusion so instead of thinking about the the brain as something that would help me to build better machines or understand you know build a better ai i almost think that focusing on artificial intelligence and mathematics and probability and statistics and information theory and robotics helps to provide a framework that people may ultimately understand the brain through. So it almost works the other way, that you have to really understand core principles of perception, planning, control, like these areas which are fundamental in AI and robotics, to understand, ultimately, you know, what the nervous system might be doing and be able to describe it. Wow. Wow. So, you worked at... Speaker 1 (26:56) You went to NVIDIA too as well, didn't you? That's right, yeah. So before starting Overland AI, I worked for about five years at NVIDIA. And this is while I was a professor. So after Carnegie Mellon, I got my PhD there in machine learning. I worked on robotics problems. And then I was a professor at Georgia Tech for five years and then University of Washington for seven years after that. And one of the things which is really cool about being a faculty member or a professor who is running a research lab is that you can also work in industry. So I had a research lab which was focused on robotics and machine learning, and I had a number of PhD students who were working in that lab. But you can take 20% of your time and work in industry simultaneously. And as part of that, I spent about five years working at NVIDIA on machine learning and robotics. How was it working there? Speaker 1 (27:59) That was awesome. I mean, I think, you know, I joined around 2018. So it was before NVIDIA was really kind of like a tier one, let's say, tech company. I think like Google and Microsoft were really up there sort of defining state of the art. But when I went to NVIDIA, I think there were a lot of good people that they were hiring. People had recently seen the power of GPUs, right, like this massive parallel processing. And I was thinking about that in the context of robotics. How can you parallelize tasks? How can you use it, not just these sorts of chips and this type of technology, not just for perceiving the environment, but also controlling vehicles? Speaker 1 (28:47) An example of this actually is carried through from work that I was doing at originally Georgia Tech and then NVIDIA and now to Overland where when that vehicle is out in terrain, so when our uncrewed vehicles are out there and looking at terrain, they're evaluating tens of thousands of possible trajectories that they might take. ranking each one of them, determining, like, is this a good one or a bad one, and then choosing how to drive after doing that. It's doing that about 10 times a second. And so how do you get that to work? Well, you can use NVIDIA GPUs to parallelize these tasks and evaluate many trajectories simultaneously and then decide how you're going to move based on that. Very, very interesting. Do you still miss being a professor? Speaker 1 (29:38) Well, yeah, I'm currently, I've got a 5% appointment at the University of Washington, which means that I'm there, you know, every couple weeks working with students. You know, I like teaching. I like interacting with students. I think that's one of the great benefits of being a professor is, you know, just engaging with students and people who want to learn. So that part's fantastic. And I miss doing that. I haven't been teaching recently since I've been spun off the company. But I think working in industry also allows you to really scale your ideas more. So there's only so much you can do in a smaller research lab. Yeah. And so Speaker 1 (30:29) In 2018, the Army Research Laboratory spotted you at an IEEE conference, demoing machine learning. What is an IEEE conference? It's IEEE, so it's Association for Electrical Engineers. It's one of the major types of conferences that folks publish in. So, you know, when you're a professor... one of your main goals is to publish papers, right? And those scientific papers further human knowledge. And so in computer science, the way that you do this is you publish papers at a lot of conferences. You work with with your graduate student you develop a new technology You then tell the world about it, right? You publish it in a paper and you do this on a pretty pretty you know fast iterative basis, so You know, that's this is one of the the major conferences in in robotics and we had a you know had some work there And you know some folks in the army were seeing what we were doing and had some Speaker 1 (31:37) cool ideas of how we could potentially like take that fundamental research and start to apply it to army problems right on right on and then the darpa racer Yeah. The crucible that birthed Overland AI. What was that? What's the story? Yeah, yeah. So, okay, so I was, you know, originally working at Georgia Tech and doing some work on ground vehicle autonomy. So the way that this started out, we took one-fifth scale vehicles, so these kind of smaller remote control vehicles. We put computers and sensors on them and made them autonomous. And we were trying to race them as fast as possible. So we were using data that we were collecting while we were driving these cars to learn a, what's called a policy. You think about it as like AI, essentially, like for the vehicle that could perceive the world and try to... Speaker 1 (32:39) drive really quickly. And these vehicles are doing things like drifting around turns and things like this. So they learn to do this, which is one of the things which is cool. Like, the vehicle's out there, it's trying to drive faster and faster, and then it's learning how to do things like drift in order to drive even faster. Oh, shit. So that's what the Army was looking at. Now, is it actually learning, or are you programming that into it? So it's actually learning. So you start by bootstrapping it. Like you have a human demonstrate, you know, this is how the vehicle should drive. And then it tries to replicate what the human does. And as it does that, sometimes it makes mistakes, sometimes it does well. But it's kind of grading itself. And then it will start to experiment. Like if I, you know, accelerate a little bit here or I brake a little bit there, does this make me... Speaker 1 (33:28) you know, faster or slower. And as it does that, it learns how to drive faster and faster and learns on its own. And so you're programing in the