[00:00:00] Stefan Seltz-Axmacher: Like the best robot in the world today, is less valuable for its owner than like an average computer was in 1960

[00:00:12] Audrow Nash: Alright. Hi, everyone. Let's start with intros. Stefan, would you introduce yourself?

[00:00:17] Stefan Seltz-Axmacher: Yep. I'm Stefan Seltz-Axmacher. I'm CEO and co-founder of Polymath Robotics. And, the the lesser good, co-host of automated, another robotics podcast.

[00:00:29] Audrow Nash: Alright, Ilia, would you introduce yourself?

[00:00:32] Ilia Baranov: And I'm Ilia Baranov co-founder and CTO here at polymath. And, I'm the silent, straight man partner of the, Automated podcast.

[00:00:40] Audrow Nash: Nice. All right. Hell, yeah. So tell me about polymath. Let's do a quick review. Stefan, you will jump in. Yeah.

[00:00:49] Stefan Seltz-Axmacher: Of course. Yeah. So, polymath, basically what we've done is we've take taken the the kind of commodity hard but unsought but, purchasable point to point navigation stack that every robotics team has to build over and over and over again, and we package it up as software that can be deployed acROSs any off highway robot. So we're we don't want to be a vertical robotics company. We don't even want to be the only autonomy organization in the world. What we really want to do is we want to take the parts of your robotics stack that you've probably built three, 3 to 18 times and make it so that you can just integrate ours instead of rebuilding your own arm. We're on tractors, bulldozers. We're too heavy for the DoD. As as neat updates from last time around the show. We're on more different types of robots than we have people on our team. And we can do everything from all of the software from point to point navigation all the way down to ripping out individual modules like a certifiable safety layer or localization or obstacle detection or or any one of those things that you don't want to build and isn't really core to, to your own program.

[00:02:03] Audrow Nash: Hell yeah. Ilia, anything to add?

[00:02:06] Ilia Baranov: No I mean Stefan, basically summarized it. It's, it's been it's been a journey to, to build the team out. I think we're overall doing well. Probably not as well as we'd love to do in our wildest dreams, but.

[00:02:19] Stefan Seltz-Axmacher: I don't think that it's going to be possible in a robotics. I don't think there's a physical area is physically possible in robotics. Do as well as your wildest dreams?

[00:02:28] Audrow Nash: Yeah, I guess it's not like that.

[00:02:30] Ilia Baranov: Or any tech business

[00:02:31] Audrow Nash: Some crazy unicorn startup. But you guys are also, I mean, trying not to have a vertical, trying to be more software focused where you integrate real well with hardware. Yep. It's like that to me from our last conversation. It's like it seems like a super good, like lifestyle startup in a way.

[00:02:49] Stefan Seltz-Axmacher: Yeah, it's it's definitely a nicer life than having to ship engineers to to west Nowheresville, on five hours notice to try to fix something. But but essentially, you know, a thing that's really interesting about this market and, and kind of, describe it more towards roboticists. But after my first robotics startup, I went and hung out with some SaaS companies, and I didn't, like, realize how much easier their life was. And it's it's not just the lack of hardware. They feel like a lot of people in robotics are like, okay, cool. Yes, this is harder than DocuSign, but like, I gotta buy actuators, I gotta buy sensors. There's regulatory stuff. Maybe it's like, that's actually not the why. Like SaaS is easier. SaaS is easier because like most people don't have to build most of the stuff. And like like if you're building a website, you should not really be doing backend stuff unless you're back at website. You should not really be innovating on collecting credit card payments. You should not really be be innovating on like 99% of the code base of that, that website or that that online service. Almost everything should be clear code at least until you're the millions of dollars that are. Whereas in robotics, kind of there's features that you've built in your last company, there's features that, disrupt their thesis about. There's features that some company just did a really cool marketing video showing off that they can do, which leads everyone else to think that it is as easy to say, make a humanoid walk up, grab a doorknob, twist it, and walk through the door and gently close the door behind them. When in reality, if you were starting with, if I sent you a Unitree humanoid today, it would still be a decent amount of work to get even to the point where you could do that, even if the hardware was out of the problem. And that's because of, of kind of this, this lack of being able to, to take components, from one robotics program to another. And that's kind of what we're hoping to partially solve.

[00:04:48] Audrow Nash: Yeah, I totally agree with that. And that's why a lot of web dev stuff feels so much faster because you, you don't have to write everything. Yeah. There's really good libraries for thorny specific problems like payments or other things like this.

[00:05:07] Ilia Baranov: Or or you just go, or you just go to the vibe coding. Right. Which Stefan has been experimenting with. Yeah. Some things come.

[00:05:14] Stefan Seltz-Axmacher: Oh. But yeah. Like, but but like, I mean, I think like and that's the, like the gap between where robotics actually is and like, where the people outside of robotics think it is, is only kind of seeming to get bigger as time goes on. Because people when you look at a web app, when you look at a SaaS application, any feature that you can you see, you can assume is commodity and easily replicable. The more money is cut into robotics, the more features people have seen and they've applied that same heuristic to it. And that's that's a.

[00:05:45] Audrow Nash: Really interesting.

[00:05:47] Stefan Seltz-Axmacher: Interesting thing.

[00:05:49] Audrow Nash: Yeah. So I agree with you. And I feel like it's hard I mean like seeing the Unitree robot do like a side flip. Yeah. And stuff. It's like pretty hard not to project a lot of competence onto that.

[00:06:04] Stefan Seltz-Axmacher: I saw you make I saw an interesting headline. And, and I said, I know this is a headline or a Reddit post, but someone saying Unitree is a scam. I bought a Unitree and I can't do some function, and I don't know if some function is walk through the walk up to the door, grab a doorknob, and walk on through. I don't know if some function is do a backflip. I don't know if some function is whenever you get an obstacle course. But they bought this thing thinking I've seen a Unitree do a cartwheel. If I buy a Unitree, it can do a cartwheel. Because that's how everything else works. It's only that's losers in robotics where you see a function happen, you're like, cool, I'll go buy a $40,000 robot. And if I do a really good nine months of hard work, I can replicate that. Yeah, this saw sort of. Yeah.

[00:06:56] Audrow Nash: For sure. So with this, you're saying everybody else does this? I understand it because or like the way I think of it. And I'd like to see Ilia your perspective, and then jump back to you, Stefan. But the, the way I understand it is that robotics is very early. We don't have standards yet. We still haven't understood all the hard problems you have. ROS and ROS solves some of the hard problems, but there's still plenty of hard problems, especially in the, like, behavioral space. Which is what you guys are doing. But, how do how do you think of this? Do you think that's correct? That it is because we're still early or or it's just harder or what do you think?

[00:07:41] Ilia Baranov: Yeah. We kind of go back and forth on this topic a lot. And if you really think about it, robots as an industrial manipulators, go back to around the same time that heavy ion computers started popping up. You know, you had the original hydraulic arms in auto plants come up around the same time as, as computers are really getting going. But computers have accelerated so much faster and made so much a deeper impact on society than robots. And it's a question of why, right? And in a lot of ways, I think a big job of computing and of UI is to get squish, the weirdness of the world, into a very specific framework that the computer can deal with. So like a screen with this type of web browser, with this kind of data or whatever, right. And a lot of work was done on here's a communication protocol like TCP, here's a standard for wiring, here's a standard for CPUs so that everybody can enter inter process. Whereas robotics never really had that push. And so industrial arms made by one company were drastically different than industrial was made by another company. And up until very recently, if you bought like a ABB arm and were an expert in installing them, you'd know next to nothing about installing a Kuka arm. Like basically you're useless. And in fact, you had to unlearn a bunch of stuff because that's the way they do things. And it's only really in the last, let's call it ten, 15, maybe 20 years that we started to have this kind of standardization layer appear in robotics. If I had to guess, I think one is because you don't have this nice screen box, you can put a lot of your interaction and code in you. There wasn't a good way to standardize on physical interaction with the world. If you even just think of grippers. Right. Like, there's no there's no good way to say this gripper will work in every single condition because it's either too expensive or too fragile or not precise enough. Right? There's always some trade offs where some like monitors like we've had monitor standards from us for 50 years. Right? Like there's no there's no magic there. And so I think it's taken time to get the fundamental tech energy efficient enough and complex enough and powerful enough to the point where we now have enough extra compute power to add on this kind of standardization layer. Because anytime you standardize something, anytime you abstract something, and we do that a lot, you inevitably pay a performance cost. Like when I was working at Amazon on the little Astro robot, the Lego model of it right here, we had to squeeze every micro out of the CPU because there was a small CPU and we had to do a lot of complicated things. And so you just don't have the space to do a lot of affordances and or abstraction layers and those kind of things. And now we're starting to. Right, and I think, again, we're seeing a little bit of this race of end to end neural models, and we can talk about that in a second, kind of again, being very custom in a sense of like, here's a camera input, here's the output, and magic happens in the middle and you don't know what it is. Battling off against more flexible applications, which can then be transported from robot to robot.

[00:10:51] Audrow Nash: Yeah I agree with you. So where do you think it's going. Ilia. And then Stefan I'd love to hear your thoughts too.

[00:10:58] Ilia Baranov: Yeah. Good question. I think I still, you know, looking at the history of technology and looking at the history of the internet and computers in particular, I think almost universally the most open source, most comprehensible standard wins and the most flexible one wins in almost every case. Right? So I think that a lot of the big neural end to end neural net models, which are starting to come out for machine learning, if they're open source and adjustable, that's a good first step. But I think especially in robotics, in this kind of idea of embodied AI, you can't, for example, take a Unitree demo, flip and put it onto an Atlas robot and like, you know, install suit up until the Atlas kickflip and like, there you go. Goes that doesn't exist, right? And I think that's hampering progress. So I think there's still going to be an underlying flexibility layer that we still have to build. And that's part of what polymath does. And other groups in introspection, in communication, in configuration management, these are all things that computers have more or less solved. And robot robots haven't. And so I think there's a lot of growth there. And then I think on top of that layer is these kind of big end to end models, which tend to be very specific per task or per robot or per environment.

[00:12:14] Audrow Nash: Okay. Stefan, what do you think?

[00:12:19] Stefan Seltz-Axmacher: So looking at, yeah, kind of the broader history, like, like related, but more from the business sense. Early crappy computers, like 50s era, 70s era, type of computers, which could be a massive amount of hardware expense, but create really, really big savings and really, really big upside, like a single piece of hardware could save you tens of millions of dollars. If that if that hardware is changing your supply chain or accounting for how you're spending money and you're a company the size of IBM or GM or whatever. And, and so the, the ROI was pretty big and pretty early. And that put pretty easy pressure on, like, hey, let's drive down compute costs. So not just GM can do this, but also companies who are only worth $10 billion and now companies are only worth a billion, and so on and so forth. Robotics has more problems than just build a compute. There's also batteries. There's also actuators. There's also sensing. There's there's that big long list of other things. And it is hard to come up with a robot, a robot that individually saves a company more than $5 million a year. Like the best robot in the world today is less valuable for its its owner than, like an average computer was in 1860. Probably even in just like, non inflation adjusted dollars. The math probably doesn't make sense. So like the pressures have been phenomenal on inflation. Yeah. Like like even if you don't include inflation like probably one of those IBM mainframes in 1960 probably cost $10 million. And for GM, save them $150 million. And like there's no robot that that is comparable.

[00:14:14] Ilia Baranov: We have a good example that, we can come back to in mining space.

[00:14:18] Stefan Seltz-Axmacher: So, so like the, the initial pressure for robots to get better, just like the math didn't make as much sense. In the 60s and the 70s and 80s, early 90s, and even in the US, I think a really cool thing happened for robotics, which is that a bunch of people made a crap ton of money on software, on web apps, on microservices, on organizing databases of information to make them quickly retrievable. And many of those people thought, hey, I'm a cool tech person. I got money to burn. Obviously a cool tech person should be working on robots. Definitely. And that poured money into things like the various self-driving car programs. And I think that led to a, a race, a series of races where people put so much money into this market that, you know, compute continued to get better and cheaper as it naturally did, but ruggedized compute became commodity. Sensor modalities got wider and more reliable, more usable. And they became cheaper and cheaper and cheaper as as massive amounts of, frankly, dumb money got poured in. Because because everyone assumed that, like, Moore's Law would work just just as quickly for self-driving as it did for, you know, microchips in the 90s. And, we now we're in a position where the components that you need for robotics have never been cheaper. There's, there's been 10 to 15 years of press release propaganda about how robots are imminently going to be here. So my hunch would be there's been more, more measurable interest in robotics in the last 15 years than the last 100 years put together. Probably like, and way more for us. An interesting thing. It's kind of happened where everyone assumes there's more robots than there are, and everyone assumes that should be more purchasable than than it is. So, you know, a big part of what polymaths businesses, frankly, is there's these big industrial companies who have said, hey, cool, if you can drive a car in San Francisco, I want to drive a vehicle in a Roomba like pattern in a weird part of a mine. 80 hours a week, nonstop. Who do I have to write a check to to make that happen?

[00:16:41] Audrow Nash: It's like no one exists.

[00:16:43] Stefan Seltz-Axmacher: Yeah, and that's that's what we're doing. A polymath, functionally speaking. And that's the that's the operations that we're doing. But I think like we the components for robotics have gotten cheaper. The interest has gotten to a fever pitch. And now it's a matter not of like, let's argue about which theoretical technology might be the most exciting ten years from now. I think we're actually now in a how do we as a market build products that seize this like incredible intersection of of conditions to make robots, frankly, a, a relevant industry at the global stage?

[00:17:18] Audrow Nash: Yeah. I think one one thing I'm curious about your perspective on and then Ilia, I want to hear your mining example, but the also I think that I see a lot is labor shortages across things, which I think is creating additional pressure. Yeah. On the factors you are describing now. So it's like it's hard to find labor. People are aging out. Yeah. What do you think of that?

[00:17:43] Stefan Seltz-Axmacher: There's. So labor shortages became pretty popular to talk about I think like post Covid. I think those of us in the white collar world realize that, we desperately need the people, the blue collar world to do stuff for, for our little tip of the pyramid to be so pleasant. But in reality, the sorts of industries that we primarily serve, you know, mining quarries, oil and gas, all the, logistics, railroads, all of them have been plagued by labor shortages for the last 50 years in, like, pretty structural, big ways. So, like, kind of fundamentally thinking about manufacturing, sku diversity is generally bad.

[00:18:24] Audrow Nash: What is that? I don't know what that is.

[00:18:25] Stefan Seltz-Axmacher: Yeah. SKU like, SKU like individual types of products that you have. Like when, when, when the iPhone came out, there was initially, you know, it only works with, one radio network and there was three battery sizes that were, sorry, three memory sizes. And that was all your options because that was a orders of magnitude cheaper supply chain to manage orders of magnitude cheaper engineering process to manage. Lots and lots and lots of advantages of we sell, you know, Henry Ford, you can have your car in any color you want, as long as it's black. Right? In, in industries like mining, however, labor shortages are so acute that there's been like massive explosions of vehicle diversity. And like, there are vehicles that you fundamentally different things like you can't replace an excavator with a bulldozer, you can't replace a bulldozer with a haul truck. Like those are those are different things. Totally. But what I mean is like caterpillar sells something in the range of 150 to 250 different types of bulldozer for, different engine sizes, different blade sizes, different configurations. All of that means and like in total, caterpillar has something like 800 different varieties of vehicle.

[00:19:38] Audrow Nash: That's crazy. Yeah.

[00:19:40] Stefan Seltz-Axmacher: Like GM, my 100 B has like 45.

[00:19:45] Audrow Nash: Yeah. Wildly different.

[00:19:47] Stefan Seltz-Axmacher: Yes.

[00:19:48] Audrow Nash: Magnitude different. Okay.

[00:19:49] Stefan Seltz-Axmacher: The, and what happens, we have all this diversity is every part becomes more bespoke, which means more expensive. What happens is every fleet ends up being more diverse, which means, you can't. You have fewer interchangeable parts. Which means there's parts that, like, are really hard to get.

[00:20:08] Audrow Nash: Just just to make sure I understand the reason that it becomes more bespoke is it it becomes more bespoke, more custom to make it so that you have things that are more suited to the job so that you can maximize the people that there are because of the labor shortages. Is that.

[00:20:25] Stefan Seltz-Axmacher: Correct? I need to basically have the biggest bulldozer that I can fit in these major operating conditions, because I can never get more people, and my mind can have a slightly smaller maximum size bulldozer than yours. Again. So you have a wild you have a different model than I do.

[00:20:40] Audrow Nash: So how how has that trend been changing over time? Has it been equally bad for the last you said like 50 years or something? We've known these labor shortages or has it gotten worse? Or how is it?

[00:20:51] Stefan Seltz-Axmacher: It seems like you there are only only more OEMs entering these markets, selling only more different types of equipment.

[00:20:58] Audrow Nash: So that implies it's getting worse and worse.

[00:21:01] Stefan Seltz-Axmacher: And like these, it's.

[00:21:02] Audrow Nash: More and more bespoke.

[00:21:03] Stefan Seltz-Axmacher: And these models, you know, if you're if you're John Deere, you want to sell people a new tractor as quickly as possible. So the the sale of replacement parts gets, sunsetting fairly frequently. Like, we know if we have organizations that are building in-house tractors because, it has a meaningful difference to their bottom line, to be able to amortize the, the vehicle frame for 20 years as opposed to ten.

[00:21:29] Audrow Nash: Wow. That's wild.

[00:21:30] Stefan Seltz-Axmacher: And all of this is because lack of having people. Yeah. And when you have all of these weird things, like going back to, like, the parts that you can't replace, the ultra class haul trucks have these really gigantic, you know, 3 or 4 meter in, 3 to 5m in diameter tires, that are 50 to $100,000 a pop. I have heard stories of publicly traded companies worth hundreds of billions of dollars buying spare tire tires from known criminals because it meant the vehicle got back to work weeks earlier than it would have. Yeah, it's wild. And that's because you have a weird special vehicle with a weird, special part if, in fact, you're just using a truck like you see on a highway, you go to the local equivalent of Jiffy Lube and buy replacement tire. The reason you can't do that is because you you can't get enough people to, to drive enough trucks to do that in that size vehicle.

[00:22:30] Audrow Nash: And so and so you think with robots, if robots are able to take over some of the labor. Yeah, we can have more general robots. And thus economies of scale kick in, cost comes down.

[00:22:42] Stefan Seltz-Axmacher: It kind of becomes a what is the most efficient size for hauling material out of a mine? And efficient starts to look like. Okay, where can we get the most, the most cheapest spare parts?

[00:22:54] Audrow Nash: Yep. For maintainability and things.

[00:22:57] Stefan Seltz-Axmacher: And that also, in turn, makes it easier to make these things electric.

[00:23:01] Ilia Baranov: Yeah. There's also the spares consideration here. Where, where when you have a critical piece of equipment. This is common in farming a lot. Is that your crop that you're growing is not only your crop, but you're in a geographic location where your neighbors are growing the same crop, too, and your harvest season is all at the same time. So your demand on your harvesters is like 0000 thousand percent zero.

[00:23:28] Audrow Nash: There is everyone's one widget is selling at the same time because everyone has peak demand at the same time. Exactly.

[00:23:33] Ilia Baranov: Exactly. And so as a farmer, it's actually a smart investment not only have one $2 million harvester, but have two $2 million harvesters just in case the first one breaks and it's that size and that cost because your most expensive thing is your person driving it. So if you don't need that and instead have ten $100,000 ones, if one of them breaks, yeah.

[00:23:56] Audrow Nash: Doesn't matter if one is out of commission. So what?

[00:23:59] Stefan Seltz-Axmacher: Yeah.

[00:24:01] Audrow Nash: Yeah. This this to me, sounds a lot like what Electric Sheep is doing where they have really small mowers as opposed to really big ones that were created by labor shortages because they wanted one person to maximize their surface area. Yep.

[00:24:15] Ilia Baranov: Exactly, exactly. It's the same thing. I'll go back to that example. You know, the size of this equipment and, and the labor issues also kind of caused this interesting story where we were talking to a mining group about deploying autonomy on some of their ultra class haul trucks. And, you know, we did the math on the business case of how much they'll save their improved reliability, those kind of things. And it came up to some number, I forget what it is, but let's call it like the $5 million a year make up a number. They talked to us, talk to us, and then they came back and they said, you know what? We're going to put this on pause because we did the math that if we just put a circuit in the thing to turn off the engine, if it's idling like an idler, same thing you'd have in a car. Auto start, auto stop, we'll save 30 dollars a year. Of fuel because the drivers have been idling their vehicles, charging their phone. Yeah, and air conditioning, of course. And so, you know, there's, there's those kind of efficiencies where all of the size and complexity has caused these weird distortions, where you need this really heavy fuel for very specific, large equipment that has to be shipped and parts and assembled on site and lives its entire work life on site. I it's buried on site, like all these kind of very, very weird, complicated things just to reduce the amount of people at the tip. So if you have a supply chain, you know, and do do things have similar concerns, right. The furthest deployed base or your furthest deployed mines is the same problem. Every nut and bolt that has to get there has to go through three stops to get there, and getting more and more expensive every time. And so the more you can reduce the amount of people you have at your furthest point, the smaller your overall it gets. And so every one driver we can save, we're saving probably ten ish people that are moving food around and providing lodgings and doing air and refueling. And so and we don't you know, Stefan and I don't suspect that we'll get a fully like fully autonomous as in no human beings, you know, there's there's no factories, for example, they're coming on the lights out factories where literally the overhead lighting is off because no human stepped foot in it. This is happening a lot in China, for example, for cell phone production. I think that's further along than what outdoor mining robotics is right now, where you still need a human to do maintenance and refueling and change the tires and fix the gears. You know, that got jammed, for example, all those things, you'll still need some level of human support, but every one person we can strip increases the value of their asset drastically.

[00:27:02] Audrow Nash: Yeah, the mine is that pyramid you're mentioning. That's a really good metaphor for it because like, you raise the top a little bit and then the like, there's a significant side added which goes all the way down. Yup. Support. People need support. People need support. People like H.R. all the things and support and.

[00:27:20] Ilia Baranov: Supply cost.

[00:27:21] Audrow Nash: Complexity.

[00:27:22] Ilia Baranov: Yeah. But some of these really far minds you know just to give you an idea, especially in Australia, are so far that not only are the people flown in and the spare parts, but there's fuel flown in. Right. They fly in jet fuel to refuel the planes that are flying. It's crazy. Right? Yeah, it's a hook.

[00:27:40] Audrow Nash: And they were like, that's the cheapest way to do it. I think that's the most efficient thing we thought of.

[00:27:44] Ilia Baranov: Yes. Wow. Yes. Yeah. They're not they're not stupid right. No of course you don't. You don't run a multi-billion dollar business by not being efficient.

[00:27:51] Stefan Seltz-Axmacher: But it's there's also like some interesting like game theory that there's some impacts where if you're going to set up a mine that size, it has to be that size like you have to you have to rip the whole piece of ground apart, things that that extend the life of the mine by six years but might be less environmentally destructive. Just don't cut it. Whereas if the labor becomes more fungible, if if that that stops being the stopping point. You can be a lot more strategic about where you go at.

[00:28:25] Ilia Baranov: And you could be more efficient generally. Resources gold, for example, there's no you know, people have this image of like a guy with a pickax knocking nuggets of gold off the wall, but that that doesn't exist anymore. Right? It's like the geologist has told us that there's one ounce per metric ton. We can.

[00:28:41] Audrow Nash: Find it.

[00:28:42] Ilia Baranov: Here. Or or more like 0.1oz per metric ton.

[00:28:46] Audrow Nash: Yeah. And then process it.

[00:28:47] Ilia Baranov: So yeah. You don't find it, you just. Yeah. You just process metric ton after metric ton after metric ton. And you need a certain size for that again, because of that pyramid. But maybe there's another site that is two ounces per metric ton, but it's only 50,000oz total. And so there's no point in building a site there because you're exhausted before you can ramp up, that all becomes more more possible with more automation, with less labor.

[00:29:12] Audrow Nash: Yeah, I think that's a very cool. I mean, that does seem like generally how technology diffuses. It's like it's starts super expensive, very difficult. And then it becomes commoditized over time. And then you get like web apps where you have payments as a service and stuff like that is kind of the end of that.

[00:29:33] Ilia Baranov: Yeah. And I think, I think for us, we're on this curve right now where we're, we're, we're trying to build kind of a standard. I hate to use the term operating system because it's a little bit overused. Everybody has their own OS. But like we're trying to build the basic autonomy layer and all the hardware abstraction to interface with your autonomous machine, such that then you can do it and go build the really fun apps, and then either the client themselves or the or other developers like Stefan is a good example on fuel versus yield usage. Which you can't do, right? Yeah.

[00:30:05] Stefan Seltz-Axmacher: Like like a lot of what I, what I would but I often jokingly referred to as like MBA porn about autonomy is is not is is pretty far disconnected from point to point navigation. So like if the, the, the, example Ilia was calling out is in a mind, you have a workspace where you're getting material from, you have a process or you need to dump the material. You can imagine that there's two routes in between them. One is more, one is faster, but less fuel efficient. Yeah, there's slower, but more fuel efficient.

[00:30:37] Audrow Nash: And so it's an optimization problem to see which is better.

[00:30:40] Stefan Seltz-Axmacher: Not not today. And today you just drive the way you drive. Whereas in a half day of work, a reasonable developer could code up language that says, you know, based on the prevailing price of, or based on the prevailing price of fuel, will route vehicles accordingly.

[00:30:58] Ilia Baranov: And like or even less you in less than two routes. It's just like your engine has an efficiency point. And so if you drive 10% slower and try try getting an operator to. right, like that's like no maliciousness. It's just as a human, you're not going to be you're not gonna be.

[00:31:16] Audrow Nash: Yeah. And you'll be impatient.

[00:31:17] Ilia Baranov: Two to oh, fuel prices higher. 10% lower.

[00:31:21] Audrow Nash: Feel like I just want to get it done.

[00:31:22] Ilia Baranov: Like even that speed adjustment. Yeah. Is not is not possible. Whereas is if you have an autonomy layer and you have a simple way to command it that becomes basically trivial. Like that MBA could.

[00:31:34] Stefan Seltz-Axmacher: my code with ChatGPT

[00:31:36] Ilia Baranov: Use ten lines of code. Basically I've like look up fuel price, look up oil price or or price, subtract, set the speeds and see which is lower.

[00:31:45] Audrow Nash: Pick one.

[00:31:46] Stefan Seltz-Axmacher: And and I think like our kind of thesis right now about like what's going on in the world. There's a lot of attention being sucked up by humanoids, by foundation models for robots. None of that matters compared to that. What we just talked about like like a fuel optimization, like what I just described for a company like BHP, which is a $280 billion publicly shared mining company that might individually give them 5 to $15 billion more year profit, more profit a year, like, wow. Like.

[00:32:19] Audrow Nash: Did you say billion?

[00:32:20] Stefan Seltz-Axmacher: Yes. What? Yeah. Like humanoids don't matter. I'm like, this is not like it's cool, it's neat, but.

[00:32:30] Ilia Baranov: Well, as as asterisk. Asterisk if BHP if BHP wants to increase their share price by being a tech forward company, I'm sure they'll they'll buy one of every humanoid. Yeah. Make them dance a little bit, have their stock price raise and that'll be enough. But the actual main drivers are.

[00:32:49] Stefan Seltz-Axmacher: And like we're in a state now where like for for industry the things that unlock that sort of efficiency hacking is, is now relatively trivial. It's now like pretty straightforward, where you could really just have a bunch of product manager types of people, vibe code out solutions that drive very, very large gains in efficiency.

[00:33:14] Audrow Nash: Since I am unsure if this is terrifying or what, but that's crazy, I do. I see what you're saying. What a thing. Yeah, so but it's it's really interesting to put robotics in this context or a lot of robotics applications that they are creating. Not that much for savings. Whereas these efficiency gains might be, hugely impactful, especially for these large companies. Yep. That's wild.

[00:33:48] Ilia Baranov: But you need the first.

[00:33:49] Audrow Nash: Because the robotics give you the monitoring and the ability to actuate it.

[00:33:55] Ilia Baranov: And that's the hill that a lot of robotics companies well enable in the control. Oh yeah. So, so so again another example by the way in agriculture is another example in agriculture is like five ish ten years ago. There's a ton of like egg data providers. We'll fly a quadrotor around with. That was super popular and it was great until it hit the barrier of like, okay, but I have to do with you, like, okay, I know that this particular point in the field is getting over watered. Want me to do about it? Right. And so you need to actually close the loop with a system that actually take that data into account and actually change your behavior of your planting your why. Yeah. Because remember. And so the same thing that it's.

[00:34:38] Audrow Nash: Just it's like quantified self stuff where you're just measuring but you're not changing stuff.

[00:34:43] Ilia Baranov: But you're not changing anything And so again like it's you need you need intelligent actors. And so again, this is kind of like AI agents is the buzzword. But basically it's like you need drivers in seats of devices that are better responded to data enough that they can do 100% per minute of their speed, or they're watering or they're planting or whatever, which humans do, and that's that's the piece to unlock it.

[00:35:11] Stefan Seltz-Axmacher: So we work with we're working with a pretty large AI conglomerate that, you know, is large and secretive. And if I, if I, if I named them, you wouldn't recognize them. But yeah, I social name them, they, they harvest a specialty crop, and they have, you know, a window of time three to 3 to 6 months, to get all of it off the field. And every year, they do not know whether they have three months or, you know, five and a half. They don't know that on January 1st. It depends on how rain happens in the first part, like in the growing season, the of.

[00:35:49] Audrow Nash: Changes during that time.

[00:35:51] Stefan Seltz-Axmacher: Changes. Changes on a daily basis changes.

[00:35:55] Audrow Nash: So you want to estimate basically and you want to do the intelligent thing based on. Now how likely it is that not you have to get it out quicker versus.

[00:36:03] Stefan Seltz-Axmacher: I mean like like they like like they're a super cool company and super data driven do a lot of really neat, analysis stuff. But like, as the season happens, they'll start to get a sense for how long it's going to be until crops start rotting in field. And we'll we'll make decisions on the fly about like we're going to speed all of our harvesters up by 25%, or we're going to drop a by 10%, or we're going to do this, or we're going to do that because that has a very substantial yield, impacts. They also like have to jump from one field to another because while they have hundreds of thousands of acres under cultivation, those hundreds of thousands aren't all next to each other.

[00:36:42] Audrow Nash: Over. Yeah.

[00:36:43] Stefan Seltz-Axmacher: So there's this kind of big, you know, traveling salesman, multi level optimization game that they have to play. And like some part of that is yell at guys on the phone with limited with limited language, to change their operations in real time. And that is the thing that robots can very quickly do and like, we think it's reasonable to think that that could lead to 10 to 20% more, more crops successfully getting off the field and also like even like down to the minute where that equipment might be in a field and it might leave that field before the field is fully harvested. I because they have to go to another field. But if you took in the data of the truck that's going to move that equipment from one field to another, and you took them that location to, to do a live, estimated time of arrival, you could theoretically be operating until just the moment that that truck arrives. You could change the the road that you harvest to make you be as close as possible to. Where are you going to get loaded into that truck, when it arrives? Because that itself might mean they get 2 to 5% more products off the field.

[00:37:52] Audrow Nash: Which makes a huge impact in terms of total sales.

[00:37:55] Stefan Seltz-Axmacher: If your costs and your fixed costs are the same, but you suddenly increase revenue by 4% through some data science shenanigans, that's the ball game.

[00:38:05] Ilia Baranov: Yeah, that's your entire profit margin. Basically.

[00:38:07] Audrow Nash: That's wild. Or it's I haven't thought about this.

[00:38:11] Stefan Seltz-Axmacher: Yeah.

[00:38:13] Audrow Nash: It's such an interesting way to view everything, and it makes a lot of sense to me. And yet so there's tons of these big companies and these big companies have very large inefficiencies. And it's partially just because people are in the loop and people are not you can't control people that well. Yeah. Robotics has covered with the or like the quadrotor stuff you were mentioning where they do a bunch of sensing. It's like, okay, sensing is not the only thing you need, because I remember this huge wave of quadrotor companies that had monitoring and they almost entirely have gone out of business, you know, and I think that is exactly why what you're saying, which is they couldn't change anything easily. And so this kind of thing, when you add robots and the robots make it possible to identify these things, you can make better. And then to quickly put in practice whatever is best.

[00:39:09] Stefan Seltz-Axmacher: Yep.

[00:39:10] Audrow Nash: That's incredible.

[00:39:12] Stefan Seltz-Axmacher: Yeah. I mean, like this AI group does like studies on how much more efficiently they can harvest a field if they eliminate left turns. But yeah, humans for that.

[00:39:22] Audrow Nash: Yeah.

[00:39:22] Stefan Seltz-Axmacher: Yeah, yeah. Then like, it's, but like, if you're, if you're ups and you want to try that out, you need a lot of data before you can slowly roll it out. And then there's a lot of change management to get rid of left turns. Whereas yeah, if you like, like we, we for example, were bidding on a program. I think this is obscure enough and, and, and unspecific enough that it's fine to talk really openly about. So we're bidding on a program to automate a vehicle in an asphalt lab. And basically they put asphalt outside in a ring, and, and drive a truck with about your weight over it. This particular.

[00:40:02] Audrow Nash: They look at tire wear or what do they do.

[00:40:04] Stefan Seltz-Axmacher: Think primarily where on the asphalt.

[00:40:06] Audrow Nash: The road wear on the asphalt. Yeah, that's a frickin lariats to just make a circle of asphalt and then drive on it until there's a.

[00:40:14] Stefan Seltz-Axmacher: Yeah. So the driving super variable. Yep. The people are variable. There's a truck that's been driving on this lab, 40 hours a week for the last 40 years. Right. But like, like if you want to experiment, in that lab, you need to write a test plan, that says, okay, for the next for the next six months, I want you to drive at 45 miles an hour, and do a hard braking event every 15 minutes or something. And then you need to coach the person to actually.

[00:40:46] Audrow Nash: Do do that.

[00:40:48] Stefan Seltz-Axmacher: Whereas, if you, if that vehicle suddenly autonomous, you could, you could operate that testing environment like you would medical testing equipment where you could be driving 24/7 there's a blizzard going on and you wouldn't want, you know, no humans should risk their life. Driving in an asphalt lab during a blizzard, I think is a easy, easily agreeable statement. Okay. But, like, you could have the vehicle do erratic braking, in a blizzard nonstop. You could have, I could be running one experiment, where I get every raining day of the year, and you could be doing a different naturalist experiment where we're driving differently, during daylight hours for for Aurora's experiment. And maybe Ilia is doing yet a different experiment about how how long it takes for there to be a change in the coefficient of friction after it's finished raining. And, and it's starting to dry out. And like, those three tests can very much be run in parallel. Those three tests can very much be orchestrated by piece of software. They can't be orchestrated by a person yelling at a truck driver.

[00:41:56] Audrow Nash: Yeah. Very interesting. Yeah. Because, I mean, basically what I'm imagining is you can tell a truck driver, you can tell a person to do these kinds of things. Yeah. The thing is, the compliance will be somewhat low. Yeah. And you'll see a lot of noise introduced because of that. Whereas you can have the robot just follow it and then you can more easily. It's like the the standard deviation of different outcomes, you see decreases, it becomes much pointier, much smaller distribution or much narrower distribution. Yeah. Okay. That's super cool. I love the idea of being able to command tests like that. So you can test. And you can also, especially if it's trivial to set up the tests, you can run more of them. So you get more data points, you understand more things and then you can optimize significantly.

[00:42:41] Stefan Seltz-Axmacher: I, I felt comfortable saying, yeah, what we're bidding for is because to my knowledge, there's like 100 of these labs in America. Like, I, I probably didn't say nearly anything that could lead you to guess where were benefiting that far.

[00:42:55] Ilia Baranov: Well, we we know it's not in Florida because of the blizzards.

[00:42:58] Stefan Seltz-Axmacher: but like, there's in in Florida, for example, other than at least 4 or 5 of these sorts of facilities where it's a test track for investing, blah, blah, blah. And I and a you don't need as many of them if you can get 150 hours a week of testing, as opposed to 35 or 40 for sure.

[00:43:16] Ilia Baranov: But I think I think coming coming back to the meta point, though, like, I think I think the storyline here of what we're doing is a lot of what I would kind of in quotation marks, but basic autonomy of moving a vehicle from A to B safely and successfully, and doing so in a way that makes business sense and that seems boring in a lot of ways. Like all of that to to most people. We talk to you though, like, yeah, of course like Waymo does it like y, you know, how hard is it to do it in a, in a space or, you know, the DARPA Grand Challenge did it ten years ago. What's the problem? And it's similar kind of discussion to, you know, one, one computer ran once, like there is a D.O.D. computer in the 70s that could run a simulation. So there's this kind of trickle down effect.

[00:44:12] Stefan Seltz-Axmacher: I played Grand Theft Auto once. Why can't The matrix be real?

[00:44:17] Ilia Baranov: Yeah. Yeah. Exactly. That's that's a similar discussion. Right. So I think I think tier two point, they asked at the start of the thing, I think there's a lot of effort and hype right now in AI and humanoids and those sort of things. But I think if you look back on history, the the lag between that being interesting and that mean, being a significant chunk of the market is much bigger than people think, almost always. Right? Like the lag between, you know, smartphones appearing as a concept and smartphones being common, which was one of the fastest ever, was still like six years. I'd have to go fast seven for everything, right? Right, right. But but tell that to a VC like that's the fastest tech ever, right? Here's my humanoid, which costs 30 to $120,000. The lag you're going to have from this to in every home is probably in the order of 12 to 20 years. So please invest $1 billion a year for the next 20 years and maybe will be one of the winners versus kind of Stefan and I have taken a much more cash efficient approach of like, where are the most expensive pain points that we can automate for the most amount of money, with the least amount of technical effort and risk now because we're lazy decision because like, that's how you actually run and efficiently.

[00:45:36] Audrow Nash: I view it as like, why do you need to make life harder? Yeah, just do the valuable thing. Like, yeah, you don't need to play life on hard mode. It's already difficult.

[00:45:45] Stefan Seltz-Axmacher: Especially when you, when you, when your life is with robots.

[00:45:48] Audrow Nash: Yes. Totally. Yeah. Let's see. Yeah. Ellie, I really like your point with all that. I wonder. Let's see, I guess. So with all of this, what I think is a lot of times people focus on the real sexy applications and things like this, or the sexy demos. But I've been believing increasingly that the majority of the value is in really unsexy tasks, and people don't get as excited for it. But the thing is, it's like budget. That's where you can make the biggest gains. Yeah, for these kinds of things. And like start to realize the promise of robotics. Yeah.

[00:46:28] Stefan Seltz-Axmacher: So and I think there's a challenge right now because in a way you can think about there being like two different robotics markets. There there's companies that I think of as like hyper robotics where the, the, the, the company exists because they're doing cool stuff, the money, because the thought is they could do more cool stuff eventually. You know, cool stuff has to be valuable, right? So if we wait long enough, it will eventually do something. And then there's a it's a thought. Yeah, it's a thought. It's a thesis. And then there's companies who are building, you know, robotics companies similar to how you would build a web app business or a SaaS company or an e-commerce site where it's like, here's a, here's a, a problem that the market has that I can reasonably build a solution for with a seed round. And now that I built a solution for them, I can get them to pay enough money that there's a line of sight to where I take in more money every year than I spend. And there's this. And I'm not just, you know, talking my own book here. There's a couple companies in the solar market, doing that. There's um, is a great example of that. And I, there's a bunch of that stuff, and, and I think that's taken me a while to realize is that jumping from hype over to realistic is like fraught with peril. And, and I don't know of many, many, robotics.

[00:47:56] Audrow Nash: Companies that have successfully made it.

[00:47:58] Stefan Seltz-Axmacher: Yeah. Because like, if you're worth $10 billion off of hype, like, like cruise, for example, I think at peak was worth $30 billion from hype. But when no one was willing to say that our worth $40 billion, that all they had to fall back on was a robo taxi business that was maybe making $4 million a year.

[00:48:20] Audrow Nash: Yeah, super low.

[00:48:20] Stefan Seltz-Axmacher: That $4 million a year or that $1 million a year was never going to cover their 3000 engineers. Huge burn rate, half $1 billion a quarter burn or whatever.

[00:48:30] Audrow Nash: What do you think of, like, Waymo with this? All right. Because they're I mean, they're still around. Yeah. And I hear people like them quite a bit. And I think they're.

[00:48:38] Stefan Seltz-Axmacher: Waymo's really cool. Go fund it. I'm really I'm really rooting for it. But let me let me put you in a scenario right. You are the CFO of alphabet, right? You have an equity plan with 1% or so of alphabet. So I haven't looked at I don't, I don't like, follow public companies that closely, but one, I think they're worth $1 trillion, maybe two. Easy $10 billion, of an equity incentive plan. If, you know, the the hype around robotaxis is, is up and down. But let's say some bad thing happens to Waymo. Public bad thing happens. And that brings down the Waymo stock 5%, even if it's just 5% for a month. So that means a loss of, $50 billion in value. And it's right around the time when you're being asked to write another 5 to $10 billion check for Waymo.

[00:49:41] Audrow Nash: Yeah. You just saw a 50 billion slide. You're like, even more than that, this kind of thing.

[00:49:46] Stefan Seltz-Axmacher: If you Audrow the CFO of of of alphabet, take that moment to decide. You know, I'm going to cut Waymo. We're going to publicly announce it. The stock's going to go down to 5% and go up another 10%. The company alphabet will suddenly be worth $150 billion more. And I as Audrow

[00:50:09] Audrow Nash: I’m responsible. Yeah.

[00:50:09] Stefan Seltz-Axmacher: No no no no no nothing that screw your responsibility who you are you. Yeah. As Audrow, we'll make an extra $1.5 billion. Yeah. Would would, Audrow would you keep Waymo alive for as cool as it is and how impactful could be in how many lives it could save? Four and $1.5 billion. Could I pay you $1.5 billion a couple? Look, look.

[00:50:33] Audrow Nash: I see how the incentives go. Yeah, yeah. That's crazy. Yeah, it makes sense.

[00:50:39] Stefan Seltz-Axmacher: I, I think it's an amazing product. I think it's really cool technology. I really hope the situation I just described never comes to a pass, but I have yet to play that.

[00:50:47] Audrow Nash: Game clear with anyone. This kind of thing.

[00:50:49] Stefan Seltz-Axmacher: Who would keep it personally alive. For what if I offer them $1.5 billion to be the person who signed a piece of paper to do less work to exit?

[00:50:59] Audrow Nash: Yeah, I know, to do less work. Yeah. That's wild.

[00:51:01] Stefan Seltz-Axmacher: Yeah.

[00:51:03] Audrow Nash: What a thing.

[00:51:04] Stefan Seltz-Axmacher: Yeah. So I really hope that I'm wrong. I really hope it's alive. But I think the challenge with these like hype programs and the challenge with these like big companies is they look incredibly stable from the outside because very few people are in the rooms where like it's like, hey guys, it's not it's not just that audio make 1.5 billion Ilia Also make 1.5 billion and Stefan and I'll make it easy 2.2 billion and I can be pretty good with that situation. Yeah for sure. And and yeah like we've seen, we, you know, aliens have been around, for long enough to see a bunch of really good robotics companies with really cool tech died because of assholes in finance who can get a good bonus check. And that's just a scary reality, which means, like, the faster you can get to a self-sustaining robot, the more realistic it is to actually have robots.

[00:52:02] Audrow Nash: Yeah, yeah, it has to make sense financially. And it's it's very hard to jump that chasm from hype to practical. Yeah. For this kind of thing. Do you, I'm interested in both of your perspectives. Do you think any of the companies will make it out of the humanoid bubble? Like, will we see any winners from it, or will we? Well, I mean, will Tesla quietly chug away at it and solve it because they have the whole Tesla business? Or what do you think? Maybe start with Ilia

[00:52:32] Ilia Baranov: Well, I mean, I think you got to you got to define what you mean by winner, right? Like, is Atlas is Boston Dynamics a winner? Because I eventually got traded around a bunch of times and ended up at Hyundai.

[00:52:43] Audrow Nash: Say they make that they jump from the they jump across that chasm from hype to reality. Where they are providing value that is justifying the cost of the businesses business.

[00:52:56] Stefan Seltz-Axmacher: Well, I think I think also what you're really trying to say is they make more money. I will well, there be any company who makes more money selling humanoids than they spend as an organization. Exactly. Not like his Ilia retort was Boston Dynamics has made a bunch of money, and a bunch of investors have made a bunch of money in Boston Dynamics without actually shipping robots at scale. Which should be fair, is a great trip that I a great trick that I'd love to be able to pull off. Yeah.

[00:53:22] Ilia Baranov: Yeah. No, I mean, you know, I, I don't know the internal numbers for something like Unitree, but they're, they're one of the few who are seemingly shipping and volume at reasonable prices. And I doubt their I doubt they're wildly profitable, but I also doubt that they're losing money on every Unitree robot they ship. It's like they're they're probably because, you know, the trick they also do is like, I don't remember exactly, exactly. But like the the humanoid robot, there's the 30,000 rises.

[00:53:53] Audrow Nash: Significant.

[00:53:53] Ilia Baranov: And if you actually do the one with it. Yeah. Like like the model that actually has sensors and compute. It's like 120, right? Like the 30,000 is a RC burns.

[00:54:04] Audrow Nash: Out when you look at it.

[00:54:07] Ilia Baranov: Right. So, so, so, you know, I think if I had to guess, you know, on the $30,000 model, they're spending 28 and on the $120,000 model, they're spending 100 and like something 110 maybe. Right. So low.

[00:54:23] Audrow Nash: Margins maybe.

[00:54:24] Ilia Baranov: They're making money. Have they made back the money they invested in it. Probably not yet. But again I, I just don't see the industry as a company that's willing to just burn money for no reason. So they're probably the closest I think all the other kind of startups are deep, deep, deep in the red, including including Tesla Optimus. Right. Like those are all into until your robot shows up on Robot Shop and you can swipe your credit card and buy now, like I don't. I don't suspect that they're quite there yet.

[00:54:54] Audrow Nash: Make sense.

[00:54:55] Stefan Seltz-Axmacher: I have, I, so someone told me that they knew of someone doing the strategy, but I have a strategy that I think would work in the humanoids which I think of as, I call it the minimally viable humanoid. And to be clear, you have to have you have to be big enough to build two robots to support the strategy. So, your first robot is your cool humanoid, and your your cool humanoid has legs. It has hands. It has, head. It looks like a head. Maybe there's little bit of face LEDs. There's a bunch of sensors in it. There's a bunch of compute on board. It can do really, really cool stuff.

[00:55:30] Audrow Nash: Yeah, it's the race car. Yeah.

[00:55:32] Stefan Seltz-Axmacher: For sure. Cost $3 million. But it's in every press release. The, the robot along side of it, is the minimally viable humanoid which is on wheels. Might even be tethered. Sure. There's probably a sensor cluster, but that sensor cluster isn't necessarily around. It's honestly in a head shape thing. It's kind of wherever the sensors should be. There might be two arms there. I'd only be one. The arms might have fingers. They might have something. It looks more like, you know, gripper.

[00:56:00] Audrow Nash: Like a pincher.

[00:56:01] Stefan Seltz-Axmacher: Yeah. And, and it's probably, you know, building materials is probably in the 35th. The 20 to $40,000 range. And the the whole point of it is it is easily deployable in smaller and non non massive factories, to do things like machine tending or, or deep, deep palletizing or getting things out of a box or into boxes and kind of rapidly be moved from one task to another with maybe with maybe only two hours of setup. And I think I've heard a company I wanted to say 1x might be doing this, but they.

[00:56:40] Ilia Baranov: Went the other way. They went the other way, they built this wheeled one and now they're building the full humanoid look.

[00:56:48] Stefan Seltz-Axmacher: Yeah. And I think that is that is a strategy where, like, people will show up to buy the cool looking one and then when like, yeah, the cool looking one is great, $2 million and has 20 minutes run time. The wheeled one, will fit your application and is economical compared to a human. Yeah, by one of the cool ones. We love. You buy one, but by 50 of the wheeled ones,

[00:57:12] Ilia Baranov: I just I also think, like on a meta level, the whole idea of humanoids sort of is, is sort of like a fallacy that in my mind, I feel I'm probably making a wrong prediction here. But like, you know, I looked at 1X video and like promo and they're like, 1x Thank you. And like the humanoid is moving a vacuum like a specific a Dyson, of course, a Dyson around somebody's house. And they're like, we're going for for home robots. Do you know what.

[00:57:41] Stefan Seltz-Axmacher: Home people have?

[00:57:43] Ilia Baranov: It doesn't look anything like a humanoid. It costs 100th the price and it works just great. I have three of them in my house right. Like. Right. And it's the whole, like, you know, we live in a world of horse drawn carriages. And somebody is like, I'm going to make a more efficient horse drawn carriage. It's going to be a mechanized horse instead of a car. Right. Like, I think that's the fallacy people fall into. And there's this thing like, oh yeah, but you know, the humanoid human world is built for humans. And like, it's sorta like, yeah, I know House has stairs, but it also has perfectly flat floor.

[00:58:15] Stefan Seltz-Axmacher: I actually rarely intersects with the part of the world that, that was built for humans and like so to be clear, what do you mean? Yeah. Humans were not humans did not come about, in houses with level floors and heating and protection from the elements and whatever humans came about in forests. Well, regardless of how you think we came about there. So it's I, you know, pick a particular side, but regardless of it, you need legs for walking in mud and sand and dirt and arms for picking up, you know, picking. You're saying.

[00:58:52] Audrow Nash: It doesn't have to fit the.

[00:58:53] Stefan Seltz-Axmacher: Circular spear? Yeah. You need. Yeah.

[00:58:56] Audrow Nash: I live the constraints that made us aren't what made the or that a robot needs.

[00:59:01] Stefan Seltz-Axmacher: I think there's a room who rarely leaves.

[00:59:03] Ilia Baranov: There's a reason wheels never in.

[00:59:05] Stefan Seltz-Axmacher: My life is between a series of flat surfaces. Like my my my house is a series of flat surfaces. The stairs are some more flat surfaces. It's flat surface.

[00:59:15] Audrow Nash: I feel like that is how most robotics way to. Yeah, but it looks like.

[00:59:19] Stefan Seltz-Axmacher: Which I am. I it outside of my ten minute walk between my house and our office. I am never more than 30ft from an outlet. Ever. That's just not a thing in my life. Yeah, I never outside of cell phone range. I'm not a big hiker, but also I don't need to buy a $10,000, device to hike with. That's. Theoretically, if I'm hiking, I should not be like bringing a computer on legs.

[00:59:46] Ilia Baranov: And I think I think I think again, like, the meta point is like, I think there'll be more robots in homes over time. And there are already. Right. And the robot vacuum clear makers are making units with arm to.

[00:59:59] Audrow Nash: Like, move simple.

[01:00:00] Ilia Baranov: Things. The ability to climb stairs right, right. And like and I think the home robots again the astro being an example but like a bunch of them I don't think need to be human shaped to accomplish their task. And so I think humanoids are very cool looking, but I think it's I think it's a fallacy, honestly. I think it's a fallacy until technology progresses a few more orders of magnitude and then.

[01:00:24] Stefan Seltz-Axmacher: Maybe, you know, like, like triple clicking on this a little bit. And this is kind of a new thought. So this might sound really stupid. A humanoid to a general purpose is a cool general idea, and it makes a lot of sense if you can only have one thing right. If you can only have one computer, it makes sense for that computer to be in your living room. It makes sense to have multiple user accounts so everyone in the family can log into it. It makes sense for it to have a bunch of central things. There's a bunch of things like that that go along with that. And humanoids are more significantly more expensive than the family computer was back in the 90s. What's happened since then? And I'm just looking around myself because I am an embarrassing person. With what we're about to talk about, we have infinite computers. We can have infinite, like, looking around me, I have two smart speakers here. I have, an Apple Watch on another Apple Watch right there, a phone, a laptop here, another laptop over there, an Apple TV, which is functionally another computer right over there, a smart TV that can also stream things. And I'm just looking at like the thing, I have a Kindle. You have been sitting on my desk. This is one of like four Kindles I own. I have an iPad. I have all the I have all these things. Yeah. I cannot imagine a world where a single $10,000 desktop computer, would be better than the series of devices I just listed off as being within eight feet of me. And similarly to Eliza's point about about, you know, roombas, for the, for for intelligent mechatronic devices that I might want in my household, it is hard for me to imagine a series of applications I'd want done by a robot that would add up to something I'd spend 5 to $10,000 on, but I can't imagine buying 15 $200 robots one by one over the course of six years. Anything one $5,000 robot that does all of my, like, eat applications that I care about a robot doing, that's probably never going to happen. That's like probably a nonstarter with my wife. Like, maybe, maybe if something really cool happens, a massive amount of money comes my way. Like, maybe I can blow ten grand.

[01:02:44] Audrow Nash: Yeah, but it's just that. Yeah, yeah, it's blowing that money on it. Yeah. The thing, because of the novelty.

[01:02:51] Ilia Baranov: Yeah. And, and I think, I think again, like humans. And we fall into this too, by the way. We fall into this fight like, all the time. Stefan and I had a discussion. Right. Probably a year ago now on, on our podcast where basically we're like, oh, you know, I want, I want the classic, like, go get me a beer from the fridge robot. And it was like, okay, so you need like an a robot with an arm that goes and opens the fridge and like finds manipulates, gets the beer, brings it back, whatever. And I was like, Stefan or like, and this wasn't in the house, by the way. This is this is for work. So it makes a little bit more sense, right?

[01:03:25] Stefan Seltz-Axmacher: Just because we drink in an office facility.

[01:03:28] Audrow Nash: Oh well, very very good. to clear up

[01:03:30] Stefan Seltz-Axmacher: Yeah.

[01:03:31] Ilia Baranov: But but like you know, and, and and he was like, overcomplicated this whole thing, I was like, Stefan, you could just have a vending machine. Like a vending.

[01:03:41] Stefan Seltz-Axmacher: Machine.

[01:03:41] Ilia Baranov: Has all the management and stuff one at a time. Yeah, yeah. No no no no no no. My point is, like, I, my point is like a Roomba a Roomba or an Astro or whatever. Like a robot. I can summon somewhere and a vending machine. Will do everything you want. Like you drive the robot under the ramp. You internet call the vending machine because they're almost all internet connected now anyway. It dispenses the beer you want, it falls into the thing, and it delivers your slightly shaken beer to your room for a fraction of the cost of any arm, or any manipulator, or any sensing or any humanoid you could possibly buy. Right? Like even cheaper than a Unitree robot.

[01:04:23] Stefan Seltz-Axmacher: Easily. Our other our other version of this, is because people always like the idea of robots for, the dishwasher, for taking the dishwasher. And I have exciting news for you. Although there's a robot you can buy. That's awesome. And for any of the listeners, for $400, purchasable at nearly any, any big box store, buy a second dishwasher.

[01:04:47] Audrow Nash: Yeah, totally.

[01:04:48] Stefan Seltz-Axmacher: Yeah. Suddenly, if you put all your dirty plates in one, take all your stuff, you just flop back and forth, just flopping back and forth and look at that. You know, you I just saved you. You 95% of your dish. Put it away, with technology and then you don't.

[01:05:04] Audrow Nash: Need space for storing them. Even better. Yeah. More efficient.

[01:05:07] Stefan Seltz-Axmacher: You already have space for storing your plates. So you just take one of your cabinets. Attorney general dishwasher.

[01:05:12] Audrow Nash: Yeah.

[01:05:14] Ilia Baranov: That's a little. That's a little.

[01:05:15] Stefan Seltz-Axmacher: Yeah. What's funny, though, is that.

[01:05:17] Ilia Baranov: I think the main.

[01:05:17] Stefan Seltz-Axmacher: Thing and having a big piece of metal go around my house. Yeah, for 10,000.

[01:05:21] Ilia Baranov: Exactly.

[01:05:22] Stefan Seltz-Axmacher: And scare them of Jesus out of my dog.

[01:05:24] Ilia Baranov: This one X group. Like I think again, looking at their press release recently, they're like, we're really targeting like we decided to target the home market as, like, why there's no I can't think of a reason other than you tried to go after the industrial market. Realize that it doesn't make sense and now have to pivot to get more fun. You could be in more hype because like in every other respect, the industrial market is so much more valuable and so much more forgiving of failures, like an industry that's testing a new technology will be okay if you drop a part somebody is testing in their home and you drop a plate. We'll put a negative review on Amazon instantly. Like it with no hesitation.

[01:06:07] Audrow Nash: Drop my plate. Right.

[01:06:08] Ilia Baranov: Like there's no reason to go after that market unless you just can't. Yeah, unless you just can't.

[01:06:13] Audrow Nash: Yeah. Okay. I wonder the thing that was interesting to me. I haven't been in touch with them in a while since I interned. Really. I mean, I saw them a few other times since, but, when I started with one acts as an intern. We were going after eldercare for this kind of thing. And that application, it was like remote. Someone would beam into it and, have it microwave food or something like this. Which the prices of assisted living are very high. Yeah. So that that made some sense to me. I like the mission there, but then it was surprising to see them pivoting a few times. And maybe it is a funding thing. I don't know, but it is funny how much you can simplify things with better design than just throwing a very expensive robot at it. A parallel that I have for my experience is, for the first year of this podcast, I was trying very heavily to use AI tools for helping me find clips, helping me, do editing, like the script will, it'll pick the person to be in the shot and things like this. And so I was trying so hard to get AI tools to work for this, and then I got so frustrated with it that I just hired someone. And they're not that expensive. They're absolutely wonderful. And it's a huge like the quality is far, far higher. And it doesn't have like weird artifact drastically better for this kind of thing. And similarly, yeah, of course hiring like so you could buy a say say it goes down to $5,000, $5,000 robots. Or you could just hire a housecleaner for this kind of thing, and they'll be excellent and custom work and things like that.

[01:07:57] Stefan Seltz-Axmacher: There are services that are like Uber for people to show up at your parents house to microwave food for them. Yeah. And those services are not that expensive. I yeah, like the services like approach like $10 a visit. Wow.

[01:08:13] Audrow Nash: That's crazy.

[01:08:14] Stefan Seltz-Axmacher: Which and like maybe I'm rounding down too aggressively, but not $100. Yeah. The math for that being a better outcome than by a humanoid is, like, pretty, not incredibly hard. For for that to work out, I think, like, there's a, a hard thing in robotics that that happens frequently is roboticists are a specialized skill set who generally don't want to go work on a SaaS app or or not or not for cats or whatever. So it's kind of like I have robotics, where can I put robots? With a little bit less thought to like, what is the best solution for for the problem at hand? And I think, like, I don't think any of us here are bearish. Is there value in robotics? I don't think any of us here think like AI tools are net negative, but like the all of us have, all three of us have gotten to play and use AI and machine learning and deep learning for like I mean, I think at least a decade plus each, it's really cool. But like, the word AI feels like a stretch. Yeah. And and that almost makes us seem like the most negative people in the world about these incredibly cool technologies. I think like, my view of, of broader AI is, is AI is going to be as big of a deal so that over the next decade as, like microservices were in the 20 tens. And like, there's a bunch of people for whom that's like, I don't know what microservice architecture is. I don't know what that is. Okay, cool. And those are the people who are probably not making good AI investments right now.

[01:09:57] Audrow Nash: Yeah, I think so.

[01:09:58] Stefan Seltz-Axmacher: Whereas instead, like, yeah, AI is going to make every company 10 to 20% efficient, not 98% more efficient. But like 10 to 20 and 10 to 20 is pretty meaningful. Substantial. Yes. Like massive a huge deal. But no, no, no one has a job anymore. We don't roll video editors anymore. And humanoids will do all physical labor for us in every situation. Yeah, which is a bummer because it'd be cool.

[01:10:24] Ilia Baranov: Yeah. It's, you know.

[01:10:26] Audrow Nash: Go ahead, Ilia.

[01:10:27] Ilia Baranov: I, I want to wrench us in another direction. Just just one funny comment, right? Yeah. Go ahead. One challenge for your your listeners and viewers is, next time you do see a humanoid ad, check if they show how this thing charges and how long its battery life is. Because I think, and I could easily be wrong because I haven't seen every single one, but I think I haven't seen one yet. That shows itself plugging itself in to recharge. And that is a trick that

[01:10:55] Audrow Nash: 1x does. I've seen a video of one Zeus, robots charging themselves.

[01:10:59] Ilia Baranov: Oh, awesome. Awesome, because that is a trick PR2 did a decade way back.

[01:11:03] Audrow Nash: Yeah.

[01:11:04] Ilia Baranov: Right. Right. And so like for the full like, usability cycle, how do you make sure that the thing can actually charge itself a little? Because again, even your Roomba fails to dock, you know, every time. Yeah. Right. So anyway, I'll stop harping on that. But like, if you're, if you're getting the itch to buy a humanoid, check that fact. How long is the battery life and does it charge itself?

[01:11:25] Audrow Nash: I think the battery life thing is a very interesting thing, because they can plug themselves in to some level. One x it was the wheeled base, one that I've seen. I haven't seen their, bipedal one. But it's an interesting thing because the battery life might be super short. I don't know, the.

[01:11:40] Stefan Seltz-Axmacher: Biggest thing that I've yet to ever hear anyone be able to justify to me, even in passing, is my legs. I have. I have yet to hear an argument for a while.

[01:11:51] Ilia Baranov: Like best one I've heard is stairs. Like for home use stairs. Yeah, like that's lots.

[01:11:56] Stefan Seltz-Axmacher: Of stairs in the middle of my airplane factory.

[01:11:59] Ilia Baranov: Or for industrial, I think industrial. Really? Yeah, I get that for people are excited about home. Yeah.

[01:12:06] Stefan Seltz-Axmacher: Well, yeah. Well, which is why computers first came to the home. Like the first place that anyone had a computer was in their living room,

[01:12:17] Ilia Baranov: To store it. Store recipes. Right.

[01:12:18] Audrow Nash: Was out there. Yeah. Yeah, that was like.

[01:12:20] Ilia Baranov: Store recipes.

[01:12:21] Stefan Seltz-Axmacher: You know, besides computer, you know, you don't need a living room if you have a computer to, to print out cards with what your recipe is.

[01:12:31] Audrow Nash: And but so, just to be clear, just so we're not losing anyone, computers first went into businesses, right? For, like, Excel, though. Yeah, and this kind of thing. Yeah. And then the applications came later as the costs drove down. So I suspect that that's probably the good metaphor. Not starting with the home as you're suggesting. Yeah. What are you. So, segueing a bit what are you thinking about? All the LLM stuff and vibe coding and things like this. Let's see. Stefan, you've been playing with it. Yeah. What were your impressions? I like, I've done it a bit, and it's very good, but you can hang yourself on the complexity very quick. Yeah. I'm. You don't structure things well, so.

[01:13:10] Stefan Seltz-Axmacher: I'd say I'm probably three hours into playing around with it. So this is this is a relatively, relatively light, the thing that I'm trying to do right now is, have it vibe coded application that lives behind an email address. Like a back end application.

[01:13:25] Audrow Nash: Emails are hard. Okay.

[01:13:27] Ilia Baranov: Let's. But by the way, what's your definition, Stefan? Of a vibe coding. So we're all on the same page.

[01:13:32] Stefan Seltz-Axmacher: I I've, I've been able to do hello World and a little bit beyond that in Python and C plus plus before. So I'm trying to describe a function that needs to happen. And and it be coded for me. And the function I'm trying to is the function I'm trying to get to is take a zoom summary of a sales call, which is, let's say, 2000 words long. Forward it to an email address, that then, runs it against ChatGPT to turn it into five bullets so that after our call.

[01:14:07] Audrow Nash: That's fairly straightforward. Yeah.

[01:14:10] Stefan Seltz-Axmacher: I've been using, windsurfer kite surfer windsurf. It will not compile. Well, will not turn on, and has struggled to be able to adjust prompts. Is the status of my my code base.

[01:14:29] Audrow Nash: Cursor is actually pretty sophisticated. For things I would suspect you'd be able to be successful with that application and cursor. You use a good model like I find claude 3.7 now. Yeah. To be the best model for these things, but, I would think you'd be able to achieve that. Yeah. For sure.

[01:14:52] Stefan Seltz-Axmacher: And, like, to be clear, I've. My three hours is a series of 30 minute increments with, hey, go, go write a complex email to this person about a complex subject. So I've hardly been flow state developing, let alone flow state vibe developing.

[01:15:07] Audrow Nash: That makes sense. Yeah. Ilia, what do you think? Have you played around with it at all?

[01:15:11] Ilia Baranov: Yeah, I mean for sure. I mean that also, you know, to Stefan's point that you still probably in the top quartile of people who've touched software before and have some concept of compiling. Yeah, you know, that sort of thing, which is like a still a very small piece of the population. And to step outside of a bubble and remember that, like the vast majority of humanity has no idea how the magic in the box works to, to run software.

[01:15:36] Audrow Nash: People that are not robotics people or like software people have not even tried ChatGPT yet that I've talked to, which is that, like, you just talk to general people. They're like, I've heard about it, but I haven't tried it, haven't tried it, which is just unbelievable because it's been a year and a half or whatever. Yeah.

[01:15:52] Ilia Baranov: So you're already talking to like a person who through experience is in a very small slice of the population. Right.

[01:15:58] Audrow Nash: Definitely. And using to especially.

[01:16:01] Ilia Baranov: Yeah. Well yeah, I mean I mean, you know, I, I code semi-regularly usually to the chagrin of my engineering team, whatever I have to do anything and and generally I use it, for two reasons. One is if I'm trying to put together any kind of complex documentation or email or something where I'm trying to get a very large thought across the first thing I do like, I've almost replaced mind mapping with prompting, where I'll be like, I want to write.

[01:16:28] Audrow Nash: Almost.

[01:16:29] Ilia Baranov: Yeah, almost. Yeah, like I want to write. Unless is very esoteric, right? But like, I want to write, you know, object oriented programing for robots, like, that's, that's the title of my talk making One up. Right. What should I consider? And it'll list out a bunch and then I'll pick like 5 or 6. Right. Yeah. Okay. That sounds interesting. I might go with that. Let's talk a bit more about that. Build a a skeleton basically, or a mind map. And then from that, that I'm, that I'm writing in myself, I'm doing whatever.

[01:16:54] Audrow Nash: Right. Then you're back to your mind map. Yeah.

[01:16:56] Ilia Baranov: Right. Right. And like and like that mind map phase. I could be sitting, drinking coffee and just trying to think of ideas from thin air for like 30 minutes. Instead, it takes me three. Right? And like, skips that phase and then it's sort of narrows my focus. And it will usually miss stuff that I do when I talk about. So I'm kind of hybridizing it with my own thoughts. But it takes me that 30 minutes. That's, that's like half the usage of ChatGPT for me. The other half, or that kind of like the other half. I was recently doing kind of like, camera distortion and projection to get a top down view from a camera. And that was like, oh, what libraries exist and how do they behave and what's the easiest? Oh yeah. That's great. Gave me some ideas. Right. And it's like, oh, here's an OpenCV function. It takes these as arguments. I'll go back, I'll try it out, I'll code it, I'll hit an error, I'll be like, okay, what the heck is this error? And that again, it saves this iteration time a little bit. So it takes the initial development that usually takes, you know, 45 minutes to 2 hours down to ten minutes to half an hour.

[01:18:00] Audrow Nash: Totally.

[01:18:01] Ilia Baranov: That's that's pretty valuable. I find it becomes less and less valuable taking from that to a final code of a complex piece of software, I think it breaks more and more and more and more to the point where it's useless. But the really early stuff, it speeds up quite a bit.

[01:18:14] Audrow Nash: Yep. Yeah. I think like de-risking things technically very quickly because you can get a prototype real fast is fantastic. You could see which libraries work. You can run into things that weren't documented. Well, this super valuable, in my experience. But yeah, it's an interesting world we're heading into. Yeah. With all of this.

[01:18:35] Stefan Seltz-Axmacher: I mean, I think, like, we, we've we fooled around with some stuff. So our architecture, basically has a common autonomy core nestled in between two hardware abstraction layers that separates us and are given sensors for a vehicle, us and the, the actual drive by wire of a vehicle. And so the most the most effort, the most, dev work for bringing up a new robot for us. Where do you type of robot is building that hardware abstraction layer? Especially with me. That's the drive by wire. So we've looked at, hey, how can we use, AI to get there? And we've seen some like early cool stuff. But my, my hunch is, I think, I think in general, much of the work that we have to do is good software, best practices more than it is like build AI to recognize a novel situation and categorize it into some, archetypical situation that we have a known behavior for because, like the types of vehicles that we're focusing on shouldn't be making judgment calls like that. Even another human driven.

[01:19:43] Audrow Nash: Is very procedural.

[01:19:44] Stefan Seltz-Axmacher: Yeah. Like like I've seen procedure docs for sites where, you know, they're driving a large wheel loader that is, say, you know, 50 to 100,000 pounds. And the rules for operating that wheel loader are, you know, if you're in reverse and you see someone standing within 100m behind you, turn the engine call, get on the radio, call for help. If you're driving from here to there and you notice that this gate is open, turn off the engine. Get on the radio. Call for help. If if this beeper goes off, turn off the engine. Get on the radio. Call for help. These large, kind of terrifying machines should not be judging pedestrian intents. It's kind of deterministic. If pedestrian, then stop moving.

[01:20:32] Audrow Nash: So that makes sense. Yeah, yeah. What a funny thing that it I mean yeah. Makes sense. They follow the simple and it probably increases how fast they can train people rather than handling all the appropriate conditions. And they need one well-trained person instead of every truck operator.

[01:20:49] Stefan Seltz-Axmacher: And I mean, I think those kind of things like that, a lot of these sites, things can go wrong with like very once very simple rules are broken. So it's a rehashing constantly of simple, straightforward rules. And in part they're simple because it's an easier for other people to notice if someone's not following the rules.

[01:21:10] Audrow Nash: Oh, interesting. Well let's see. So one thing that I wanted to make sure we talk about today, so I'm having, a baby girl in bed about six weeks from when we're recording. You guys both have kids? Yep. I don't often hear this discussed with founders and this kind of thing. I'd love to hear your perspectives on how it is to have a family. But also, you guys are doing your startup. Startup life is notorious for being chaotic and busy all the time and everything. I'd love to hear what you think about it and how your experience has been. Yeah. You want to start, Stefan?

[01:21:51] Stefan Seltz-Axmacher: Yeah. I've done a startup with no kids I was at a startup with. I get. So if it's, It's good. It's got a good middle ground.

[01:21:58] Audrow Nash: Well informed.

[01:21:59] Stefan Seltz-Axmacher: Yeah. It is significantly harder. It is. That is not a joke. I mean, I think, there's an old, Sam Altman quote from back when he was just, a measly VC partner and, you know, the CEO of an AI. That was, you know, when you're a startup founder, you get to have, like, one hobby. Like that hobby might be that you have friends, it might be that you like kite surfing. It might be. It might be that you go backpacking, but you get like one. And when I didn't have kids, my hobby was I had friends. And now my hobby largely is I have a kid. Yeah. And that is, that is where my my free, not work nonproductive time goes. Which is. Yeah, that is a trade off. That is a thing. But, you know, that's, that's kind of that's kind of part of the deal helps that, you know, in my 30s and I'm a I did a bunch of fun social things in my 20s. I'm okay with like the. Yeah. Now the change. Yeah. Yeah. The there's, a lot of parallels, I think, my, so I, I don't know if you guys have signed up for this. You and your you and your partner. There's a great, Montessori toy subscription called, love every.

[01:23:14] Audrow Nash: Okay.

[01:23:15] Stefan Seltz-Axmacher: Which I highly, highly recommend. My kids, my kids, you've.

[01:23:20] Ilia Baranov: Got, like, you've got, like, five months until this starts becoming relevant. To be clear, the first three months, they're not going to be.

[01:23:26] Stefan Seltz-Axmacher: Oh, no, no, but Fritz Fritz loves his, like, first three month toys. He like, he like. Wow. Yeah. Like, I mean, there were a lot of them were like, black and white cards with patterns on them, that you like to look at. Like, we'd probably plop them in front of them anyway. Yeah. Like, what is that? Yeah.

[01:23:43] Audrow Nash: Oh, contrast. Very interesting. Yeah.

[01:23:45] Stefan Seltz-Axmacher: But, with that toy cat in the first box, there was a coffee cup that had the phrase, the the days are long, but the years are short. And that is exactly, a parallel experience having a startup, because actually, it's the thing that Ilia was joking about earlier on the podcast of a, a beer getting humanoid. Actually just being a vending machine with, autonomous cart. That was not a year ago. Earlier that was like two and a half years ago. We had that. Oh, there you go. The, startup time is like a black hole. Where I started my first startup when I was, like, 26 or so. And now I'm 35, ish, I think. I don't know, and I'm not really sure. And just like, in my mental model of myself, I'm still, like, maybe 26.5. And.

[01:24:38] Audrow Nash: Only six months is advanced in some sense. Yeah.

[01:24:41] Stefan Seltz-Axmacher: And.

[01:24:41] Audrow Nash: I think years has gone by otherwise.

[01:24:43] Stefan Seltz-Axmacher: Like the fact that having a kid is similar is is useful in that, like, I can, I can grok the joint mental model of it all where there's so much happening all the time, but also not a lot. It's,

[01:24:55] Audrow Nash: Yeah, it's like a paradox. Yeah. For everything.

[01:24:57] Stefan Seltz-Axmacher: Like, we've like polymath has achieved much less than I would have wanted to achieve in the years that we have. But like, it's impossible to achieve what I want to achieve in any, anytime. But we've also done incredible, amazing amounts of things. We've been, we've, we've gotten to work on some really cool programs. We've been on lots of different robots. We've seen really neat things. And I think the same thing is true with like kids where like, if Fritz, my son is now, we just moved from saying, months to he's a year and a half.

[01:25:29] Audrow Nash: And wow, he's a big transition.

[01:25:32] Stefan Seltz-Axmacher: Yeah, yeah, he can do lots of things, but, like, he still can't talk in sentences, so I don't know. It must be dumb. But yeah, it's it's cool. I know it. At least being a dad for way longer than I have.

[01:25:46] Ilia Baranov: Yeah, I have, I have a four year old daughter and, one in the three quarter year old son. More than three.

[01:25:53] Audrow Nash: Quarter. That's great.

[01:25:54] Stefan Seltz-Axmacher: Yeah.

[01:25:56] Ilia Baranov: I keep saying one and a half, and then I have to keep mentally pumping a lot of things. Yeah, far beyond that. Yeah. But the thing, the thing I want to highlight a lot because my experience is a little bit different than stuff. And my daughter was already born when, you know, Stephan and I met and started discussing, co-founding and doing this. And the thing I really want to highlight is the enormous role that your significant other spouse or whoever plays in this. Right? Like, I, I at the time was working at Amazon, relatively well-paid, you know, decent relative stability, decent program, was still working on robots. I was still doing something I like. And I brought up this crazy idea of like, let's drop all of that.

[01:26:36] Stefan Seltz-Axmacher: Move.

[01:26:37] Ilia Baranov: To a startup. But, hopefully it works. And, what do you think? And, and, you know, all credit to my wife. She, incredibly, was on board with this, she really wanted to understand the scope of the idea and need my co-founder, you know, try to understand if we're building something reasonable. And what is it? You know, is it sane? And, and. Yeah. So I think that's a huge chunk, right? Like that, that makes all of this possible. And her infinite patience to have me suddenly have a trip planned and needing to have her take care of two children and by herself. And then a day before that, the trip is no longer planned. But now there's a new trip, but some other point. And now I have to go to this event and have to go to that event. And that that introduces a lot of chaos. And so that that is, you know, the only thing that keeps the sane, I would say the only way that this would be harder would be to be a single parent and also run a startup. And I can't even imagine that.

[01:27:37] Stefan Seltz-Axmacher: I, I feel totally I don't know how anyone is a single parent. That seems to me like not even that's inside of a startup. I mean, like like and just like, take a shower, like, yeah, yeah, yeah, yeah.

[01:27:49] Audrow Nash: Yeah. Well, we super, super.

[01:27:51] Ilia Baranov: Human levels of ability.

[01:27:52] Audrow Nash: We have a puppy now and the puppies too. You might have heard Embarek a little earlier, but he, is so much work and I could not even imagine. I imagine, like, single parenting a puppy. Like there's no way in hell I could do it. And I can only imagine a kid will be more for. Yeah, one or more kids. Oh my goodness.

[01:28:12] Stefan Seltz-Axmacher: But it's also really cool. It's it's great. It's really cool. It's really amazing. It's, you know, ten out of ten would do it again. Highly recommend.

[01:28:22] Audrow Nash: Two thumbs up. Yeah.

[01:28:23] Stefan Seltz-Axmacher: I don't know, five stars. Like, there's annoying stuff. Like he wakes up screaming bloody murder in the middle of the night, and, there's disagreements on parenting strategies in the middle of the night because kids screaming in the middle of the night. Yeah, and everyone's at their worst. And then there's, like, you know, when, when he went out for the day with, with his, nanny, you know, he kept on hugging me and my wife on his way out the door, and it's just like, oh, my God. Or like last night, I'm given a bath. And, like, I figure this, like, kind of rubbery bottle. If I held it underwater, I could squeeze it and, like, a burst of water. And he, like, laughed incredibly loud. It's like it's it's the coolest thing ever. It's pretty amazing. And yeah, I mean, I think it's just a work and and kids, it's a all the time always like if, if, if instead we had done, you know how to not for cats. Probably we could just be hanging out right now either either rich or unemployed. Yeah. But it's just it's a high level of intensity. And I mean, I think, you know, honestly, before, before alien, I teamed up when I was thinking about, like, service and what I wanted to do, I desperately did not want to start another robotics company. I, like, desperately like I wanted to do some easy mode SaaS company. And the thing that I realized really quickly is that none of those things were exciting enough for me to want to work on on a Saturday morning, instead of literally anything I could do on Saturday. On a Saturday morning, even even during like peak Covid, it was like, I'd rather watch like Netflix reruns than work on like a trucking software company. And like, robotics has this incredible intensity that makes it incredibly compelling and rewarding. And I think life with a kid is a similar sort of uplift where, yes, it is really hard that they scream in the night. Yes, it is really hectic. Yes, it is like a puppy on steroids, however, but it's this like extra cool thing in day to day life, so I don't know, I'll just be tired until he's probably like 8 or 10 or something. And yeah, and until polymath is in the billions of dollars a year revenue and I can mostly be the the founder who shows up and waves his hand, hit his hands, in a black turtleneck every six months. But like, yeah, that's I feel like it's better living than than the other way around.

[01:30:49] Audrow Nash: I suspect that's true. Yeah. What do you think?

[01:30:52] Ilia Baranov: Well, I was just going to give a hint to the, to yourself and to whoever might be listening on, on on the waking up screaming and disagreements on parenting styles. The the golden rule I read was that you you have to make up your mind on how you're going to handle that crying during the day. And the rule is, you cannot change your mind at night like there's no discussion. You cannot change it. Whatever you do in it. That is the plan. Period. Like that doesn't even enter your thoughts, and you have to be ready to reevaluate that during the day when you have a little bit more sleep every few weeks, because it will change pretty rapidly. But like the plan is the plan, you don't discuss it like that. That saved a lot of chaos.

[01:31:38] Audrow Nash: That is great.

[01:31:39] Stefan Seltz-Axmacher: So have you and, you and your, partner, my wife, Michelle. Your wife? Yeah. Have have you and Michelle had your parents try to chime in, not raising your.

[01:31:48] Audrow Nash: But they haven't. So, I mean, we're already getting parents talking, like, giving us advice all the time for things for the baby now, puppy. They didn't really care. Our dogs pretty well trained. Okay, so, like, I don't know, we generally impress with them.

[01:32:02] Stefan Seltz-Axmacher: We I, we had a really useful experience, when we just had a puppy, pre pre having a kid. Yeah. Where, we went to my mother in law's house, and my dog is great and lovely and charming, but, you know, neurotic and crazy and scared of some of the other. And, Cassie, my wife and I would be arguing about some course of action for the dog in front of my mother in law, who then would, like, take my side. And then then, like, Cassie felt teamed up on. And then, like, I tried to switch to her side, and there's this musical chairs. Who's our brother in law? And it was just incredibly stressful. And no one likes it. And yeah, another another piece of advice I'd give you, similar to Eliza's on, Night Cry is, in front of anyone else. You guys agree? Don't argue or just you're on the set. If you're arguing about what to do about Nate crying in front of someone else and they chime in supporting one of you, you know, support whoever is being, who.

[01:33:04] Audrow Nash: Yeah. Or ganged up on. Yeah.

[01:33:06] Ilia Baranov: Better just not to discuss it.

[01:33:08] Audrow Nash: Yeah I think yeah. Don't discuss it in front of anyone else or this kind of thing. Not not a, like an impromptu passion argument.

[01:33:15] Ilia Baranov: Yeah. Impassioned argument. There's many wrong answers, but there's no right answer.

[01:33:19] Stefan Seltz-Axmacher: Realistically.

[01:33:21] Audrow Nash: So that's hilarious. Yeah. One thing that you may laugh at with this is we moved to San Antonio, Texas. And, when we ask people or when people ask, about like how we chose it, we say it was the least bad.

[01:33:36] Stefan Seltz-Axmacher: Move, all that.

[01:33:39] Audrow Nash: In a similar vein.

[01:33:40] Stefan Seltz-Axmacher: Yeah. Yes. Yeah.

[01:33:43] Audrow Nash: Any any other advice of kids, that you can think of or any other other things to share? Maybe we'll go around one more. Stefan. Anything for you?

[01:33:55] Stefan Seltz-Axmacher: I think on the kids side, I said all my stuff earlier. Probably got more. Yeah. I mean, you'd think so. If you're working on robots, we'd love to. We'd love to talk to you in general. Where. So this, this, this approach that we're talking about, about mostly focusing on these big, chunky, industrial things that approval that are, you know, obviously vulnerable, have put us in an interesting position where we're starting to have cash flow, positive bumps who, and that that's so exciting. Yeah. Like, there's a big fundamental difference from the groups that I know, who are saying, like, hey, we need raise $100 million to get a bunch of synthetic data to then maybe get $500 million and then maybe get 3 million. And I think, I think we're in a time where there's a lot of noise going on elsewhere about those sorts of approaches, and 99% of those companies won't exist in three years. But like the folks who can, can actually build a robot that actually creates real value are, you know, half of us will be around, you know, like, you still screw up a pretty generous product. But like.

[01:35:07] Ilia Baranov: Me and us will be around.

[01:35:09] Stefan Seltz-Axmacher: That most of the companies will die, not 100%, but most.

[01:35:13] Audrow Nash: Maybe not in three years. Yeah, or they'll be acquired or something.

[01:35:16] Stefan Seltz-Axmacher: Yeah, but like, many of them are too big to be acquired. You know, there's not a lot of natural acquirers for $10 billion products. Humanoid robotics company fair point. Like there was when that company was worth $200 million and it was 2016 and interest rates were low and like, cool, like Google or whoever, cash was expensive now, for sure. Like they did. Fine. But now at like the current valuations, that's just not feasible. So I think this there's, there's this incredible like, slope of enlightenment that we're currently on for robotics around building things that are actually buildable and actually sellable for, but, but can be built today and I encourage more people to look at getting on that slope instead of the the next hype cycle.

[01:36:04] Audrow Nash: Oh yeah. Let's see. Yeah. I guess, wrapping up, because I see that we are coming close to our end time, so, Stefan, that sounds like what you want people to take away from this, which makes good sense. Yeah. What would you hope sticks out in people's heads? From this whole interview?

[01:36:23] Ilia Baranov: Yeah. You know, definitely, definitely. Stefan making the points on the on the business and robotics side, I'd say I gave kind of a talk, last year at Robot Business on running a startup and and running a robotics started, and I think it's on our YouTube channel if people are interested. And, that what I ended up, that one was kind of two critical things. One or let's add one more, but two to start. One is even if you're building robots, you have to be humans. And I think that, you know, like having a family, both myself and Stephan and many of our team members are younger and starting families. There's no blood from a stone that works well with complex technical things like kind of crunch time that comes from gaming industry or heavy software dev does not work well and is, in my mind pretty clearly the less capital efficient way to do things over time, like even for shareholder value, like it's almost always a negative and so be humans. While building robots is number one. And then the other thing is the lesson we keep having to learn over and over, and as core to robots is you have to do less work, love, do less, but do a better love. Focused approach like do that as much as you can. And I think both of those things, are applicable almost equally to child rearing as they are building robots. Very, very common skill sets of like, don't try to bring your kid to 30 different events like, yeah, do three, but do them really nicely.

[01:37:57] Stefan Seltz-Axmacher: What is child rearing if not building a really slow dev time robot.

[01:38:02] Ilia Baranov: Yeah, yeah. Oh yeah, I think of that all the time, especially with a with a newborn. You're like, oh like I keep, I keep giving this example. When my son was born, I noticed he would only smile if my face was oriented the same way. His face was like, he can only recognize faces in this orientation. Wow. And then, yeah, and that's apparently pretty common. And then like, a some amount of time, I forget exactly when suddenly it snapped that like, even if my face was rotated this way, it was still a human face and it was like an overnight change of like, rotation works now for a vision like that job.

[01:38:34] Stefan Seltz-Axmacher: Yeah, it works.

[01:38:35] Ilia Baranov: Right. And like, you could see these things turning on and.

[01:38:38] Stefan Seltz-Axmacher: I don't think even had it enabled.

[01:38:41] Audrow Nash: Yeah. Yeah.

[01:38:42] Ilia Baranov: No, no, nothing real impressive, you know.

[01:38:44] Stefan Seltz-Axmacher: Yeah.

[01:38:45] Ilia Baranov: Reinforcement learning I don't know. But yeah. So, be humans and, do less but do a better.

[01:38:54] Stefan Seltz-Axmacher: Oh, yeah. Well, thank you so much for having us on today. So lot of fun.

[01:38:57] Audrow Nash: Oh, yeah. Well, thank you both. Do you want to plug your podcast real quick?

[01:39:01] Stefan Seltz-Axmacher: So you can learn more about us at Polymath robotics.com. And our podcast is automated.com or automated parts.

[01:39:08] Audrow Nash: I'll include the link.

[01:39:10] Ilia Baranov: It's at the bottom of the webpage.

[01:39:12] Stefan Seltz-Axmacher: It's called Automate It. It's wherever you get your podcasts.

[01:39:15] Audrow Nash: Okay. Hell, yeah. All right. Thank you both.

[01:39:17] Stefan Seltz-Axmacher: Thank you.

[01:39:18] Ilia Baranov: See you next time.

[01:39:19] Audrow Nash: Bye, everyone.