Google's Secret Robotics Play That Nobody's Talking About

22:49And they argued earlier, the hardware is actually really capable. The software that is built in to the robot controllers is really capable for what it does as well. So if you buy a robot, it comes with a robot controller. It's like a special purpose computer. You can do really capable things with it in a relatively narrow context. But what's difficult is that that software, because it was kind of developed in a time when you were, the applications you were building were of that kind of high volume, highly repeatable, sort of blind applications, not strictly, but sort of like you're kind of closing your eyes and then you know where the object is and you're just going to pick it up and you're going to put it down and you're going to optimize the time to do that, right?

23:33There are good tools for that. What has been more difficult for that software ecosystem, because it is software, right? But what's been more difficult is how do I bring in the latest sensors? How do I bring in the latest AI models? Is there a place in there where I can host the inference that we talked about earlier? That's certainly part of it. Another part of it is that historically the companies making the software have also been the companies making the robots. And the two have tended to be kind of locked together. And that's, you know, that's worked well for commercial reasons for some of the players.

24:10Very often the customers, the end customers who are getting the robots would prefer to have a little more flexibility. And so this is, you know, this is one of the things that, you know, if you go back to the Android analogy, an Android application, a well-formed Android application can run on kind of any Android device. And it doesn't matter whether the manufacturer was a, you know, Pixel or Samsung. And that's what we think is going to really change how the willingness of people to be able to deploy these applications is knowing that they've got the freedom. Sometimes it takes some adaptation, like, you know, you could have two robots that are about the same size, but the joints have different limits.

24:50The workspace might be a little different. There could be some tweaks that are required. But by and large, being able to write an application once and then ideally only configure it in order to run even on a robot hardware from a different company is that's a that's a game changer. Yeah.

25:06Grant:So as we're thinking about where this goes from here, is the real challenge or the real bottleneck, if to the extent that there is one, on the training and the data side? Or is it on the actual deployment of these tools at scale? I think it is. So it depends on what problem you try to solve. So I think on the training side, we're really, really good now as a, not just intrinsic, but as an industry, we're really good at generating synthetic image data. So if what you want is to be able to, you want to train a system to understand images or even video, boy, we're excellent at that. Now we've got, you know, we've got these loops where I can start with some data.

25:49I can train a generative system that will then generate synthetic data, which at a higher volume, which then I can train, use to train the next step, right? That's working really, really well. The training data for physical contact with the world, that's harder. And that's because we understand from first principles how to, like, we have an understanding of how physics works at different levels, right? But it turns out getting very detailed models that are accurate is pretty hard. And most simulators take what's called a rigid body dynamics approach. So I just imagine infinitely rigid things that are just coming into contact and they probably have like a single point of contact, like a billiard ball model, right?

26:33Like a simple Newtonian kind of model. And I can compute the point of contact and I can compute what the reaction force must be and then where they ricochet off. But the real world is so much more complicated than that. There are many points of contact. The surface properties matter a lot. The texture, the friction properties, like friction is famously difficult to simulate very well, which also makes it difficult to generate synthetic training data for. So that's one gap is how do we get data for that physical, for the in contact kind of tasks? And that's why you see some companies having, you know, essentially armies of human teleoperators who are moving robots around and then gathering data from physical systems because it's so difficult to generate synthetic data.

27:18We're getting better, but that's still one gap.

27:21Grant:Is that just because you need the physical world contact to really model it effectively? Or is it because there's some fundamental property of friction that we haven't quite modeled mathematically or in the code? This is where I've got some of my colleagues are smarter about this than I am. But I think if in principle, we understand how, let's say, frictional contact works. But in practice, it is very, very difficult to implement a way to compute it that is tractable. It is if what I want to have is a if what I so if I'm studying like in like like there are great models for cars. So if you, you know, you think about you're going to design a new car or you're going to tweak something about the tire design and you want to understand how does that behave as it's coming into a corner at a particular speed?

28:10And there's this much water on the surface like those specialized models we have because of the need for them. The industry has a really, really good understanding of how that works and we can compute that. It might take a long time to compute, but that kind of specialized model we've got. If what you want to ask instead is, well, I've got a robot and it's in a factory or a warehouse and it's going to interact with all these different things, which have all these different properties. And I want to get the right answer for all of those interactions. That's a lot harder.

28:35Grant:Yeah. There's just so much more. It's just like, I don't know the exact term, but it's a greatly expanded. I don't want to say exponentially, but it's a vastly larger space that you have to model for. It's much more difficult. And so the places where very, very, very detailed simulation has gone really, really far is in those domains where you can narrow it down. You can say like, you know what, we care a lot about tire to road surface contact. That's super important. Let's build a very, very deep simulator for that. You can look in aerospace and say, you know what we care about? If I'm building rockets, I care a tremendous amount about how the fluid dynamics look like inside the thruster cone as the fuel is being burned.

29:18And I want to understand everything about that. I can build a very detailed model of that. And it's incredibly good and incredibly predictive. Expanding that idea to cover the variety of interactions that a robot can have in the just physical world we find around us, that's a higher bar. But I want to come back to something else. You're asking where the gap is. So certainly on the training side, there are some areas around like contact and manipulation where we can improve. Frankly, I think one of the things that's missing is knowing what the right applications are for the robots. I think as, and I can, you know, speaking as a roboticist, I think robots are awesome.

29:53I just want to build robots all day. Right. So I've got, I've got this hammer, which is a robot and I'd like to make it the best hammer ever. Now that, that's not the same as knowing what to hit with it. And that's where I think we, a lot of the attention, which has pluses and minuses to robotics today, is much more about the technology push than it is about a pull from an application that is really going to make sense. And by make sense, I mean, you need to believe that you're going to be able to deploy a system that has the reliability and the cost so that a customer would actually pay for it.

30:33You can maintain it in the field. There are all these extra constraints. And so what I like about working at this place where we are at Intrinsic and this software platform ecosystem model is we put these tools out there and then it lowers the barrier to entry and brings many, many more people in who they might have the best ideas for what the applications are. Like we at Intrinsic, we've got some hypotheses and we're exploring. We already have deployments with customers and we've got some, you know, what we think are like really great ideas for those applications are. But we don't have a monopoly on those ideas, right?

31:08Like I think where the more, the bigger impact play here for us is we put the tools out there and then we get what that means is like somebody who doesn't even come from a robotics background. Ideally, I think they generally have the best ideas. They're going to come in not enamored of a robot as a technology and look for a place to plug it in. Rather, they're going to come at it from the point of view of, you know what, I don't know anything about that technology, but I understand what this customer over here wants. I understand what the need is. Now you, and not know anything about robots, you've given me the tools I need to build that application and then go work with that customer, right?

31:46I think that's that is what is that's what historically has been missing in robotics is the ideas for where to apply the robots.

31:55Grant:And I think, I mean, I suspect you agree with this, but I'll get your take. Making it open source in the way that you have and making the tools as accessible as the way that you have enables that so much more. Because now, you know, the 80 % of people who work in manufacturing who couldn't, you know, automate things before, now they have a platform where they can actually do that. And you have a good distribution system for it. 100%. I think on the, you know, one of my favorite things to do every year is go to ROSCON. It's the ROS Developers Conference. That's awesome. The Open Source Robotics Foundation, which is the nonprofit that is responsible for the Ross ecosystem, that organization puts on this conference every year.

32:34It'll be in Toronto in the fall. For anybody who's watching, please come join us in Canada for RossCon. And that's a place where one of the reasons I like to go there, other than just being able to see lots of friends and colleagues, is you get to see the variety of applications. So if you roll back to that origin story for Ross. And, you know, I mentioned we built it for the, it was a robot called the PR2, which was this large kind of human-sized robot on wheels with arms and cameras. Okay. Looked a little bit like Rosie, the robot from the Jetsons, if you split. So, but having said that, like we built it for that for use in research, but now today that because it's open source, that platform is used by people across like every application that you can imagine.

33:18So you've got people running off on not just robots, well, who else, but robots with legs and also robots that fly and also robots that float on the surface of the ocean or dive underneath it to understand what's happening from an ocean science perspective. you've got robots doing uh like campus security you've got robots doing food delivery you've got robots doing autonomous driving on the road like basically every application you can imagine most of which we did not have in mind when we built the platform and that that to me is just amazing that's that that's one of the one of the reasons that i personally identify as a tool builder because that's if you put the right tools out there and and especially if you provide them in a way where people are free to do as they like, where open source is one very permissive model for doing that, not the only way, but it's a very good way to do it, then you don't know what will come and you get to be surprised by it.

34:13Grant:I want to talk about the idea of like you mentioned, some companies are hiring lots of groups of humans to manipulate robots and move them around. And one of the things that people always think about when it comes to automation is what's the impact on jobs and how is the future going to look? And as someone building tools to help manufacturers and all of the different use cases that you just explained, create new automations to enable new use cases, what do you see as the role with humans and we'll say like autonomous platforms going forward? Do you think that we'll be all just builders like you are and spending our time building?

34:49Grant:Do you think that there will be a lot of people who are data labeling and that's still going to have a lot of room to grow? What's your suspicions as far as where this goes, like the human and robot relationship, let's say? It's a super important question. I think that the, you know, first of all, I don't, I think people, you know, we talk about autonomous robots. Autonomy is probably not exactly the right term because autonomy, it implies like, it implies is intent and desire. And, you know, things that we create don't have intent or desire. They have some kind of capability to perform tasks.

35:29They don't necessarily desire to do anything. So I think that what that means is that they are always going to be working with and for people. We're the ones who give any system that we build, whether it's a, you know, at home, it's a washing machine. It's a laundry machine that I say, you know, I give it intent by saying, go. I have a robot in a factory. I give it intent by telling it what I want it to do. I'm going to be, there are going to be humans interacting with these systems. So thinking about how the humans interact with those systems over time is that that's going to be, the need for that is only increased as we see more applications.

36:09Like sometimes people think about having, Like introducing robots means we take the people away. I completely disagree. Like it means we, it means we don't need to think about the people anymore. I think we're in a, you know, we are people and we are a world that is, you know, populated by people in the places where we want to put robots by and large, unless they're like in the depths of the ocean, even then you still need an interface to talk to them from afar. You're always going to have people in the loop somehow with these systems. And so I think the human interface aspect of it cannot be overemphasized.

36:42The kind of like labor market side of it, there's a lot of concern about robots. You know, I don't think I've done an interview with anybody in the media for the last 20 years without someone asking me, you know, are robots going to take all the jobs or are robots going to rise up and kill us all? Okay. So I think, I think neither is going to happen. I look at robotics as a technology development in a long, and even AI, it's a long history of technology that we as a creative species have developed. And there's been, you know, you can pick any point along the way. There's been human cry over, this will be the one that ends, you know, life and society as we know it, right?

37:22And it's never turned out to be like that, right? Things are genuinely disruptive, right? And I think robotics and AI today are going to be, they are and will continue to be genuinely disruptive. That disruptive is not the same as, you know, catastrophic or world-changing. And also, none of these things is going to change at the rate that I think most people are concerned about. So my expectation is that this technology will, like pretty much every technology that came before it, it'll cause some disruption. It may actually cause some, and it will likely cause some pain for some individuals. There will be people who will perhaps have a bad time because of the technology.

38:04the technology. Ideally, we make, we have a way for those people to be able to adapt, upskill and so on. But in the long run, I look at like just greater productivity for society. And that, that's been the story so far of technology development. I think it's a little bit like, there's a little bit of hubris in imagining that like, this is the one time that it will be different versus all the other technologies, right? Like this is because we're very proud of this technology or because we, you know, we think that we're lucky to live in this time, this will be the one time where we've developed something that the impact is just so qualitatively different from everything that came before.

38:40I don't know. I think it's much more likely that we as a society find interesting and creative ways to incorporate, to adopt, to adapt to it. And the future is going to be, it's going to look different from how it is now, but I don't, that doesn't keep me up at night.

38:54Grant:That's fair. Well, I mean, you studied robotics in undergrad. I mean, do you still see that as being like a really good path? Like let's say, Because I see two benefits to many, many benefits. But if I'm thinking just of the more, the most obvious benefits possible is a lot of manufacturing in this country. You know, we're both based in the US. You know, other countries, I'm sure have their own concerns here. But a lot of manufacturing in this country has left, you know, we don't manufacture as much here as we used to. So perhaps bringing more of the ability to automate things here could lead to more companies automating things here, and that could lead to more jobs here and around the world, wherever your technology gets used.

39:34Grant:And then I also see a benefit in, you know, everyone having the ability to create robots. So could everyone or people who are potentially displaced, you know, an automation way of hitting, you know, the place they're working now, could going into robotics be a good path for them? Should they go to school for it? Or what would be your recommendation if you're interested in this field? I think it depends what, you know, so first of all, I agree that there are the benefits that you laid out. I think that does totally make sense. I think that there, I do hope that people continue and even more people study robotics, whether they're coming at it like in early or mid or later career.

40:13And I think that where you are in that journey, probably there are different aspects of it that you might be studying, right? So I think that there are, I mean, you can already see this in factories today as automation gets introduced. You have people who were previously doing the tasks themselves manually who are learning to be robot operators. And the way that I think about that, like it's too easy to dismiss that as like, well, the robot took away my job, right? I think it's much more productive to think about it as look at this power tool that we've got. This is like, you know, you're probably already in this job using some tool, just a really great tool.

40:54It happens to have a bunch of software in it, but it's still here to help you do your job. And I think thinking about how do we make it possible for people to be able to figure out how to interact with these robots. And then increasingly, this is not something that is production ready yet. But those like we traditionally we've had a separation between developers and operators. They're kind of two personas that we think about, right? You've got like the people who build the system. And we think about those people as needing to be, you know, engineers who have a background in, I don't know, say computer science.

41:27And they can develop all this code and they can build the system. And then there are the operators who have a different background. They understand the work to be done. They know the details of, I don't know if I'm going to weld two things together, which I know nothing about. Like there's a ton of what's called process knowledge that goes into that. I'm like, exactly how am I going to ensure a high quality result? And we've tended to separate those two people and personas and say, well, the second one, they're not in a position to change how the robot behaves because you need to be an engineer to do that.

41:58Well, with, if we make the software more capable and we start to take advantage of some of the generative AI techniques, maybe that person can change the behavior of the robot. Maybe that's something where like you're like you, instead of thinking about yourself as an operator, I don't know, to very loosely borrow the term, maybe you become a creator of a sort. And you have, you take your on-the-job process knowledge skills and you use it to prompt or whatever the right interface is to the robot to get it to change its behavior to help you do your job in a better way. And so I think that line becomes a little bit blurry between are you like one of the, you know, trained engineers who's allowed to change the system versus a user?

42:40Maybe

42:40Grant:the users are now effectively developers right it goes back to your earlier point about a lot of times the customers are the ones who know what needs to be built because they're like getting the you know whether they're in the middleman who's getting the requirements from their customers and then they needed something that fits that requirement or they're the ones actually on the ground like hey i need to build something that helps me yeah so it makes it it kind of empowers everybody in the field in a really cool way which is why i think i love what you're doing that's the hope. And, you know, I think where we are right now is just what I want to see from us, but also from all the other kind of peer companies out in this space is let's get more and more systems out into production.

43:22I have one of my colleagues actually had a t-shirt made for me that just says deployments, deployments, deployments. I'm known to be kind of dogmatic about saying, okay, that's an awesome demo. I love that. How do we get that deployed to a customer? For me, that's where you prove the value of what you've built. The value manifests itself when there's an end customer, which for us, again, there are lots of applications for robots. For us, we're focused on manufacturing. So our end customers, they've got factories and they've got work to be done. There are jobs to run, there are widgets to make.

43:56For me, the value of what we've built manifests itself when there is a robot that's using our tools that is doing useful work for that customer. That is, you know, they're, that's where I want all of us as an industry to get to.

44:10Grant:Yeah. And to your point about generative AI helping their, like a lot of people now, because code is somewhat trivial to create. I mean, I'm sure you might have thoughts on that. You can create demos a lot quicker. And the benefit of creating demos quicker is not just like, hey, look, I made a cool thing. It's like, no, no, get it out there and have people test it and get feedback and just reduce that feedback loop as much as possible. As a CTO, are you using any coding tools in the software you're building? Absolutely. Yeah. Or, you know, I encourage our team to use the increasingly I encourage, I should say, you know, I learned a program a long time ago and that, that gives me maybe like a somewhat skeptical bias of the, against the new tools.

44:51Grant:I think that's fair. I'm really, but I, I'm, I'm coming around on that and now actively encouraging our team to pick up the latest tools because they are, they're, they're getting better and they're getting better so quickly that the, you'd be crazy to ignore them. So the kind of productivity that you get out of the latest so-called agentic development tools, it can be extraordinary. So yeah, we're absolutely using those. And we're expecting our customers and our users to use them as well, which means we need to think about how do we make our software kind of amenable to use by those tools. You went to space camp as a kid in the 80s and now software help you helped build literally runs on the International Space Station.

45:30Grant:How how does that feel? And like, what would you want to do with like, where where do you want to take this software in space? Like, what are your thoughts on this? Like, yeah, I yeah, I did. I was when I was in middle school, we were living in Georgia, and I had the great privilege to go to the space camp in Huntsville, Alabama, which was amazing. You get to wear a jumpsuit the whole week and kind of learn a lot about the space program, kind of pretend to be an astronaut in a lot of ways. That was fantastic. At that point, like every kid who had that experience, I thought, I'm going to be an astronaut.

46:06And it turns out I'm not an astronaut. That's okay. But we, along the way, especially when I was running the previous company, Open Robotics, I took every opportunity to work with NASA. Like any time that our friends and colleagues at NASA called and said, hey, we've got this like crazy research project. You guys want to help out with it? I was like, yes, every time yes. Just because I think there's such like there's just such inherent virtue in space exploration that it's always worth doing. And one of the projects that the folks at NASA Ames did is they built a what's called the Astro B. It's a cube shaped robot that's like 18 inches on a side.

46:44and it's they've got i think three of them inside the international space station they got fans on each side and they free fly around inside the space station and they help astronauts to do their work they can do some kind of like surveillance tasks inside the space station measuring things they could look over the astronaut shoulder and give ground uh mission control a uh like a over the shoulder view and that runs the that runs ross knowing that there's some code that we wrote that's up there is just extraordinary so for me if you ask like where do I want to take this. I want to see, I want to see like robotics and automation broadly construed.

47:17It's not, it doesn't just have to, it doesn't have to look like a robot arm. It doesn't have to look like a humanoid. Like what I just described, it's a cube with fans on the sides. Still, it's a cool robot. I just, I want to see that technology go as broad as possible. So, you know, at Intrinsic today, we are building a business and it makes a lot of sense to focus on manufacturing as an industry. That's not where I personally want to stop. That's a stepping We go from there to adjacent industries and so on. Eventually, maybe we're providing software that is helping with space exploration or ocean exploration or who knows what the next frontier is.

47:54Grant:Yeah, I mean, you could at least manufacture rockets and things like that that will take you to space. We can certainly help with it, yeah. Yeah. Brian, thank you so much. Where can people go to learn more about Intrinsic and ROS? Where would you like to direct everyone? Sure. Intrinsic.ai is our company website. You've got some jumping off points there. In fact, we're running a, together with the Open Source Robotics Foundation and some other partners, a challenge right now. We call it the AI for Industry Challenge. So there's a open competition to try to solve some of the problems that we've identified in manufacturing.

48:27That's one jumping off point. More generally, you want to learn about ROS, you go to ross.org. You can find all the documentation, the tutorials, download the software, go get an open source neural network model that does, you know, perception, train it in simulation, go get one of the low cost arms that you can even, or buy one or 3D print one at home. Lead robot has a good model and you can start playing with this stuff yourself and just, just get familiar with it. Even if you don't, you don't have a particular application in mind, just, just learn about the technology and see what it sparks in you.

49:00Grant:I love it. I think that's so cool. Thank you so much for your time. All right. Thanks, Grant. And thank you everyone for watching. If you haven't yet, make sure to check out the Neuron.ai and the NeuronDaily.com, where you can catch our newsletter. It goes out to 700 ,000 people every day. Farewell for now, humans.