Zachary Jackowski on Generalization in Robotics and the Reality of Deploying Robots in the Real World

Transcript

[00:00:00] Zachary Jackowski:

One of the things I enjoy most about our team here is we understand designing robots well enough to understand when we want to conform to a natural form or an industry standard, and we're totally comfortable with just going off and doing our own thing. We actually understand the fundamental constraints on the engineering, the fundamental physics here. And the team is just completely unbounded in going in a new direction because there's so much understanding and confidence about the fundamentals of the design problem.

[00:00:43] Brian Heater:

Hello, and welcome to another episode of Automated. My name is Brian Heater. I am the managing editor at the Association for Advancing Automation. And boy have we got a good episode for you. This week we're speaking with the head of Boston Dynamics' Atlas program, Zachary Jackowski. I learned a lot about the company's approach to humanoids from the chat and how Boston Dynamics is approaching research platforms after shifting much of its efforts towards commercialization. If you like the show, don't forget to like and subscribe. Please check out our newsletter over at Automated.fm. Also, Automated is coming to Boston. We're taking our first ever road trip in a few weeks. We are recording a bunch of episodes in person, including some in front of a live audience. We are also hosting a happy hour from 4 to 6:00 PM at MassRobotics on April 8th. And thank you to Maxon for sponsoring that event. And with that, please enjoy this conversation with Zachary Jackowski from Boston Dynamics.

I figured a good place to start would be to reflect on Robert Playter's time at Boston Dynamics. Obviously, the history of Boston Dynamics is Robert Playter's time at Boston Dynamics, but the big role that he has played as CEO is making that transition into commercialization. That's essentially what he took the throne from Marc to do over the last few years.

[00:02:11] Zachary Jackowski:

It's still kind of sinking in a little bit that we're gonna have to figure out how to do this whole thing without Rob by our side at every step. It was Rob almost 15 years ago now who gave me my offer letter to Boston Dynamics. I've been working directly on his team for such a long time now. It's hard to express how much of what you've seen about how the company has excelled both on the product side and the commercial side is due to Rob's inexhaustible energy and perseverance - the number of major corporate events, moving from an independent company to Google, to SoftBank, to Hyundai. The fact that those all went well, it was not a thing that just happened. That was Rob's perseverance, just working day in and day out to make those transitions work and negotiating all the details. Because of that work, we are very much on our feet now - the robotics product design research organization - and on the commercial side, things are just cranking along. That's what gave him the comfort to say, this is my opportunity. You guys take it from here.

[00:03:26] Brian Heater:

I was matching up the timeline. You started in 2011, coming on from really pure research - you were over at MIT CSAIL. And then two years later, Google buys Boston Dynamics, and then four years later SoftBank, and now it's Hyundai. So obviously the company has been through a lot in the 15 years since you've been there. Your first two years at the company must have been very interesting, really undergoing that period of being purely research and, in that sense, to a certain degree, very similar to what you were doing over at CSAIL under Russ Tedrake. To then going to this big corporation under Google, an investment bank under SoftBank, and now a commercial arm.

[00:04:16] Zachary Jackowski:

There are a few things to tease apart there, though. I've always been pretty entrepreneurial and product-oriented at heart. Even during undergrad and grad school at MIT, I had a startup at the time, so I've always had a passion for doing robotics, but also a passion for doing product. At Boston Dynamics over the first four years that I was here, the big thing that was happening was it was becoming increasingly clear that the research we were working on was working. That if we wanted to push it to the next level and do bigger and better things and more impactful research, it was clear that the next thing we had to do in the pursuit of the frontier of robotics was go into product and be more commercial.

If you look at it from the perspective of the bunch of people at Boston Dynamics who came here to work on the forefront of robotics and to make an impact in our work, it makes sense that that transition was not as big as you would've thought. Everyone's excited about being on the forefront and doing bigger and better things, and everyone was mentally acclimated to the idea that the blue sky research part - we kind of tapped that out. It's time to figure out how to go bigger and better.

Now, obviously, the transition from a 65-person, fully independent private company to learning how to live at Google was interesting, but on the other hand, we've kind of always been Boston Dynamics, and we are to this day. The company itself has a very supportive, very tight-knit engineering culture and now commercial and product culture. It's a bigger family than it was, but it's still very much a family here. Everyone is now in a few buildings, but it's a cohesive whole, and it's always felt like a cohesive whole. We learned how to be a cohesive whole inside Google. We learned how to work with SoftBank, and now we're learning how to do the same thing with Hyundai - where it's still Boston Dynamics. We all have Boston Dynamics badges, everything that goes along with that. But there is this big entity that knows how to do things that we don't know how to do, and has access to resources that we don't have on our own. It's still comfortable. It's still Boston Dynamics and it's still working.

[00:06:45] Brian Heater:

That's really interesting, because one of the things that you and I spoke about the last time we spoke - maybe a month or two back - was that the original version of the electric Atlas that we saw, I think R1 is what you were calling it in-house. That it was designed to be a research robot. And in that sense it was similar to the original hydraulic Atlas or the earlier robots that we saw. And in so far as Boston Dynamics does continue to be a research lab and does do these boundary-pushing robots, is there still somewhat of a distinction between the research side and the commercial side?

[00:07:19] Zachary Jackowski:

Up until last year, I would've said yes - that each of our products follows this arc where it's first in research, then the research is kind of working, and then we go full commercialization. We've got Spot, Stretch, and Atlas. Spot was first and it was on that arc, and then Stretch started and it was on that arc, and now Atlas is on that arc. Even though we are focused on product and we're driving there really hard and we're designing this organization like a commercial entity, it is still very much research. It's just now it's all centered around VLA design and agentic VLM behavior and how we're gonna do dexterous manipulation. It's part of a bigger product story, but it's still very much research. The size of the research team on Atlas is as big as Boston Dynamics in its entirety was when I joined the company.

[00:08:14] Brian Heater:

This was something I was trying to figure out. I was looking at the video of the R1 doing the flip and everything else. There was something about the title that really made it feel like this was kind of the R1's last hurrah. Is that right? Are you effectively retiring the research version of the robot, or will it still be around in some form to continue to push that side of things?

[00:08:53] Zachary Jackowski:

It's really done. We got back from CES, we had a bunch of working robots from CES, and our team did a big exhale, letting out all that stress and pressure, and then we decided, hey, let's go do some really fun stuff with these robots for a couple weeks. But behind that is the new version of the robot, which is so much better on every axis. So yes, the R1 robots look a little bit more human and they have a couple specs where they have some really special performance, but the new robots are so much more reliable, predictable, and repairable. The sensing and compute aspects of them make so much more sense. We were ready to let the R1 robots go. We're making static display models. We've got one up in our museum now. We're putting them away. They're on a code branch. There are a couple people still using one, but that branch is getting increasingly divergent from trunk.

[00:09:56] Brian Heater:

I got the sense, the last time we spoke, that the idea behind the R1 and the idea behind starting as a research robot first was: let's build a robot where price isn't a factor, weight isn't a factor, certain safety concerns around working alongside people aren't a factor in the same way that they will be when we deploy these things. Therefore, this is gonna be bleeding edge, state of the art. Now we're moving towards productization - weight, pricing, repairability, scalability are all concerns. Don't you necessarily have to make some compromises and lose some of that functionality in the process?

[00:10:33] Zachary Jackowski:

It's a little bit more complex than that. If you launched from day one and just said, oh, we want to build the whole humanoid robot product enchilada, there are so many decisions and technical threads to pull on, and information you fundamentally do not have yet about what makes the robot work well, that you just get bound up. So part of the purpose of making a research robot is just an artificial simplification of the problem so that you can go gather the information necessary and design the whole thing.

But the second aspect is, now that we designed that R1 robot and we understand how all the pieces fit together - when we built that robot, we didn't yet understand how all the pieces fit together precisely. So yes, you can spend a lot of money and buy really nice materials and kind of push the limits of performance in a lot of ways. But in a lot of ways, when you build a first-generation robot like that, it's an unbalanced machine. The fact that you went to the extremes on some parts means they don't all play together perfectly well. And so when you go and do a second-generation design, you now know: if I make this a little bit weaker, if I make this a little bit heavier - I didn't need that whole thing.

So the R1 robot has three computers in it - two x86 computers and a Jetson in the head. The new robot has a single NVIDIA Thor, and that's an enormous simplification and actually an enormous jump in compute power. We get that in a lot of places in the robot. So not just the compute, but the actuator designs. We know so much more about how to build super performant actuators, and we actually know how to build super performant, low-cost actuators, because we figured some things out that let us have our cake and eat it too. When you design a product, it's not always just making it worse to make it cost less. There are plenty of places where you can legitimately be smarter or use information that you didn't have the first time to make it fundamentally better in every axis.

[00:12:55] Brian Heater:

If I'm remembering correctly, was it two actuators on the entire body, and three if you count the hand?

[00:13:01] Zachary Jackowski:

Yeah, that's exactly right.

[00:13:03] Brian Heater:

The thing that really impressed everybody in the first video of the R1 was it standing up and being able to change direction - that's a sign of incredibly powerful actuators, that it could flip on the spot. Were you able to maintain that kind of power and adaptability while paring it down to this many actuators?

[00:13:22] Zachary Jackowski:

Yes. In short, the actuators are phenomenal. In fact, it's actually the opposite direction of that. The one piece about the production version of the robot that leaves us a little queasy still about the two-actuator decision is these giant elbows. There are actually more places on the robot where it has excess strength and excess power capability than places where it's actually underpowered. You saw the specs on how much the production robot can lift. In some ways, that's a consequence of our simplification work - we got so good at making these actuators really strong that we were able to simplify the robot down to mostly the stronger version of the actuators.

[00:14:08] Brian Heater:

You pointed out too that you were surprised by the reaction at CES - you thought people's reaction would be that it was more of a radical departure design-wise. You pointed to the legs, because the legs aren't a really big design change. Why necessarily did you have to make such a big design change to the legs?

[00:14:36] Zachary Jackowski:

The biggest driver for that design is actually making the robot safer. One of our top goals with the production version of the robot was to drive out as many pinch points as possible. If you look at most robots, you can find a pinch point behind the knee - literally somewhere a finger can get caught. Some of the worst examples even have exposed linkages, but anywhere that you can get a finger in and crunch someone, we really don't want that. So that pushed us down this path of trying to make as many of the axes on the robot offset axes as possible. That's a kind of a common design pattern in robotics - you look around and you'll see it repeated in most industrial cobot arms and such. Second, maximizing the range of motion capabilities - you want these offset joints, otherwise your joint interferes with itself.

Those are the two main drivers. One of the things I enjoy most about our team here is we understand designing robots well enough to understand when we want to conform to a natural form or an industry standard, and we're totally comfortable with just going off and doing our own thing. We actually understand the fundamental constraints on the engineering, the fundamental physics here. And the team is just completely unbounded in going in a new direction because there's so much understanding and confidence about the fundamentals of the design problem.

[00:16:11] Brian Heater:

We know what a lot of the pros of the humanoid form factor are. We know what some of the drawbacks are, but the way I've started framing it recently is that I'm no longer thinking about the humanoid form factor as being an endpoint for robot design, but rather a starting point.

[00:16:31] Zachary Jackowski:

That pretty much mirrors our philosophy too. There are lots of places where we can see making wheeled platforms with arms, fixed arms, forearm robots, all sorts of things. But we do have to pick a thing to do first. In our judgment, it makes the most sense to do the humanoid first because it has such compelling generality. And I really think that the theme of modern robotics that everyone's discovering is the importance of generality. We're kind of being taught that by the success of LLMs and VLMs. It's cool that these things appear kind of intelligent, but the big revelation is, hey, we've been building specialized software tools. And once we got shown a way to build a generalist tool that can do these specialized things, the lid just got blown off software engineering and software in general. We're seeing the same thing in robotics - the magic is the generality, and the magic is it's one instantiation that can do everything.

[00:17:41] Brian Heater:

I mentioned your work at CSAIL before. I think the work you were doing alongside Russ - I know this is something that he had been teaching for a long time - and I'm wondering if this dovetails, because I'm not super familiar with Underactuated Robotics, but is that something that kind of informs the direction you've been going in with Atlas?

[00:18:04] Zachary Jackowski:

Underactuated Robotics is really just the term for the field of control system engineering where you don't have direct control over every degree of freedom in the system. So like an industrial robot that's bolted to the ground, you can apply direct torque control to every portion of that robot's state vector. But with a balancing robot, a walking robot, you can control the joints of the robot, but you don't have direct control over the base link of the robot - your body is not attached to another arm holding it to the ground. You have to influence the evolution of the dynamics of the robot through how you interact with the world. A foot can only exert so much torque on the world. So that's the definition of the underactuated term and how it relates to robotics.

The things we were working on at CSAIL back in the day, we were focused on the problems of balance and manipulation and how you design control systems for robotics problems where you don't have full control. That's kinda where we are today, so of course that thinking has influenced where we are. When I started my grad school work - back in 2008 or 2009, what folks would call the ice age of machine learning - Russ was one of the true believers still in reinforcement learning. He always had that saying that birds aren't learning how to solve the Stokes equations - one of the equations of motion for fluid flow. Model-free learning. So that's what we were working on. At the lab we spent most of the time working on model-based control methods and optimization-based control, normal robotics stuff. Finally, in the late 2010s, the world kind of snapped back to the original design philosophy of that lab and what we were all working on initially.

[00:20:10] Brian Heater:

Just yesterday, as a matter of fact, we had a great conversation with Sergey Levine - he's at Berkeley, but also Physical Intelligence. We were talking a lot about data collection. I use the phrase "brute force" to describe this idea of just getting robots out in the world to collect as much data as possible. It strikes me now that if that's the method we're going to use, if that's really the way to create generalized robots, then we're in this liminal space right now. It's a chicken and the egg. How do we get robots that are generalized enough that we can get them out in the world to collect the data we need? I assume the approach - and I do see Boston Dynamics and other companies taking this - is sort of baby steps, right? You get it out there, you get it doing one job at a time, and then you start taking on more work.

[00:21:05] Zachary Jackowski:

That's a correct characterization of the situation. I haven't heard it called liminal space yet - I'd say that's pretty accurate. There is a really important thing that isn't super widely appreciated about the data, though, which is we all have this hope and a fair expectation at this point that you get your data together and you're going to see generalization of behavior capability. But I can tell you it's pretty firmly known now that if you get out there and you do all of one thing - like we're showing the automotive part sequencing behavior - if we deploy robots and just do automotive part sequencing, we are not going to produce the kind of generalization that we're hoping for.

[00:21:53] Brian Heater:

You mean, like, you do automotive part sequencing and it's not gonna be able to do my dishes. Something like that.

[00:21:59] Zachary Jackowski:

Exactly. Quality and diversity of the data is every bit as important as the quantity of the data. It's not as simple as, oh, just get good at one thing and get out there and collect lots of data. The smart folks are constantly pushing a level of discomfort about doing new things - new kinds of actions, new environments - rather than just kind of hill climbing up, getting really, really good at this one specific thing.

[00:22:31] Brian Heater:

I'm kind of curious to get your approach to generalization from a research standpoint. There are companies out there that are saying we're going after the home first. There are companies that are starting to experiment with the home. There are companies that can see use cases beyond industrial on the horizon. Generalization is something you'd love to have in a general purpose robot. So what does that mean - does that mean you look at the home, does that mean partnering and seeing how Atlas would react in a different environment?

[00:23:13] Zachary Jackowski:

There are a few things buried in there. Even in industrial and commercial environments, you need generalization - even just limiting yourself to working in a car plant. One of our big messages: we're part of Hyundai, and Hyundai is so beautifully good at designing car plants. If you go to a Hyundai car plant, there is a very good reason why every task in that plant that isn't automated is not automated. They know this stuff backwards and forward, and the reason is that there is such a diversity of behaviors around all of those things that aren't automated yet. And so you need that generalization to do that stuff. There's plenty of diversity to engage with just in that context.

What you do get with that context is some regularity. So there's a huge diversity, but you get to engage with that diversity in a repeated form. And you get to work around trained people - trained adults who understand how to work around a piece of equipment. You don't have those two things in the home, and I think that's going to make the home obviously a place that we are going to get to, but it's a little bit more complex to engage with if you're trying to build something that's commercially successful pretty early.

We're highly interested in home applications. We're also just highly interested in getting our commercial juices flowing and deploying robots to customers that genuinely see a return on investment on deploying them. That's what we're gonna get with the industrial environments. The opportunity in commercial service and home markets is huge, and of course we will be playing in that area.

[00:25:11] Brian Heater:

Maybe this is a better way to frame it. We spoke about the research part - you said that's a big part of what the productized version of Atlas is being used for now. Outside of selling to a Hyundai or deploying these systems in the world for commercial reasons, does the current production Atlas have a role to play in research being done outside of that environment?

[00:25:42] Zachary Jackowski:

Oh yeah, we'll definitely be doing our first experiments in commercial and service robotics with the production version of Atlas that you see today. I'm sure we'll spend some time in home environments too - making beds, folding laundry, cleaning countertops and all that stuff. It's an amazing platform. It's amazingly capable. Probably a little bit too capable for a home environment, but that's a great thing if you're going and doing research.

[00:26:07] Brian Heater:

What does it mean to be too capable?

[00:26:09] Zachary Jackowski:

I think the robot is slightly heavier than we would want for a home environment. In these industrial applications, you have the luxury of having the people around you be a known thing. You are around able-bodied adults who are trained on interacting with your robot. In a home environment you don't have that - you've got kids running around, you've got pets. And I think that's gonna lend itself to a pretty different kind of platform once you get to mass scale. But again, great for experimenting and research.

[00:26:42] Brian Heater:

It sounds like a lot of these things just come down to bandwidth, right? We're using words like generalization, what do we prioritize, what can we take on? You have a lot of resources. Obviously Hyundai has a lot of resources, but you still need to prioritize things. So what do you put first? Just because Atlas has legs doesn't mean you believe there aren't gonna be applications where a humanoid with wheels might not make sense.

[00:27:07] Zachary Jackowski:

That's completely fair. This is really about focus. No matter how many resources you have to apply - whether that's your number of people or number of dollars for a project like this - the fundamental constraint you have is how much focus your researchers and leaders can apply to directing those resources. You fundamentally have to have people at the top that are paying attention to the details. That's kinda where you run out of bandwidth.

[00:27:37] Brian Heater:

One of the things I'm curious about - you obviously spent quite a bit of time on the Spot team. I think you were heading up the Spot team as well. One of the innovations there, especially towards the end, was this idea of turning Spot into a developer platform - like let's do the iPhone, Android version of a robot, where we don't have to develop all the applications. Is that something that continues to make sense even as we're talking about VLAs and VLMs? Opening up these applications to third parties?

[00:28:08] Zachary Jackowski:

Spot actually launched as a developer platform, and the evolution of it was us figuring out that just throwing a platform out there isn't enough to make a product successful. It turns out that that's really hard work. And if there's one thing that other people aren't going to do for you, it's really hard work. So yeah, we put the developer platform out there. It was exciting. We had great engagement from a lot of partners, but fundamentally we needed a team of very focused, very well-resourced roboticists in order to get promising applications over the line in terms of feature completeness and reliability. We kept the developer platform aspect, but we picked some primary applications that we made successful ourselves.

With the humanoids, we're gonna cut to the chase a little bit. Having done that story with Spot, we're picking manufacturing to be our successful application. We will certainly engage partner companies and have research efforts across the board - all of those will benefit from platform aspects, and we'll do those. And you're totally right that the evolution of the technology stack toward VLM-based systems that are fundamentally prompted rather than painstakingly coded makes that a lot easier. It does push the work of making that system reliable back onto the training side now. So just because it's easier to interact with that system and fine-tune it by prompting it differently for different situations doesn't get around the fact that when you find real problems with a system, you need to understand them, root cause them, and go retrain the system for proper behavior. So that's still gonna be a limitation.

[00:29:59] Brian Heater:

It sounds like for you and the team, to a certain extent, you were really able to hit the ground running with the Atlas project because of the work that you did on Spot.

[00:30:08] Zachary Jackowski:

Exactly.

[00:30:09] Brian Heater:

When did you personally make the transition between teams?

[00:30:13] Zachary Jackowski:

I'm kind of losing track of time. Maybe about two years ago.

[00:30:20] Brian Heater:

Okay, so fairly recently. When did work on the R1 start?

[00:30:24] Zachary Jackowski:

It depends on what you call the iteration. The electric version started about two and a half years before I started on Atlas.

[00:30:33] Brian Heater:

It's funny - I was also looking at the timeline from this standpoint. So 2011, and then I guess officially hydraulic Atlas is announced about two years after you had started, right?

[00:30:46] Zachary Jackowski:

Yes.

[00:30:46] Brian Heater:

Was that a difficult transition at all - from the Spot team to the Atlas team? How dramatic of a difference is that?

[00:30:53] Zachary Jackowski:

Like I mentioned right at the start about how Boston Dynamics feels - the Spot team, the Atlas team, and the Stretch team are all literally down the hall from each other. We all have lunch together and we all build on a common tech stack. And so even though the teams operate relatively independently so that they can choose the fastest path to success, fundamentally they're all pretty similar. Moving between them is pretty fluid. All of them are very familiar spaces to folks inside Boston Dynamics.

[00:31:25] Brian Heater:

Stretch doesn't get enough love.

[00:31:27] Zachary Jackowski:

No, it doesn't.

[00:31:28] Brian Heater:

What are some of the learnings on that side that have informed the way Boston Dynamics approaches robots?

[00:31:34] Zachary Jackowski:

One of the lessons is if you want to move boxes at super high rates and super predictably, even for the company that pioneered legged robotics, sometimes wheels are better. The biggest thing we've learned from Stretch is that even within a problem like logistics, and even within truck unload, every time you look one level deeper in a problem, there's always an explosion of complexity. You might think that truck unload is a straightforward problem to solve, but it is its own kind of crazy kaleidoscope of complexity that they have to engage with. Not appreciating that is definitely a way to unfulfilled expectations. Thankfully we figured that out, and we've figured out how to go really deep in truck unload.

[00:32:28] Brian Heater:

I talk to a lot of physical AI companies, and there's a lot of conversation now around generalized models versus platform-specific models. Companies like Physical Intelligence are building across embodiments. Maybe Boston Dynamics is heading in that direction too. I see the benefit in sharing a platform and collaborating together. But beyond that, what is the benefit of having three very distinct embodiments all working on the same platform?

[00:33:03] Zachary Jackowski:

The benefit of having three different platforms is when you get back to first principles about how those platforms interact with the world - there are things that each one is fundamentally better at. Just like in the natural world, quadrupeds versus bipeds: there is clearly a giant niche for quadrupeds in the natural world, because fundamentally they have this incredible capability to place their feet where necessary to maintain balance. A biped, with its heavy legs and only two of them, will never be as good or as fast as a quadruped in maintaining stability in tough environments. So for a data collection platform - which is mostly what Spot is - you just can't beat four legs. For Stretch, similarly, you go back to fundamental physical principles: you're just not gonna move boxes faster than a machine built like Stretch.

Back to your question - will those machines increasingly be built off the same learned backbones? Definitely.

[00:34:10] Brian Heater:

What specifically is the benefit to that?

[00:34:12] Zachary Jackowski:

This is a place where we're just pushing the frontier in research. VLMs are kind of lighting the way for the industry, but it's all super early for the robotics stuff. The expectation is you should see stronger generalization and stronger reasoning capabilities through the diversity of the data and the training. It's just kind of a rule that more diversity is better.

[00:34:40] Brian Heater:

Like effectively robots can potentially do more based on the data that they're fed?

[00:34:47] Zachary Jackowski:

Yes. To use your example - we haven't done this research yet, but I would expect that an Atlas that has a VLA trained to do sequencing, if you bring VLA data from Stretch for truck unload in, you should see a performance improvement from the introduction of that new data diversity. Maybe not a huge improvement, but it should be an improvement and not a loss.

[00:35:16] Brian Heater:

Now this is really obvious as I'm thinking about it, but in effect, having Spot out there doing security or inspections in the factory, and having Stretch doing truck unloading - all of those things are ultimately collecting data that will go to inform and make Atlas a better robot.

[00:35:41] Zachary Jackowski:

There are a lot of details to get right about exactly what data we're out there collecting, how we get permission to use that data, and the relevance of the data. Everyone kinda understands this in the industry now, but being good stewards of your customer's information is actually one of the hardest parts about this whole endeavor. We go to really great lengths to make sure that our customers understand what the robots are collecting and how that data is being used, to make sure we're really partners in that. But it does make things slower.

[00:36:17] Brian Heater:

Here's a place where it makes sense to be owned by one of the world's largest car manufacturers.

[00:36:24] Zachary Jackowski:

Yes. Unequivocally, yes.

[00:36:28] Brian Heater:

Great. Well, Zach, thank you so much for taking the time.

[00:36:32] Zachary Jackowski:

Thank you. Great questions.

[00:36:35] Brian Heater:

All right. That was a great chat. Thank you to Zachary and to Mia and Carrie at Boston Dynamics for helping set that up. Thanks to you for tuning in. If you enjoyed the conversation, please support the show by liking, subscribing and maybe even leaving a rating. And don't forget to check out our newsletter over at Automated.fm. Thanks again, and we will see you soon for another episode of Automated.