Boston Dynamics： 1 on 1 with Atlas director of robot behavior

If you look at this Atlas compared to the previous generation, there's almost like an order of magnitude reduction in complexity. It's much that has way, way less parts. Wow. Way less like unique parts. And the process of manufacturing it is like much faster and simpler, which leads to higher reliability and lower cost. So many things are happening right now in humanoids in general, but Boston Dynamics as well. Hyundai obviously took full control. You've released significant capabilities. I mean, arguably your robot is the most capable humanoid on the planet right now. And I say arguably because it's it's really hard to tell. Specs are not all out there. Demo videos are great. But I mean, there's some really impressive things you're doing. It's probably expensive still right now. But tell me kind of state of the union where you guys are. Yeah, definitely. That's along along the lines of how we feel that we've really brought up a new generation of Atlas that sets us sort of at the forefront of capability. It's interesting when we mention capability because it's really the combination of hardware and software. So when you bring up a new robot, you have in your head sort of capability and potential that you've designed it for. But it's not really until you figure out how to get it to do those things that you really like verify. And we're really happy that we've been able to do that quite fast after designing this new robot. It is amazing, right? I mean, you want the hardware to be able to do the sort of tasks that you want the robot to be able to do. But without the A.I., without the software, the physical A.I. behind it, it's kind of useless. I'm seeing some robot manufacturers overengineer in some sense, if that's possible. And humanoids, what the hardware is that they're initially shipping and then they're going to continually build the A.I. and software update, you know, like modern products. Yeah, absolutely. And in many respects, I would say that the things we can show with existing hardware are still limited by our ability to know how to control it. Right. I think that's definitely a plus. But I'm sure it's the same for many other robots that they're capable of much more than we've been able to squeeze performance out of them still today. So I would say that the A.I. capabilities and the control algorithms are still one of the main bottlenecks on getting value out of hardware. Maybe just to highlight one thing on the hardware side that we're very excited about is that in the design of this new generation of Atlas, we've been very explicit about focusing on simplicity and reliability. Right. So if you look at this Atlas compared to the previous generation, there's almost like an order of magnitude reduction in complexity. It's much that has way, way less parts. Wow. Way less like unique parts. And the process of manufacturing it is like much faster and simpler, which leads to higher reliability and lower cost. So we're happy that we've been able to demonstrate like the same level of performance or higher with a robot that is fundamentally way, way simpler, which we think puts us in a really good place for sort of the next step of mass manufacturing. Order of magnitude. That's that's a big statement. I know you said almost, so you're not saying it's a full order of magnitude. But wow, that's super impressive. And that's obviously a critical step towards, as you say, manufacturability, cost, so you can price it appropriately, all that stuff. Talk about the A.I. side a little bit and the physical A.I. side. It's a good time to talk about it. Abtronic just released the news this morning. I just literally published a story on Robot Park's 90,000 square foot facility. They're doing several around the world as well. And it's a data factory for robotics. We're going to have robots doing many, many tasks in many, many areas and many different things and learn from that. They obviously have a partnership with Google DeepMind as well. Talk about your A.I., your foundation models. How are you training Atlas? So you can, I like thinking of the A.I. layer that controls the robot as like a two system, right? So on one hand, you have what I call like physical intelligence. It's sort of the core component of the controlled algorithm that worries about things like balance and worries about things like physical skill and agility, like jumping, grabbing things and moving them with speed and generality. And then there's the reasoning intelligence. It's the thing that you look around and you understand, OK, to do this task, first I have to do step number one, then step number two, then step number three. And the way that this thing looks seems like it's going to be heavy or it's going to be lightweight or I'm going to have to push this way. So we are investing on both sides. I think that we are uniquely positioned to have as a core strength competency on the physical intelligence part of the behavior stack, right? Like for many years, we've sort of demonstrated how we can push forward the frontier of what these robots can do. And in the past with we've been inspired by areas of like human agile skill, right, like you name it, it could be dancing or it could be parkour or it could be gymnastics. I guess now we've done it sort of in the context of soccer and using sort of the World Cup as motivation, which has always forced us to sort of have a very strong grasp on how to squeeze physical strength and agility out of the robots that we build. And we think that that's clearly a competitive strength we have. Over the last couple of years, we've been investing more on the second kind of intelligence, sort of the reasoning kind of intelligence, because it's also essential if you want to get to the point where you get these platforms to be to have some degree of like ease of retasking or generality in the capabilities that they have. If you want to deploy a robot in a factory and it has to do a certain job, but two weeks later, that job changes because the workflow changes or there's a new exception that it needs to handle. You want to make sure that the robot can adapt to that without having to invest another few months of effort in validating and sort of creating the new programming. You want the robot to learn through either experience or through demonstration in a way that is more natural. So we've been heavily investing in that second kind of intelligence the last couple of years. That's super interesting because that's like the human crawlery, right? You get a new job and somebody, your supervisor or a co-worker, says this is what you need to do and this is how you do it. And if your robot can learn like that on the fly and not spend two months learning the new job or two weeks, sometimes those things get a little faster. That's generally speaking, a good thing. Are you also working on a layer of intelligence for interoperability? So I was just at Humanoid.ai's offices in London and they've got like three, four levels of intelligence. One of them is sort of a, you know, an agentic overseer. Others like Tudor Intelligence connect into a warehouse management system and they get sort of tasks from above. And then there's even layers where you think about robots working together on a task like, you know, nobody told them to, but this is heavy and it needs help. And it gets what about those layers of intelligence? Yeah, definitely. There's certain things that we think are essential. For example, being able to understand and communicate with factory manufacturing systems, right? So a big cost today when you're deploying new technology in a factory has to do with integration, right? Like you have to get your system to speak factory language and be able to communicate with the management system that is orchestrating everything. So getting these robots to be able to speak that language and have that ability to either be commanded from a higher level system or being able to provide feedback. That's important. And that's one area where we actually have significant experience, right? Like from the other products that we have that are deployed with hundreds of customers, like Spot, for example, has a fleet management system where you tell Spot what to do, not by programming each individual Spot what it has to do, but rather by distributing sort of a set of inspection points in a factory. And then there's a coordination fleet management system that decides, well, I have to send this robot or this other robot and distributes the load of that. That's something that we've been doing for a while now with some of our other products. Do you have any thoughts around legs versus wheels? Jeff Gardinas from Aptronic, when I talked to him yesterday, he said, you know, we see wheels as doing most of the job, but legs have a higher ceiling. Thoughts on that? Yeah, lots of thoughts, of course, and many internal discussions. So we've actually tried both, right? So one of our products, Stretch, is a big base on wheels with a giant arm and then Atlas or Spot, they're legged. With Atlas, the humanoid factor, one of the main drivers for investing in legs is that, on one hand, actually the mechanical complexity that goes into building legs today is not that different from building sort of omnidirectional mobile platform. The number of wheels they usually put, it's four, and you end up putting two actuators per wheel because you want them to turn and also reorient. So you end up with eight actuators, which is almost the same as you have on two legs. The second thing is that legs give you, I guess, the obvious thing is that allow you to go to more places, right? So many factories, yes, they might look flat on first inspection, but then it turns out that, oh, you also want the machine to be able to cross the gap between your loading dock and your trailer, or there's some places where there's a mezzanine and you have to go upstairs, or there's some places where so on and so forth, that legs actually give you that extra reach that becomes necessary. The other important thing about legs is that allows you to have a more slim form factor, so you can squeeze in places that are narrow, which happen to be very frequent in factories or warehouses because space is a premium. So just as an example, if you have a mobile base and you don't know where your arm is going to look, you have to make the base big in all directions so that it can sustain the load when you're tilting your arm in all directions. If you have two legs, you can choose whether your legs are facing one way or the other way and be thin in the other direction so that you can squeeze in between places. So there's an accumulation of reasons why legs actually end up making sense. Maybe the last thing is that balance and locomotion used to be a difficult problem, but it's actually not that difficult anymore. We've figured out the right recipes for how to do it reliably. There are some humanoid startups that are still working on that problem, of course, but you've been doing it for a while, so that makes a ton of sense. Well, great to meet you guys and have a wonderful day. And yeah, I might have a few more things and we'll chat about that later. Sounds great. Thanks so much, John. Thanks, Sharon. Bye.