Loading summary
Transcript160 lines
- [00:00]
A
Do you think it's like cooking a steak at some point?
- [00:02]
B
Yeah, why not? If you think about at the end of the day, you've, you have pick in place and simple manipulation. That's what cooking is. They're just like a much higher degree of reliability. And there's other things around food safety and bacteria and other things that come in. Cookie cooking and temperature sensing and what like that. So it's all doable. It's just like, I would think at.
- [00:17]
A
Some point it's like, hey, robot, I'm at work right now. There's a steak in the fridge. Please cook it and clean up everything by the time I like, 15 years from now, that's, that's doable.
- [00:27]
B
Less than five?
- [00:28]
A
Less than five.
- [00:28]
B
Yeah. All right.
- [00:30]
A
I'm really pumped to be here with Kyle Vogt. Kyle, thanks a ton for making time for this.
- [00:33]
B
Thanks for having me.
- [00:34]
A
So I want to start with talking about, like, why robotics seems to be having such a moment. You know, it's obviously been really important for a long time, but in the last few years, it seems like a lot of really good entrepreneurs, a lot of good investors have started to pour a bunch of time, money, resources and effort into this. And I guess I'm curious just to start with sort of laying a foundation of like, can you put this in some context and like, what used to be the case and what has changed that's like making people so energized right now.
- [00:58]
B
Yeah, it is. It's like the most excited I've ever seen people in robotics. And, you know, I guess as an engineer, there's something like romantic about building machines to do the stuff that we don't want to do. And that's, that's why I've been doing this for so long. First with, you know, a decade on self driving cars, but for me, even going back to like teenage years doing BattleBots and then going to MIT to basically build more robots. But, you know, during that entire spectrum, it's also, it's always been this niche thing, and frankly, like, robots have never really lived up to their promise. There's always something. They're always overly fragile. Like in a factory environment, we put them in cages, and if things don't line up like within a millimeter, the whole thing doesn't work.
- [01:33]
A
Except the BattleBots. Those were good. Actually, now I'm thinking back, BattleBots were. Yeah, you made them with like the saws and everything.
- [01:39]
B
Ours had like a hydraulic ax, which was, which was pretty cool. But calling these robots is a bit of a Stretch. They're basically glorified RC cars. Yeah, right.
- [01:46]
A
That's right. With a weapon.
- [01:47]
B
Yeah. So what's different now is, you know, for the first time, you have robots that are powered by, essentially, they have all the brains of an LLM built into this robot, and we're controlling them with neural networks instead of classically engineered algorithms. And so the difference was before, if you have a robot that's like, in a room like this, even saying, like, go to the whiteboard is almost like an impossibly hard computer science problem. It's like, okay, I have to build an exact 3D map of the world. Like, have a detector that can figure out what a whiteboard is, train it on millions of examples of what whiteboards look like, just to be able to do this. And even then, the failure rate would be high if you put it in a different room and it doesn't have a map. But now it's almost like cheating. You can take all the common sense that's on the Internet and inject it into a robot brain. And so if you're like, where's the whiteboard? It knows instantly you, like, open the chatgpt.
- [02:36]
A
If you open, like, video, you can, like, show it anything. And it, like, knows what it is.
- [02:40]
B
Yeah. And so imagine, like, robots before started with zero knowledge of the world, and now, like, suddenly have this kind of.
- [02:45]
A
Knowledge of the world better than us. Like, they can look around the room and see stuff better than we can.
- [02:48]
B
Yeah. And then on the motion side, you to have to have a PhD to compute these complex trajectories. You have, like 12 joints on a motor or on a robot. How do you get all 12 joints to move in tight coordination to move an arm to a place? And this is a very difficult and computationally intensive problem. And now we just kind of jump over that whole thing. And now if you have a way to teleoperate a robot or to put it in a simulation, you can just learn how to move all those joints to mimic the human operator or to accomplish some reward function or to maximize it. But, um. And so you can skip that whole computational challenge. And so those two things together basically mean that everything we thought we knew about robotics or, like, what kind of businesses were good businesses or bad businesses, all, like, that slate has wiped, been wiped clean.
- [03:30]
A
Yeah.
- [03:31]
B
And so I think you're going to see this Cambrian explosion of different robots for different applications that now suddenly just work. Whereas before, they would, like, really struggle to do the most basic things.
- [03:39]
A
And when you say for different applications, is that are you saying it won't be terribly generalized? Will it be medium generalized? Like, what made you say, for different applications or like different environments maybe?
- [03:50]
B
Yeah, I mean, classically, like a lot of robot businesses, like, try to get really, really narrow the successful ones. We're going to focus on this one problem, like Factory automation for 3 PLs, for putting things in boxes and putting them on a conveyor belt, like, very specific. And that's just so you could narrow the problem enough to be good at it. Now I think you're going to see people broaden the horizons a little bit because it's much, much easier to go from, you know, a piece of dumb hardware to something that's performing a useful task. You know, I say multiple applications too, because it's my view that there's going to be a whole bunch of different shapes and sizes of robots, each optimized for different types of work, as opposed to maybe a humanoid robot, which is very, very expensive, but in theory could do everything. I think we're probably going to see some of those. But the vast majority of robots will be more special purpose in nature.
- [04:32]
A
I feel like there was a moment in AI where the researchers who were sort of closest to the work were very sure it was going to work before the rest of the world knew. Is there an equivalent thing in robotic. Have people crossed a similar threshold to whatever that was at the pre chatgpt moment in robotics where people who are at the very front have been working on it for decades are like, this is definitely happening now?
- [04:57]
B
Yeah. If you had like secret microphones in like robotics labs across the country, right now you'd just be hearing, holy shit, holy shit, holy shit. It's like constantly happening. And I think finally, like the light bulb moments are happening and it all like in the early days of this stuff, it all looks very rudimentary and kind of simple, but if you know what you're looking at, you see the signs of life that mean over the next three to five, 10, you know, even less years of development, this will go from an interesting technology in a research organization to broad mainstream appeal. And yeah, those signs of life are happening. Those light bulb moments are happening all over the place right now.
- [05:32]
A
So what are the components? Obviously we talked about there's vision, there's the ability for the robot to do manipulation the right way for it to have the right sort of dexterity. There's gotta be something around reliability. I don't know about decision making, if that's its own sort of. What are the components basically to this uplevel.
- [05:52]
B
Yeah, you've touched on a bunch of good ones. Depends on the type of robot for the ones we're building. Like robots that operate in your home, they need to navigate through a home, they need to remember where things are in the home, they need to interact with and manipulate these objects, like you said, and probably have some way of incorporating your user preferences into all of this. And so you've got a reasoning component like I see a certain thing in a home, plus I know the preferences. You've told me in the past about how you like things organized or how you run things in your home. And then I'm going to take that and reason about what the next steps I should take as a robot. And then once you have those next steps, you're going to take drive to the oven, put the towel on it, then go over here. Once you have those discrete steps, then you can move to more one of these end to end models that basically given a simple task can go execute it.
- [06:36]
A
My implicit assumption here is that on some timescale you're extremely confident this will all work. But what are you unsure about in the Next, let's say five to 10 years or what will drag?
- [06:46]
B
One of the biggest challenges for something like this, it's a brand new product, is how do I use it? Like how does my life change and how do I adapt the way I live to best, you know, make use of a robot like this? And that could be, you know, in a home environment or maybe it's like a manufacturing business that, you know, its entire workflow is organized around people standing in work cells doing a task and handing things on a conveyor belt. Like how does all this change? And so I think that the technology part will come pretty fast and I'm pretty confident in that. The part that I think traditionally takes longer is the world has to adapt to basically now that this new thing exists, how does everything about how I run my business or how I live in my home or how do I operate my hotel or whatever it is need to change or should change to best make use of this new thing.
- [07:26]
A
It's like this is like where like AI software is, where like it's obviously much better than like what's currently being deployed and used. It like takes time to get from like the tech is good to now it's like implemented everywhere. So you're basically saying it's like the robots will be good enough at some point soon, but then figuring out how to use them in daily life and like where does it Actually fit into, like a life workflow, that kind of thing.
- [07:48]
B
Yeah. And I think the companies building these technologies have a responsibility to help us figure that out. You know, they're closest to technology, and I think they need to think not just about, like, what, what does, what does technology building do? Or, you know, what's the fancy new thing I built in there? But, like, you know, three steps removed from that. How, how do businesses actually make use of this? And like, you know, what do they need to know about it? And like, what things do you need to build in so that it's as easy as possible to sort of go on this adoption curve and make it happen?
- [08:11]
A
Are you more in a mindset of like, we are Apple and we're going to like, bulls tell you the product kind of, and like, this is how it's going to work and this is what the robot will be, or is it more of like the yc? Like, let's just get it into some homes and iterate. Like, which mindset do you think you feel closer to?
- [08:27]
B
Well, frustrating answer, but a little bit of both. It's one of those things like strong opinions weekly held. So I think you have to have an opinion. You have to have your taste and your preferences built into the design of a product. Or it feels bland, like a product with no opinions. It's just like, you know, you wouldn't even notice it. So I think you have to start off with strong opinions and then be willing to put those in people's hands and then quickly abandon them if it's not, you know, if it doesn't work the way you want to. I think if you're not stubborn enough, you end up with a just product no one is interested in. And if you're too stubborn, then you end up with a flop in the market once it's out there. And so, you know, I think it's a careful balance.
- [09:00]
A
Why did you feel compelled to go for the home? Like, you, obviously, even with this generalized robot sort of idea, there's a lot of things that you could do that aren't just like, pack a box in a warehouse type of thing. That's sort of like more dynamic than that. But, like, you picked home for some reason.
- [09:14]
B
Yeah, for some reason. So I just turned 40. This is my third, you know, company that I'm working on. Big one. Yeah. Yeah. I bring this up because, like, at this point in my career, I kind of know how I want to spend my time and, like, what's important to me. And first of all, I want to have a lot of fun and working on home robots that I could use, all my friends can use. Like, couldn't think of anything more interesting than that or more fun, especially compared to, like, robots that are hidden in a factory that no one would ever see. Totally. I also think that, you know, one of the great promises of working on really cool technology is, you know, you certainly get some dopamine hits when you're solving a problem and you make it work. But like 10 times that or a hundred times more is when you see your hard work go in someone hand, someone's hands, and they use it for the first time and they come back to you and say, oh, this is so cool. Or my life changed because of this. I remember, you know, one of my favorite stories from or examples of this was when we were working on Twitch and there was this guy who was like a carpet cleaner in Minnesota or something who started streaming on the side and, like, became had a really popular channel. And he was making, you know, he was like the first one of the first streamers to make six figures just playing video games online. And he's like, this completely changed my life. And so, like, moments like that when you build some cool technology, but then it actually, like, moves the needle for someone and they tell you their stories. That's. That's the really motivating thing for me. And you're just not going to get that, you know, if you don't have millions of people using the product.
- [10:28]
A
I mean, the idea that you could get like a robot in everyone's home is. It's really. It's actually totally believable to me. Like, I could see a future, which I guess this is what you're building towards. If it's the right form factor and price point and it does the right set of things, it seems very beloved, believable. And I guess you probably had some range of considerations where you're like, we could make the smallest possible, cheapest possible thing all the way up to we could try to make a $50,000 humanoid. And you picked something at some point along that spectrum trying to be somewhere there. Did you think about it in sort of like a range of what was technologically possible? What future you thought kind of made the most sense? How'd you pick what sort of complexity and price point to live along?
- [11:11]
B
Because you're not doing humanoid from day one. My concern is that there's always going to be an expectation for what the home robot product can deliver and what reality is, especially in the early days. And that Expectation versus reality kind of goes into value, how much value you perceive you get from this product. There's a scale, there's like cost on one hand, value on the other. And we want to do everything possible in our favor to tip the scale towards value. And so that means being really aggressive on cost to get the price down and make these affordable. That has the dual benefit of, on one hand, you know, making it so that people are delighted by the product because it's not something they spent as much as a new car on. They spent something much, much less and they're pleasantly surprised, hopefully. And the other is if you get the cost low enough, you can sell these to a lot of people because lots of people can afford them. And at this day and age, data, real world data, is one of the biggest bottlenecks in robotics. And so if you can get lots of robots out there, you're going to have lots of data much sooner, which then creates this feedback loop where the product gets better and then it's worth more to people and that more people buy it. There are a lot of trades where you can build a cooler robot or add more capabilities or you can reduce the cost. And we've almost always been in the reduce the cost kind of thing. Yeah.
- [12:17]
A
Do you think that the, like, the humanoid vision, which is obviously extremely sci fi and cool, like, does it make sense? Obviously you could build a robot a bunch of ways and like one way you could choose to do it is just like shape it like a person, but it's a robot, you know, maybe there's some reason for it. But like, when you think about like the humanoid question, like, does it intuitively make sense as something that like, ought to exist or is it kind of random?
- [12:41]
B
First of all, when, when I see the videos of, of human, humanoid robots these days, having worked in the field for a long time, it is just so cool. It's so amazing to see what people are able to come up with these days and how fluid the movement looks and, you know, how dexterous they're getting in terms of the things that they can do. And so I think they're amazing machines and I think they need to exist in the world. I think the question for me is if we're talking about putting these robots to work or people owning them, the question is, at the end of the day, is this the most cost effective way to deliver the most value I can to that customer or to that person? And I think for humanoids there are very few uses for which the answer is yes. Most of the time. The answer is no. I can build a simpler machine that works in this environment. If it's a factory thing where the floors are all flat and you're just moving things from one place to another, that robot should probably have wheels. If you're in a home environment and a humanoid presents all these safety issues with walking upstairs, if it slips on a banana peel and falls, it becomes a ballistic missile, basically going down your stairs. These are not good things for the home.
- [13:40]
A
That's true. Actually, a big heavy robot falling down your stairs is a huge problem.
- [13:43]
B
Yeah. So for the home, you probably want to optimize more on low mass, low cost, and try to maximize what you can do, but not running into some of the challenges of a humanoid. That said, there are some things that'd be really hard for a non humanoid robot to accomplish. Like if you're on a construction site and you're climbing up and down ladders and using hand tools, design for humans and all these things. I buy that argument that there are some uses where we'll want humanoids, but I think currently, I think people advertising humanoids are trying to get hype in the space, get more investment in the space, which we need. But I think the actual practical uses of them will be a little bit smaller than what is being portrayed currently.
- [14:16]
A
It also could make sense that they don't make the most sense in a home, but they live other places. Like, it would be good, for example, if a lot of defense was carried out by machines, because that could, in some world, I hope, hopefully, that could save lives, for example, or you could imagine it sort of guarding at a stadium or taking care of big sort of patrol areas and things like that. So I could see that because it is a very mobile thing in the home. That example that you just gave, it slips and it falls on the stairs and it hurts a kid or an animal or something like that.
- [14:49]
B
I mean, maybe in the distant future we can solve these problems. I think just near term, you're less likely to see them in the home.
- [14:53]
A
First along that curve, though, between now and, of course, 50 years out, obviously these things are going to be. I think it's like cars, where it gets safer than people one day, I assume. But on the way up, what's the regulation going to be like for robotics? Do you need to be really involved with the government to put these robots in a home or is part of what you're doing with the design to avoid a lot of that stuff?
- [15:15]
B
Right now it's very different than some of the industries I've worked in are defense things or automotive things where they're very, very heavily regulated industries and for good reason. I think you're going to see a lot of products in the home, and it depends on your view. On one hand, we have these little robot vacuums going around today. You could make an argument that this is kind of just a step up from that. But you don't see for consumer products a whole lot of targeted regulations for individual products. We have general product product liability laws and other things that are generally applicable to everything from chainsaws to blenders or other things that you might have in your home that carry some risk associated with them. But I think there's an immense responsibility on the developers of these products to try to make them safe and to do everything possible following best practices, regardless of whether or not there's regulation. One thing that we may see more of is looking at how the data is used from these products. You know, the security of these products. I think that's really important. Obviously, like, the home is one of the most intimate spaces in your life. You know, there needs to be a great degree of trust and responsibility that goes with companies who have these machines that are likely covered with cameras, you know, running around our homes. And most of us don't even think about today. Like, when we buy a robot vacuum, where does it come from? Like, who is the company behind it? Are they trustworthy? Are they going to do the right thing in my home? And that's where I'd like to see a lot more scrutiny.
- [16:22]
A
So what does that mean you're going to need to do? Because you're right, it's like, you know, I remember, you know, people got comfortable with it at some point. But like the Alexa problem where there's like a microphone in your home, like now there's like a microphone and a camera and like whatever else in your home. So, like, what, what does that mean? You need as like a company to sort of be, you know, a trusted brand there. Like, you have to go from day one pretty hard at that, I guess.
- [16:41]
B
Yeah, yeah. You have to have some principles and opinions and be able to talk about it publicly, I think. But, you know, all these products are going to. Every, every new category of product. Like this goes through weird snafus in the early days. And when you mentioned Alexa, I was thinking to mind when those first came out, wasn't there something where there's like a TV commercial that came on and said, hey, Alexa, something, something, and then like across the United States, like thousands of People bought toilet paper, you know, and then recently with the Meta meta glasses, Zuckerberg was on stage and he said something and all the people in the audience, their device ping the server at the same time and the demo failed. You know, so there's going to be these weird moments and things that come along in the early days that. But anyways, on the, on the data side, for us we have two things we care about. One is transparency. So if there's data being collected in your home, like what was it? I want to be able to know what that data was and what's going from the robot to anywhere else. And the second is control. If this product is in your home, you own it. You need to have the on off switch and be able to control what that data is used for. And I think if you have those two things and you are principled about those things and hold true to them and basically fulfill your promises and you give the control to the user, I think that's the best starting position for something like this is just to establish those principles up front.
- [17:51]
A
One last question on robotics and we can go to another topic. AI models behind robotics. How distinct is the concept of robotics? AI versus other AI.
- [18:02]
B
So, I mean, there's a lot of similarities. And I think In a way, LLMs that started off as chatbots that exist purely in the text world, and robots, which are like physical machines, very multimodal in nature, you can see these things kind of converging because the latest models are multimodal. They can take in audio, images, other things in the same way that your robot is expecting that. And so over time, I think they're converging a little bit. And in fact, a lot of the training approaches, pre training, post training, those concepts exist in the robotics world. However, there's still a lot of things that are unique to robotics that you would never do if you're working purely on an LLM. And that's a lot basically like mixing in real world data, different ways of collecting it, different ways of using simulation and figuring out how to tie that to all the intelligence that's embedded in like a modern LLM.
- [18:47]
A
And then the data is like super important here.
- [18:48]
B
Obviously, data is important today. I think this is like a now problem. You know, if you look at LLMs, I think the reason that you can see so many different companies, like, like 20 different companies, all building foundational models and get, you know, within a stone's throw of each other in terms of performance. Small teams, large teams, whatever it is, is because essentially they're all starting from the same data set, which is the Internet, and everything that can be downloaded from it. And that data kind of determines the quality of the model that you can get. And there's certainly some alpha on top of that from individual teams. But in the robotics world, there's no corpus of data like the Internet that exists. There isn't an entire Internet of point clouds or camera images of robots manipulating objects. And so right now, we're in this early days where you've got to either bootstrap that data yourself, you've got to pay people to collect it for you, or you've got to try to interpret or generate robot data from other things, like watching YouTube videos and trying to infer from hand motions how a robot should do the same thing. And so we're just kind of in the early, early days of that for robotics.
- [19:45]
A
Do you think there should be like a scale AI for robotics data, or will it be that a company like yours just generates its own data and gets smarter as a result of that?
- [19:54]
B
I don't know. I think there'll probably be both. In fact, I've probably talked to at least a dozen companies who want to be the scale AI for robotics, and I think that there's going to be plenty of customers for that in the near term, especially as you know, as this data void exists. But when that starts to be filled and we start to see useful robots in the world, I do think the majority of data collection will come from robots and less from people getting data.
- [20:17]
A
I would also think that for, like, for your product, for example, any data set that is not your products in the wild is going to be approximating the data. And the perfect data set, I would imagine, would be if you had armies of robots out in homes giving you data.
- [20:30]
B
If your technology is sufficiently advanced that you can do transfer learning from other forms of Data, other robots, YouTube videos, whatever it is, any source of data, and you can use that to train your robot. That's like an advantage because you don't, you know, then that total size of that data set may be much larger than just the data set that would be collected on your specific robots. However, where we are today, it's much easier to get robots to do amazing things if the data collected came from the exact robot that you're trying to deploy a model on.
- [20:59]
A
Totally. Yeah.
- [20:59]
B
And we'll see if that that changes over time.
- [21:01]
A
Yeah, it makes sense. So we alluded to this before, but you obviously started Twitch, you started cruise, you're doing it again. First of all, why are you so motivated to keep doing these hard companies and so many people after this much success, wouldn't go back to the beginning and you've had two really successful companies, which I want to talk about, particularly Cruise, because I think it's related. But I guess to get started, what's driving you now to do this again?
- [21:30]
B
I mean, I had, you know, a very, very short lived existential crisis after Cruise. I was like, oh my gosh, I'm done with this company. This, this is like, you know, practically my identity for a full decade. What next? I spent some time thinking about that there's, you know, you could retire, you could become a venture capitalist, which is kind of like, yeah, are those the same thing? I think.
- [21:48]
A
I don't know if half of.
- [21:48]
B
Just kidding.
- [21:49]
A
We work super hard.
- [21:50]
B
And then after, after thinking about that for a while, I realized like the thing that, you know, outside of spending time with my family and friends, the thing that brings me the most joy is solving really hard, really hard problems with really smart people. And so that is retirement for me. That's like the most fun, satisfying thing that I could possibly think of to do. And it also ends up being you can do more of that and do it at a larger scale. If you work with a big team of people and you do it in the form of a company as opposed to a hobby or something, that you're doing it on your own. And so to me, I think there's no better thing. And maybe at some point I'll run out of energy to go hard like I am right now. But for now, this is great. We have a brilliant team. We're going on this, building this exciting new product in a big market. And that is energizing to me.
- [22:32]
A
I want to talk about a couple of the things that you've said about how you want to build this time. One that stuck out to me was that you never want to be more than a hundred people.
- [22:40]
B
Yeah.
- [22:40]
A
And first of all, I actually didn't know, is that like literal or is.
- [22:43]
B
That directional to be seen? Okay. Yeah. I think right now we're taking it very seriously. So if that is actually your belief and talk about why, but if that's actually your belief, then you make very different hiring decisions. It's like, well, you know, if I think about the Future company having 100 people in it, I can allocate this many people to this type of role. That means every person in every seat has to be the best in the world at this for the company to be successful. And so you end up, you know, passing on a lot of people that are great people, really talented, but they're not, they're not at that specific level we want for that particular role. And I think, you know, if you're successful in doing that, you end up with this. There needs to be a name for it. But like in the early days of a startup when everyone is like on the same page, like maybe just the founders, they're all in it 110%, they're all usually like brilliant working together, they're almost mind melded. And then you have insane productivity for some period of time until you get bogged down by the organization growing and adding more functions and teams of people and management layers, incentives, disconnected communication issues. And so you get this drift away from this pure force of energy that is in the beginning stage of a company. And so the reason for trying to have a cap on the size of the company is to keep it so that we're always in that pure high output zone. And you can't get that if you have like too much of a range of people in the company. I really think of it more like a pro sports team. Like you're not going to have, you know, the Lakers. I think you're gonna have like LeBron James and a bunch of high school kids on the team. It's like they're all players that are the best in the world so that you know, when they work together as a team they can outperform a team that is like a mix of talents.
- [24:13]
A
Like if you have like the LeBrons with the high school players, like the LeBron's are like what are we doing here?
- [24:17]
B
And they want to, they want to go play with the best people in the world against the best people in the world. And, and that's how you get better and grow. And you know, people who are people who are the best in the world at what they do typically got there because they have this growth mindset. They constantly want to get better. And you know, what better way to do that than to surround yourself with people of different skill sets that are all the best in the world at what they do and sort of absorb from that.
- [24:38]
A
I think so much of what gets hard is as you start getting into like scaling operations and you get into like that side of things, it just gets so hard to keep it really small. You know, like even you think about like let's say you only had like 10 non engineering roles, it's like, well some, someone's got to run finance. They probably can't do it alone. You've got like, you know, you're going to have all these like physical parts. You're going to have to have buildings for things. Like, so how do you think you'll actually try to keep a limit on that? Like will you partner? Do you go sort of like work with people outsourced or do you actually think that like, you know, maybe with like new AI tooling you could just go way further with people and it's just sort of like a demand deal place?
- [25:14]
B
Yeah, it's a good question. That's, that's part of why I said to be seen like this is a great mental model now and it may not scale.
- [25:20]
A
And by the way, I think most of the, one of the healthiest changes I feel like I've seen from five years ago is the shift from thinking that like big teams are cool to thinking big teams are lame.
- [25:30]
B
Yeah, I mean things seem to ebb and flow right. Like I'm taking the extreme position here, but I do think if that has the effect of, you know, causing a small shift in that direction, that's probably net good for the industry and good for these companies. So it'd be a question for us, do we partner or outsource things? I think, you know, keeping the team small also forces you to focus on like what are our core competencies, the things that we need to do uniquely because we think we can actually do them better than any other company that we could potentially work with with and for things like a lot of operations or facilities or buildings. These are things where maybe we have no reason to think we would be the best in the world at this. So we should partner. And a lot of companies, they have lots of funding, they have lots of teams. It's almost like they take on these responsibilities because they can, not necessarily because they should.
- [26:10]
A
I feel like one of the most important things which I think you've obviously shipped in self driving in a way that very few have. But I think in a lot of these sort of more sci fi areas it's very easy to not be in like shipping mindset. And like I think you did this really well at Cruise. Obviously like OpenAI was doing this like well before ChatGPT. And so you basically probably are in a mindset I assume of figuring out like how quickly can we ship and like iterate and like that's gotta be the mindset rather than just like hang in a warehouse building the perfect robot forever.
- [26:42]
B
I think for that it's, it's starting the thing you want to build and then working back to what is the main, what is the constraint, what are the constraints or bottlenecks that we need to be, that we need to make our number one priority because it cannot go faster than, you know, what that one bottleneck or constraint would dictate. And for self driving, that's a combination of safety, trust and public acceptance. And so, you know, those are different work streams where basically like unless those are all green, you don't have a product, it doesn't matter how good the technology is. And there are similar things, you know, for a home robot or really any business. And so like you know, mapping out what those are and basically making that the company's top priority. Like at Cruise for example, metrics were the single thing we talked about every week, week over week over week, making progress towards those. And I think for any company, like what you talk about, what you make design your metrics around kind of sets the tone for the company and it's gotta be aligned with that, you know, whatever the constraints are.
- [27:28]
A
What do you think you can do in a home first? Like what do you think will be the first activity that can like really be done well in a home? And then like what are the things that you think are close but maybe follow in the, you know, next 12 to 24 months or something?
- [27:39]
B
Yeah, I mean there are hierarchies I think of tasks for a home robot and if you look at I think two, like a classic two by two grid, I guess one is maybe the technical complexity of the task, like how hard is it to get a robot to do this successfully? And then the second is like what is the success rate that is acceptable to a customer of a product like this? And I'll give you an example. If you are, you know, in the easy side of things, from the technical capability and also the very forgiving side of things in terms of success rate is probably like picking up your kids toys. So I have, you know, two kids, a one year old and a seven year old and they're between the two of them are constantly making messes and toys are all over the house, running around picking up toys. And so if you have a product that you can buy, you can go to the store, buy this thing, put it in your house, push a button, turn it on, and then when you're gone for the day, all the toys are magically put away by the time you get home. It's like a mind blowing experience and let's say it screws up and like two out of the hundred toys are still on the floor when you get home.
- [28:34]
A
That's okay.
- [28:34]
B
Doesn't skip better. Yeah. So that think about nines of reliability for engineering. Like maybe 1:9 is fine for that particular task. There are other things like putting a wine glass in a dishwasher where the technical complexity is a little higher. And the what's hard about that by the way?
- [28:52]
A
Is it like the grabbing or is it like.
- [28:54]
B
Yeah, so if you think about picking up objects, this microphone which is going to make a noise when I squish it is compliant. And so if I'm off a little bit on where I grip it or like how much I squeeze it, I'm not going to shatter this microphone into a million pieces. For a wine glass, the margin is very thin. And so from a dexterity standpoint, it's a little more fragile.
- [29:11]
A
Actually sometimes I think about that's like a good example of a thing where I'm like, it's amazing that people can do certain things like squeeze a wine glass the right amount or like hit, you know, a ball, you know, with a racket or a golf club with the right angle or something like that. Or like catch something that's flying while you're moving. Like it's actually pretty crazy what you can do like mechanically it is.
- [29:31]
B
And the evolution to how we get there is interesting too because my one year old daughter, her hands are like open, closed, there's nothing in between. She grabs objects, the wine glass is shattering. And at some point along the way we developed much more new on skills and abilities. But so wine glass is another one. The other thing is challenging is if you're putting a wine glass on a rack and you know it's a thin stem or something and you nick, you like bump into something, you might break the stem off. Right. And so not only is it more difficult from a technical standpoint, but if you shatter a wine glass in someone's dishwasher, they're probably not going to be your customer anymore.
- [29:59]
A
That's right.
- [29:59]
B
And so that's like maybe several nines of reliability. And so I think that this, this sort of spectrum of technical difficulty and basically forgivability is going to dictate the types of things you see home robots do first. And, and I think we'll work our way up towards, you know, I think the holy grail of a home robot which is like dishes, laundry, end to end, maybe cooking these things, all of them have like all of these little, it's like a minefield. You do one thing wrong and you ruin the Whole process. Like, for laundry, if you put the red sock in with the whites, you now have pink laundry. That's like, game over. Right? And, you know, so there's things like that for cooking. It's the same thing. You put too much salt or pepper in there, and the dish is ruined. You know, so these are things that I think we'll get to, and I think it'll happen pretty quick. But, you know, I think you think.
- [30:43]
A
It'S like cooking a steak at some point.
- [30:45]
B
Yeah, why not? If you think about at the end of the day, you've. You have pick and place and simple manipulation. That's what cooking is. Yeah, they're just like, a much higher degree of reliability. And there's other things around, food safety and bacteria and other things that come in. Cookie cooking and temperature sensing and what like that. So it's all doable. It's just like. I don't know.
- [31:00]
A
Do you think at some point it's like, hey, robot, I'm at work right now. There's a steak in the fridge. Please cook it and clean up everything by the time, like, 15 years from now, that's. That's doable.
- [31:10]
B
Less than five.
- [31:11]
A
Less than five.
- [31:12]
B
Yeah. This stuff is going fast. Again, if you saw. If you look at the robots you can buy today in the world, like the nice robot vacuums, you may not think that. If you see what's happening behind closed doors at the best robotics companies in the world, you might think that. And if you're the leadership of these companies, the technical leadership, and you kind of know where things are going, you absolutely believe that.
- [31:31]
A
The hand seems really. As we're talking about this, I was sort of, like, stupidly like. I was like, actually, a hand's pretty good. Like, your fingers are, like, pliable. You have, like, a lot of degrees of freedom. You have, like, multiple grip points. Is the hand the optimal thing?
- [31:43]
B
The hand is really important to get right because it is the robot's interface to every object that it interacts with. If you make it too simplistic or not enough sensing capabilities or whatever, then you have to have a much, much smarter brain to figure out how to use this primitive tool to accomplish a complicated task. And so the more mechanical complexity or capability that you add to a hand, the more sensing ability. In theory, it would require less, you know, sort of rocket science to figure out how to do a task with that hand. The trade, of course, is the more technology, the more degrees of freedom or motors that you pack into a hand, the more complicated it becomes, which Impacts durability and also cost. And so there's push and pull there to find that sweet spot where you can basically come up with the simplest hand possible to do the tasks you want to do at the lowest cost, while while also being able to accomplish everything in a fairly straightforward manner. But I think in the limit there's a lot of, we think by analogy, a lot. And we have two hands and two arms. And so a lot of the robots you see today have two hands and two arms. But it is really interesting thought experiment. What does the ultimate hand arm thing look like? And I think it was Rodney Brooks who said this the other day, but I actually do kind of think maybe it ends up being some crazy octopus tentacle looking thing in the future that's very adaptable and can reach into small spaces.
- [32:55]
A
Interesting.
- [32:56]
B
Well, I think that the human hand was ended up where we are due to probably some impossible to unravel sequence of evolutional pressures. Right.
- [33:04]
A
Well, it's like you start down some path and then you do your best. Evolution does its best given some somewhat random starting point, I suppose.
- [33:10]
B
Right, yeah. So if you could go back like a million years and hit the reset button on human evolution, maybe something, a new fork would emerge and it would be more tentacle like, or who knows what. But I am skeptical that the way that human hands and arms evolved is the ultimate. And so the challenge will be like, can we figure out what that is?
- [33:26]
A
I have a couple stupid questions about the robot at home. One is, how strong could it be? Like, is it is a hundred pound robot, like it must be ridiculously stronger than a person. Right.
- [33:36]
B
I would think for 100 pound robot you could certainly make it maybe stronger in some dimensions. There are some things that like our sort of soft biological muscles are pretty.
- [33:45]
A
Good at are stronger than like, like a physical robot, pound for pound kind of thing.
- [33:49]
B
Yeah, it's really hard to say. I think so. I think probably the state of the art Boston Dynamics robot seems like it's on par, if not, not more capable than a human. And if not now, I'm sure the next couple generations will be. So that's kind of interesting.
- [34:00]
A
That's surprising that a soft muscle is stronger than like. I don't know why I would think a robot could be dramatically stronger.
- [34:06]
B
Yeah, I've been going down the rabbit hole on this a little bit thinking about like, you know, as, as again, our focus on affordability and cost. Like is a electromagnetic gear motor where you've got magnets and copper winding and a bunch of gears in A housing. Is that the most cost effective, durable way to generate motion for a robot? And the answer in the short term is probably yes. But I think there are some interesting things happening where we're trying to mimic either some of the chemical processes or electrostatic actuators or other things that are similar in how they work to like a human muscle. And the benefit there is you can get a higher cycle count, more silent operation, and potentially more power density. Like how much strength can you get into a physical volume than what we have today in gear motors and then potentially much, much beyond what humans have in our muscles.
- [34:47]
A
Isn't like hydraulic, is pretty strong, like a hydraulic. Hydraulic. Hydraulic pressure is pretty strong.
- [34:51]
B
Hydraulics can be extremely powerful, but they have other trades. Typically noisy, the valves and things are pretty expensive, can be harder to control and get, you know, high fidelity motion. And so in terms of power density maybe good. But you know, there are other trades and the reasons you don't see these on a lot of robots, that makes sense.
- [35:05]
A
Another question I had that's sort of like probably off spec, but while we're talking, is this going to be something that would have like home security applications as well or does that then take you into weird territory that's just not worth going to?
- [35:19]
B
Yeah, I think so. I mean one of the challenges with a, with a home robot is it's kind of general purpose. And so like, you know, what are people going to use this thing for? And I think it's, it's hard if you just have a laundry list of 50 different items that the thing can do and security is one of them. But I do think a lot of people will be out and about and with their home robot at home be like, oh, I wonder if I forgot to turn off the gas on the stove and send the robot over there to just, you know, tell you or even take it on for yourself. In the same way you could be like, hey robot, like, you know, if you see any person in my home or any doors open, like let me know.
- [35:47]
A
Yeah, if you see me getting burglarized, like, do something.
- [35:51]
B
But I, I don't know if you will think of it as a security robot so much as like, this is just one of the many responsibilities of my home robot is to keep tabs on my home.
- [35:59]
A
Totally alerting probably is good. Taking action is probably not.
- [36:02]
B
I hadn't thought about that side of that. I, you know, that's not really in our. That's good.
- [36:06]
A
That makes sense. I'm just thinking because like, you know, in my Head. I'm like, okay, if there's this brilliant, capable robot in the house, my guess is you're going to have a lot of people want it. To start Doing a ridiculous number of things for them would be the arc of time, and then you'll have to choose from that set. What goes in?
- [36:22]
B
Yeah, I think so. But for sure, on the security side, I would hope, though, rather than having physical deterrence and having your home robot turn into a security guard with a baton or something, it's more so that it just becomes unattractive to. To rob homes or do, you know, break in and enter into a home. Maybe in the same way that, you know, a world full of cars where everyone has, like, that Tesla Century Mode, there's very little incentive to break into cars. It's not worth the risk.
- [36:48]
A
Well, I mean, even like a security system just makes a loud sound and calls the police, you know, I wonder.
- [36:53]
B
I think that's pretty effective, isn't it?
- [36:54]
A
I think it's extremely effective. Yeah. And I think those systems are pretty old and, you know, they're deeply embedded and.
- [36:59]
B
But yeah, it's like you may figure out how to disable the alarm and sneak into the house, but if there's a robot, you know, rolling around and.
- [37:05]
A
Then a siren's blaring and all that stuff, I just think it'll be interesting where, if this gets in there, my guess is people will start to. I could see a future where people expect a ridiculous amount from these things.
- [37:16]
B
Well, I mean, it touches on something interesting. I've. I've thought about is, like, when you ask people, or we ask people, what would you do with a home robot? You know, immediately what comes to mind is, like, the thing that's most annoying to you today to do in your home, and I think that's good. We want to help with the annoying stuff.
- [37:28]
A
And what comes up most, like laundry, probably.
- [37:30]
B
Yeah, laundry, dishes, picking up after my kids, you know, wiping surfaces, cleaning. Like, these are the things you would expect. And so we're going to chip away at those things for sure. But what I also like to think about is the things that we don't do because we value our time more than that. You know, the example is if you've ever gone to, like, a really nice hotel, you know, the slippers are laid out for you. There's a glass of water on the nightstand, a little chocolate on the pillow, all these, like, little bushes, like, I don't know. I think that, you know, robots should not only automate the things that we don't want to do, but also like elevate our standard of living to some degree.
- [37:59]
A
Yeah.
- [37:59]
B
And so I love the idea that if you can afford a really affordable home robot, we're going to give you a lifestyle that that would otherwise be inaccessible to you.
- [38:06]
A
Totally. I mean that's actually a really interesting point that a lot of the types of things you're talking about don't require any new inputs. It's just about taking care of your home in a certain way that's beyond what you would normally need. But it's like you've got a bunch of towels that are sitting in the laundry room that are clean, but can you put those by the shower and roll them up nicely?
- [38:25]
B
Yeah. And maybe you don't need all these things, but the point is your time is more valuable than that. It's very scarce. Humanity's time I think is really important. But for a robot that's got 24 hours to sit around in your home and like try to make your life better, what could we come up with for it to do. What could it come up with to do for you? That's an interesting question.
- [38:41]
A
Wow. I'm curious about reflecting on what you've learned from the way self driving cars played out and how it might matter here. Maybe one interesting sort of case study is the Tesla versus Waymo approaches. Do you think in any way that how that played out or any learnings there that poured over to like what could be impactful in the robotics land?
- [39:05]
B
Well, it's hard to say that. Very different approaches to getting to market. But it does seem like they're both trying to converge at the same thing which is self driving cars everywhere. I think one thing that was really brilliant about Tesla's approach, they found a way to sell the product essentially before it was fully complete. If we're looking purely at the self driving side and generate billions of dollars of cash flow which they could use to, to bolster the core business but also continue to invest in R and D to to make this self driving product. Waymo by comparison, you know, has taken almost a couple decades at this point, maybe not quite that long and probably tens of billions of dollars of investment. And the revenue relative to that has been fairly meager.
- [39:46]
A
Right.
- [39:46]
B
Compared to that total investment over time. Which basically means that the only companies in the world who can do this are the ones with that kind of capital on their balance sheet to basically fund this crazy amount of R and D year over year.
- [39:57]
A
Yeah.
- [39:57]
B
And I think it's no coincidence that the only Companies who succeeded in that approach or are on track to succeed in that approach are owned by, you know, Amazon, Google, or you know, like a major car car company. And that was even a struggle for a company like General Motors. And so in the home robot space, I hope we don't repeat that. I hope it doesn't become the case that the only companies that make it are ones that are basically kept alive through billions or tens of billions of dollars from, you know, a corporate benefactor. And instead we can find clever ways to get to market that.
- [40:26]
A
Which I guess is why you need to get to market and be selling something along the way to fund all of this.
- [40:31]
B
Well, I think if your development cycle means you don't get to meaningful revenue for five to 10 years after the company has started, it means you're entirely dependent on either being acquired or the capital markets being pointed in the right direction. And historically things tend to cycle back and forth in a five to ten year timeline, you're getting awfully close to almost guaranteeing that you straddle a down cycle as well as an up one. And that can be a killer for these companies.
- [40:55]
A
I, we don't need to talk about sort of Cruise and GM too much, but I am curious about sort of. You know, I saw you share on Cheeky Pint about like not wanting to sell and I'm just curious like your mindset about the sense of autonomy and how you think about selling a company since you've been through it, you know, a couple times and everything.
- [41:15]
B
I think, you know, and I said this before, but my conclusion is like, if, if you are selling a company, it should be because the reason you started the company or the thesis that you had in mind or the thing you wanted to build, something has changed and maybe like you're no longer interested in it, your life circumstances have changed, whatever. But I think it is a fantasy to believe that you can sell your company, like have your cake, you need.
- [41:35]
A
It too, sell your company in order.
- [41:37]
B
To further the mission and further the mission. I think in theory this can happen sometimes, but it is so, so rare.
- [41:41]
A
It is rare.
- [41:41]
B
And I think more likely than not you'd be disappointed with that outcome. And therefore, for me, I can't imagine being a situation where I would trade, you know, the opportunity to build this amazing thing and control it and make sure it happens in the way that I want it to for some kind of partnership or liquidity. So that just doesn't make sense to me. Maybe not for everyone and maybe that's just because I'm so excited about this thing and bringing this new idea of a home robot into the world that it just wouldn't even cross my mind, the thought of handing over the rans to someone else. Yeah.
- [42:08]
A
You're also at a point now where this type of company is such a forever project, and you're now able to start a company that's like, you know, it's not. It is not some little thing, like, if this works, it's just such an important thing, which also, I guess that probably also drives you to want to sort of hold on to it indefinitely.
- [42:24]
B
Yeah, perhaps. But I think I'm not selfish about it. Like, if I, you know, I feel like I have an obligation to stay true to, you know, our investors, the employees, and the mission. So, you know, even though I certainly want to hold onto this, I'm not treating it like a pet project. I actually do want to, like, fulfill this broader vision that. That we all share.
- [42:41]
A
Yeah. Maybe as a final thing to touch on, you did this crazy, like, marathon around the world experience. What was that? And like, why'd you do something that seemed so, you know, hard deep in the.
- [42:52]
B
In the middle of cruise? I was. I think I was frankly, like, kind of frustrated that, like, we were putting in all this energy. And sometimes there would just be periods where the metrics wouldn't always go up and to the right. We'd take a regression and then go back and forth. And so, you know, the result wasn't always proportionate to the energy going in. In. And for running, for me, at least, that was not the case. You put in the time you're running. Yeah. And so that's very deterministic and satisfying. And so I needed something to balance that. I feel like in my life. And as I do, I went down a rabbit hole reading about, like, extreme marathons that you can do. I was like, sort of an amateur marathon runner and came across the World Marathon Challenge. It's this thing you can sign up for. They take you to each continent, one continent per day, and you run a marathon on each one. And then, like, the next day you fly to the next continent. And I thought, that is insane. And then in fine print on the website, it's like the world record is like 5 days and 10 hours by this one guy. And then I got the wheels turning. It's like, well, I wonder what the theoretic engineer brain clicks on. I wonder what the fastest theoretical time you could do is if you optimize the. It's like the traveling sales. What part of the Continent, Yeah. Where you land, optimize for customs in and out and logistics, and really dial it up to 11. And that turned into an 18 month obsession. Got stuck in my head and I ended up writing some software to find the shortest route between the seven continents. That's crazy spending.
- [44:05]
A
My problem is that I couldn't run a half marathon. That's where I would struggle.
- [44:08]
B
This sort of stubbornness and attachment to this idea meant that part of this was I had to train my body physically to be able to do this.
- [44:13]
A
Because like you're running a marathon and then you're not resting afterwards.
- [44:17]
B
Yeah. So the cycle, for example, is you start in Cape Town so that you fly to Antarctica. You have to start there because the weather is so unknown.
- [44:24]
A
Oh yeah, the Antarctica one. That's tough.
- [44:26]
B
You need like at least a, A call it a six hour window of decent weather and you're like looking at the weather forecast. When it clears, then you fly in and you land, you run the marathon and you get out because that can throw off the.
- [44:35]
A
Where in Antarctica do you do this?
- [44:37]
B
On like the most temperate outer part of Antarctica. So it's on the continent, but we're not talking like South Pole. Yeah. So it's cold, but it's not. It's not like.
- [44:44]
A
But it's not snowy.
- [44:45]
B
I mean, it's icy. And you can't say it's like barren, all ice. And you land the plane on ice, you run on ice, you're running on ice. On ice. It's kind of like crunchy ice. Like there's like a ski slope groomer that did a course down there. It's like a six mile loop or something. And so it was like running on. It's like trail running. Got it. Still pretty crazy. That's crazy. Anyways, yeah. So organizing the logistics for the training. Training my body and everything. Doing it in the end, ended up doing this in about three and a half days, which blew the world record aside. And then after 18 months doing this and finishing it, you know, I gotta say, like, it was, it was. You ever like finished the last item on your to do list and it's just like a dopamine hit and you're like, oh, that feels so great. Yeah, that's what it was. It was like relief. It's like, check the box. Now I can. My tormented brain, which wouldn't let this go for 18 months, can finally relax.
- [45:28]
A
Where do you go? Cape Town. Antarctica. South America.
- [45:31]
B
Yeah. Southern tip of South America. And then to Panama City and then I think to Madrid. And then Oman and you're just like.
- [45:37]
A
By the last one, are you just crazy fried or were you in shape to just not be, you know, so beaten down?
- [45:44]
B
I was pretty fried. Pretty fried, yeah. But it's like, you know, the training I did peaked it. Doing three marathons within a 24 hour period in three different cities, that was like the, the stress test for this, if you will. And my coach was like, you know, I was like, only three marathons, that's not seven. Like, is that actually the right amount of training? And he said, I promise you, when you're doing the really real thing and you have this whole crew of people with you and everything is on the line and the adrenaline going, you're. If you can do three in 24 hours, you can do the rest. And he was right.
- [46:11]
A
That's wild. That's like the hardest physical task imaginable.
- [46:14]
B
You know, mental toughness is important for startups and I feel like it really helps me quite a bit in that domain.
- [46:19]
A
Yeah, totally. All right, Kyle, this is really fun. Thanks for making time for it.
- [46:22]
B
Thank you.