From Teach-and-Repeat to SelfPath AI: The next robotics leap

0:00Gene Demaitre: Hi, I'm Mike Oitzman, co-host of the Robot Report podcast. If your company wants to reach decision makers in robotics, automation, and AI, you're in the right place. Each week, thousands of industry professionals listen to this podcast for the latest news, expert insights, and conversations with leaders building the future of robotics. This spot could be yours. To learn more about podcast sponsorship opportunities, visit therobotreport.com slash podcast.

0:38Gene Demaitre: Welcome to episode 259 of the Robot Report podcast. I'm your host this week, Mike Oitzman, Senior Editor of the Robot Report. Thanks as always, folks, for tuning in. Today I'm joined by Gene Dimitri, Editorial Director for the Robot Report and Automated Warehouse. Gene, thanks for joining me today. Thanks for having me as always, Mike. and I'm glad to be here. Have you been following the world humanoid robot games in China? It's sort of the, every time I look at my feeds, it's the top news, even though there's a lot of other news that we'll get to in a moment. But the Chinese humanoid games seem to be capturing a lot of attention.

1:16Gene Demaitre: I think it's definitely reached their goal of filling everybody's feeds on LinkedIn and everywhere else. But man, it's been crazy. I think the funnest, most interesting thing out of all of that has been those sprinting robots, right? Literally falling apart while running down the track. It's been crazy. Crashing into the walls and catching on fire. Can you believe that? I mean, it's self-destruction mode, but it makes for great video. It sure does. It is a little bit like roller derby right now. It also reminds me of the DARPA robotics challenges of several years ago. And it shows how long a ways these systems still need to go before they're actually useful.

1:59I know there was a weightlifting robot that fell into the judge's table pretty spectacularly. Again, I think on the one hand, I draw different conclusions than a lot of the other headlines and news outlets that are out there. I think one, it does illustrate that China is absolutely serious about scale, is absolutely serious as a global competitor and producer of humanoid robots. But I also think we should not be wringing our hands and saying the end is nigh for humanity because these robots are pretty primitive still and entertaining, certainly. But, you know, I got a notice that Agivot won the most medals.

2:44So good for them. But we know that Unitree, Furrier, Beijing, Humanoid Innovation Center all developed competitors. And other companies to watch or other noteworthy participants included Limax Dynamics, Pudu Robotics, and Honor. So there's a lot of companies that are out there. And you and I have talked before about this. We're tracking more than 250 humanoid makers worldwide, many of whom are in China. So we shouldn't ignore what's going on in that sector and that part of the world. But I don't think we need to take this games too seriously at this stage. There's a lot of other news that I think is getting overshadowed, perhaps, by this.

3:32Gene Demaitre: Yeah, but at the same time, I mean, everyone loves the competition. and as engineers, it is fun to sort of put the yardstick out there, compare what you're doing. And again, it's outside the lab, right? So a run, a lift, a jump, they're all things that we can measure. So in terms of being able to sort of measure progress, I celebrate that aspect of this, pulling the competition together and having real things that we can measure and discreetly compare things as opposed to, I don't know, how many picks can you do in an hour? whatever the other case may be. It is quite interesting that this is a great way to compare things.

4:13If these games become an annual thing, I'd love to see it become truly international. I'd love to see some of the best US-made robots compete and see how they do. I don't think they're necessarily going to be able to perform in quite the same way. They're more expensive, perhaps a little less capable on that side of things right now. but I do think that there's going to be progress year over year in that front and you're right that competition from an engineering point of view is good because it shows us what can be done or

4:45Gene Demaitre: what's possible and Gene you know I'm a big photographer and love the night sky tonight we're recording this on Thursday afternoon tonight's the lunar eclipse I'll be out there grabbing some images and the other big thing to celebrate that this weekend is the the launch of the Nancy Grace Roman space telescope that's happening on Sunday. I think it's like a$4 billion project. So a lot of, a lot of stake on Sunday with the launch of this. And you know, what's interesting, Gina, about the Nancy Grace space telescope. It's, it's an infrared telescope. Construction was completed a year ago in November of 2025.

5:25Gene Demaitre: It has a mirror that's the same size as Hubble. so this this telescope's that the the mirror is the same size as Hubble but it's um got far greater uh uh field of view so the actual eyeglasses that are on the lens that's on it is way wider and you know it's going to be able to take uh pictures you know in an hour what it would have taken uh Hubble uh multiple months to to acquire that same amount of space because of its field of view and still has the higher resolution. And that's thankful now to the higher resolution equipment that we can deploy on there. And so this is really going to go out and look for dark energy and all of that and prove some of those physics questions that are at stake.

6:14Gene Demaitre: And I'm just thrilled that we'll be able to see this in our lifetime. So let's pray that everything goes safely on Sunday, get it into space. In six months, we'll come back and talk about some of the great science that this thing is doing. Yeah, fingers crossed. I'm also a big fan of aerospace and further exploration. And I can't wait to see what we find out once it's in operation. Now, we have a great show for you. Our guest today is John Black. He's CTO of BrainCorp. We haven't caught up with BrainCorp for a while. So this conversation you and I had a couple of weeks back with John, and we'll hear that interview a little bit later in the show.

6:52Gene Demaitre: So stay tuned for that after the news. Before we do that, Gene, we're going to talk about the RoboBusiness event that's coming up almost a month away. We're about still about six weeks, but it won't be long before it's a month away and then it'll be here. It's taking place October 20th and 21st in Santa Clara, California. This is the sister show to the Robotics Summit in Boston. Doing it out here in my neck of the woods. And, Gene, the neat thing about this, it's the 20th anniversary show. You and I and Steve have been involved with this event for various incarnations over the last decade plus.

7:29Gene Demaitre: And so it's going to be fun to celebrate, you know, the 20th anniversary. We've got some great sessions planned now. And that agenda is all up online on the website. So if you haven't checked it out, go check out the agenda. A lot of great speakers. We plan on having over 100 exhibitors. The show floor is filling up, but there's still some space. So if anyone listening wants to get out and reach this audience, there's still a chance to exhibit. We're planning on over 2 ,000 attendees. So that'll be fun. By the way, I think also Monday Night Football is happening just down the street that Monday night.

8:09Gene Demaitre: So if you're a big football fan and you want to come out and go to the 49er stadium and watch the Monday night football game, get out there by Monday night, it'll be just up the street from the event. It's going to be pretty crazy. We've also opened a call for participants in the Startup Alley. Gene, we've been already filling that up. So there's a few slots left. If you're a startup, it's a great opportunity to show your wear here at the event. Those are free, but you need to submit a participation entry, and you'll find that on the website as well. So, hey, we've got some great keynote sessions.

8:48Gene Demaitre: The Future of Robotics will feature Rodney Brooks, Ken Goldberg, Jonathan Hurst. We've got another AI keynote panel that'll have Andrew Davidson, Chief AI Officer from Universal Robots, and Michael Boglesong, Head of AI Foundation Models at Collaborative Robotics at Amazon. That'll be great. And Gene, I have to blow my horn here. I'm moderating a panel on the state of RAS, the RAS business. And I've landed one of the best experts in the industry. So everybody's going to try and come to this. I hope it's Rick Falk from Locus will be on the panel along with some other folks. We're going to dive deep into why Locus is so successful.

9:29Gene Demaitre: I think it'll be a great education on how to do RAS right. And we've got some other great panelists on that session as well. Last thing I'll say is it's still time to save some money with our early bird discount. Early bird pricing ends on Monday, August 31st. So if you're interested in going, go check out robobusiness.com. Now, just a reminder, we've got all the latest news and information on our network or website that include the robot report and automated warehouse. Let's get on to the news. So Gene, just a quick update on Unitree. You know, it's lost over, you know, half of its market value from what it closed at on that first day.

10:09Gene Demaitre: But I think that's pretty par for the course for a lot of the hype that happens around many startup IPOs as popular as this was. I think people are still ahead if they bought it, were able to buy it earlier in the day on that opening day. But, you know, I think the real story isn't that, you know, the stock price. It's what's happening since this, since the timeline. I mean, lots of visibility on Unitree. I think there's still some value there. And, you know, I think that the founder, Wang Jingjing, you know, talked about they're still trying to define, you know, that chat GPT moment for humanoids.

10:47Gene Demaitre: And I think you and I and Steve and the rest of the crew are still looking for that to happen with humanoids as well. And so it'll be interesting to see what happens here. I think we're really talking about something we ask a lot of humanoid companies, you know, when are we going to be able to tell the robot to clean up the room and have it listen to us and actually do that, you know, quickly and efficiently as well as other tasks in a zero shot. And I think that's going to be the real defining moment for this class of robots. And Mike, I agree with you that the drop in value is not necessarily a sign of weakness.

11:24is probably the fact that there was that irrational exuberance of investors during Unitree's IPO. And we've talked for a while about is there a humanoid bubble that remains to be seen. But we do know that even if it dropped by value by half, it still made money from where it started. And it's also still a sign of the level of interest in humanoid robots. I think it was a course correction, a natural market correction. we'll have to see in the long term how things even out right now it's very spiky because of of the level of interest and the the size of some of the companies involved uh but as with all robotics funding it's one of those things that you have to take it with a grain of salt to aspirin uh because uh the long-term value is what we're looking for and and that goes back to what you were talking about your your session at rebel business is that's what we're really looking for.

12:20But on a similar note, one of the big items of news this week was Dacotex, which is the humanoid robot unit of XPeng Motors, raised$900 million, or the equivalent. And that's not a small amount of money. In fact, it is reportedly a record single private financing round for China's embodied AI sector. It is noteworthy that Xpeng is an electric vehicle maker and we've seen a similar phenomenon here in the West where a lot of big automotive companies are the ones investing or in some cases buying the humanoid robot developers. So we know that BMW is testing a number of robots including Agilent's Digit.

13:08We know that Hunai owns Boston Dynamics. which makes Atlas. And so I think it's sort of an extension of that phenomenon that we've already seen. Xpeng or Degotex's iron humanoid robot. The goal is to produce 1 ,000 humanoids per month by the end of the year. So even though we've been on this podcast quite skeptical about the claims that there'll be millions of humanoid robots in the coming year, there will certainly be thousands. Now, what those robots are doing, if they'll just be dancing and racing and greeting people in shopping malls, or will they actually be conducting useful work? That remains to be seen.

13:54But XPeng's unit has already some experience with quadruped and tractor robots and logistics, security, power inspection, and smart homes. So there's, again, I think some potential there. But we should also note that they just raised$900 million, but they had losses the previous two years. So, again, we're talking about, as we were with Unitree, it's going to be up and down for a little while. We don't really know whether manufacturing at scale and being able to raise a lot of money are going to translate into successful or viable commercial humanoid robots. So we're keeping an eye on that. We're certainly tracking it.

14:38Gene Demaitre: I think, you know, X-Bang is sort of equivalent to Tesla in some regards and Optimus, Tesla with Optimus. You know, one interesting thing about this story, Gene, that I found sort of odd about what you just talked about was the initial deployment is going to be in retail stores and industrial campuses. Not so much assembling cars, but greeting people in the dealership room, right? So, again, more on the hype side, trying to bring people in, trying to, you know, lower the barrier to entry for future robots in the same space yet, but not quite doing useful work. So, I think that's going to be interesting to follow with XPeng here.

15:21Gene Demaitre: Yeah, they'll have 1 ,000 robots a month. That's a crazy number in the year. It looks great on paper in terms of the total number of units manufactured, but what are they doing? And I think that'll be the interesting thing to follow. Now, we have another story that we're going to publish this week about Generalist. And Generalist, they put in another, bringing in another$200 million in funding, right, according to their SEC filing. And Generalist is a, you know, a Gen AI-ish robot trying to build foundation models, right? And so their latest model was released earlier this week, Gen 1.5, of its robot foundation model, where the company is talking about what we just talked about, learn a new task in seconds from single examples without gradient updates or fine tuning.

16:18Gene Demaitre: Again, I think this is the kind of yardstick that we're going to begin to measure these foundation models by is, you know, how quickly can they learn? Can they do, you know, zero shot learning and execution? That's going to be very interesting to watch now in the next 18 months, I would say, as these things mature. So that's interesting. Again, processing video input and trajectories and being able to learn from that and then repeat and rinse and repeat and execute against this. So it's quite an interesting story from the perspective that, again, it's got to continue to raise money to feed the model.

17:00And they had just raised$400 million not that long ago, a couple months ago. And so I think it is noteworthy. And you and I and the rest of our team have had a lot of conversations lately about building the right models for the AI and then be able to train the models in order for robots to act more independently and be able to learn new skills on the fly or do new behaviors in these dynamic environments or applications. I think that is going to be the main story for making heleroids and other types of robots more useful out in the world is as much the AI evolution and the money going into that.

17:41Another, speaking of AI evolution, another one of our big stories this past week has been NVIDIA announced Jetson Orin Nano 2, which is their entry-level computer for robotics for edge processing. And so what's interesting there is that NVIDIA has said that they're able to perform at the entry level what used to be their high end just a year ago, and with less power consumption, with more efficiency. and when we're talking about inference at the edge, that's really part of that secret sauce, right, about making the robots more independent, about making them easier to train, about enabling them to perceive things better.

18:31And so we're going to see the application of VLMs, VLAs, all of these models be more enabled by this. And so NVIDIA has already announced a couple of developers that it hasn't worked with that were already using Nano 2, including Wing, which makes delivery drones, Matic Robotics, which does horrible cleaning robots, and then a range of companies like Cotnyx and QCM Bobcat and Advantech. And it's interesting to see where they're taking this based technology and then what they're building on top of that. But the actual Jetson or Nano 2 won't be available to the general developer community until early next year.

19:21But I do think it's noteworthy that, and I was on their briefing call, I had asked Deepu Tal a question about energy conservation or power usage. And it is interesting that they are continuing to make strides in this, which is very ultimately going to enable robots to be both more performant, but also hopefully a little bit cheaper going forward.

19:47Gene Demaitre: Yeah. Of course, power consumption is key for any of these edge devices, Gene. I actually saw another story last week that Apple, the latest Apple workstations are going to be more powerful than you need some of the NVIDIA data center processors with way less power consumption and cooling and be able to run on your desktop. So, you know, the competition's heating up in terms of the architectures and the hardware layer for all of the AI compute going forward, whether it's plugged into the wall or whether it's plugged into the data center or whether it's plugged in on battery on a mobile robot of whatever form factor.

20:26Gene Demaitre: It's very interesting, all the options and opportunities for hardware developers to determine what their compute architecture looks like going forward. Yeah, I think it's a moving target. I think NVIDIA is going to produce this or make this available, and then there'll be a little bit of delay as the community starts building on it, and then we'll see, again, the next level of improvements. And you're also right to point out that while NVIDIA is the market leader, it is not the only game in town. All the other processor makers are also developing architectures to enable better use of power to allow more capable models to be run on them.

21:07And I'm excited for that arms race because I think it will help robotics developers across the board. Another item of news this week is Symphonic has taken number one in the warehouse automation market in the Americas, according to Interact Analysis. Now, there are a lot of qualifiers there, right? We're not saying across all types of robots. We're not saying across all markets. But it is interesting that Symbonic has become the number one, according to this one analyst firm. And I posted a Q &A on our automated warehouse site. And largely, it's driven by two things, according to the analyst firm.

21:57One is it has some large-scale projects with Walmart. And obviously, Amazon helped create this market by pulling Kiva into itself in 2012. And so Walmart, as one of the largest retailers in the country, them working with automation providers makes a big difference. But then the other thing that the analyst at Interact Analysis told me is that some of it is based on the focus. In the past, it was point solutions. And companies that focused only on one type of robot or only on one type of application or task, they grew, but then there seems to have been not necessarily a ceiling, but a limit to how fast they can grow.

22:48Whereas Symphonic has a working on end-to-end automation from induction through depalletizing to autonomous storage and retrieval to assembly pallets to sending them out. And I think that's noteworthy. Steve and I had the opportunity to go and visit a million square foot warehouse earlier this year. And we got to see all that in action. And it is really impressive because, yeah, you still need some people in the warehouse for certain tasks or to supervise. but to see all of these boxes of varying sizes moving around and being stored super efficiently and then raft and palletized, depalletized, all that is, you can understand why a company on the size of, on the scale of a Walmart would be like, yeah, we want that.

23:44It's not for everyone, but it certainly has helped Symbotic. And to its credit, Symbotic has continued to innovate, continued to acquire. They acquired an autonomous workflow provider, Fox Robotics, and then an intelligence provider, Arms Innovations. Just this week, they joined the Stanford Robotics Center Industrial Affiliates Program. So they're continuing to move forward, and they're diversifying, right? They're as important as Walmart is. you don't want to have all your eggs in one basket. They are making sure that they serve the wider industry. And we'll see how that continues to play out.

24:24But it is interesting to see how they're growing.

24:28Gene Demaitre: Yeah, I guess they're doing this right, Gene, as a support fulfillment arm for Walmart, right? For online ordering and delivery, that kind of piece, that's the basic use case here. And just a reminder, we've got you covered with all the latest news and information on our network of websites, including the Robot Report and Automated Warehouse. Great. Well, thanks for covering the news with me, Gene. Now it's time to talk about our interview with John Black, CTO of BrainCorp, who we talked to a few weeks ago. What are we going to hear? Well, for those of you who may not know, BrainCorp creates BrainOS, which is a software that helps operate autonomous mobile robots in a variety of settings.

25:15Most of us have probably seen BrainCorp or BrainOS-driven robots that are cleaning robots in large facilities. But Brain has been continuing to develop its platform. And John will talk to us about how the team there is using real-world customer feedback, fleet data, and new models like SelfPath AI to scale safely in public spaces.

25:47Gene Demaitre: Well, on the show, Gene, we've got John Black, CTO of BrainCorp. John, welcome to the podcast. And hey, great to have you here. Yeah. Hi, Mike. Hi, Gene. Thank you for the invitation. Look forward to speaking to you today. Yeah. So we're curious because it's been a while since we've caught up with BrainCorp. So you guys are busy growing all the time. And, you know, arguably you stepped into the CTO role after years leading engineering and strategy at BrainCorp. So let's start there. Tell us a bit about your transition to CTO and what's changed for you, you know, moving from building products every day, running projects to build engineering and deliver it to your role as CTO now.

26:32Yeah, it really comes along, I think, with the scaling. So I would say, you know, when I joined the company, it was very much about making the first product and commercializing, you know, one product, right? Like proving that we could do autonomous navigation in public spaces, really going from like research to commercial. You know, we've just crossed 50 ,000 robots now that we've got out operating in the world. You know, so I would say we are really, really looking at these next levels of scale now. And it makes you really have to think about the problem differently, right? So you're really looking at how do I manage whole fleets and not 50 ,000 robots in one building, but 50 ,000 robots in almost every facility and type of application that you can think about.

27:23Right. So that's kind of that's been the fun part, I would say, about stepping in the C2 role is really thinking about that roadmap. How do we really get platform effect? How do we really leverage all of that data to kind of accelerate and get to those next levels of scale?

27:43Gene Demaitre: Yeah, I know everybody listening would be jealous to have a 50 ,000 installed base. That's just an amazing number. And you guys have done that quietly. I don't think you celebrate that number enough of just how big of a milestone that is. There aren't very few companies that have that big of an installed base. And so, you know, the other part that comes from that in your role is listening to customer feedback, obviously watching all these deployments and learning from those deployments. So how do you bring that knowledge back from the field and leverage that feedback from your installed base to inform your product roadmap?

28:23We do a lot directly with customers and then also with our OEM partners. You know, the whole premise, I would say, of the automation is we're working within customer spaces with customers and with their customers. Right. So these these robots are are really intended as, you know, call it appliances. Right. So it's it's not about the hype of buying a robot. It's do I have a really a really effective tool to help me do the job that that I need to do and to help improve my operation, help improve my work environment, help improve my, you know, my outcome from a retail store. And so to do that, like you really have to understand those application problems and ultimately, like, what are you trying to solve?

29:14And so we do that, I would say, kind of two ways. One, very traditional is just get out into the field and meet with the customers, right? You never get better feedback than just talking to and, you know, spending some time and a day in the life. So like a lot of our original, you know, a big pillar for us and original products is within the cleaning industry. So we spent many hours overnight in stores that are closed and with janitorial staff and custodians and operations staff really understand their pain points, help introduce the technology to them, help create like automation that's easy to use, right, and drive that adoption.

29:57And a lot of that is, you know, there is change management, right? There's a lot of just familiarity and making, you know, making the experience comfortable and intuitive and enjoyable, right? So, like, you have to want to use the products. And that comes really by having a lot of face-to-face, like with customers. We leverage that for scale through our partners. So, you know, a couple of key partners with Tenet Corporation makes cleaning equipment, shelf optics that helps us with our scanning business. And I'll say Walmart as kind of a big, big customer partner. You know, really the scale of our partners is what has helped us and allowed us to get to the scale of the deployments.

30:45So you think about, you know, Tenant with hundreds of sales and service folks across the globe and especially across the U.S., you know, they have a lot of direct interaction with customers and a lot of good insights for that. And we get to leverage that, you know, as well. So you're kind of rolling up the insights and the experience. And then on top of that, our whole fleet is a connected fleet. So that's one of the biggest powers and like called the data engine or the corner case engine for doing automation at scale. So the aggregation of all of that data and the ability to just deep dive into it and get the insights and figure out all the corner cases and, you know, almost like read between the lines.

31:34So when you talk to a customer and understand their experience, and then you have the data to look at that and understand, like, where does that really correlate? And then where doesn't it correlate? And both of those are great. So one is reinforcing. The other is the opportunity to go ask more questions and understand, hey, I'm seeing this or, you know, these are the challenges we're seeing on sort of the operational back end. Right now. how do we address those together with you in your operation? So I think it really, these two aspects of just good direct face-to-face conversation with a really rich insight and data source of the customer's operation, those two together, I would say really allow effective deployments and effective operation, good ROI.

32:32Well, John, that sort of gets to our next point. Obviously, you have these great use cases, you have these growing fleets, but let's talk about the technology itself. How has BrainCorp's stack evolved over the past couple of years? Because you're getting this feedback, you're incorporating it into your offering. For our listeners who may not have followed you as closely, what are some of the things that are evolving with you guys? Yeah, so Brain, we call BrainOS, is our full stack solution. And really, I want to emphasize sort of the full stack aspect of this is that we look at this as a robotics business and a software suite to power a robotics and automation business, as opposed to just a stack of, say, the navigation where you have perception and motion and kind of the traditional robotics.

33:27So the big evolution, I would say, maybe two different vectors for that. One, from the buildout of the stack as an AI and robotics business, we really have expanded all of the tools necessary to operate, manage, and deploy fleets at this scale. Right. So we're talking a lot about cloud infrastructure, safety and security, trust centers over the updates, like basically like switching partitions for like, you know, loading or pushing out updates to the fleet, for example. right? Manufacturing tools so that our partners can actually build a robot from, you know, beginning to end, have that fully tested out pretty much by the time it gets to the end of the line based on manufacturing tools and, you know, like test suites and things like that.

34:23So from like the manufacturing side of things, from the service tools, from the fleet management side of things. We've really built out, I would say, an enterprise-level set of those tools. And that's been something that we've built out over time. They just keep getting more mature. Probably the latest badge there is SOC 1, SOC 2. So really getting to a maturity level of infrastructure. The second is, if we look at the advancements on just the autonomy side, you know, we're on what we call like our third generation of robotics. First generation is pretty prototype. Second generation was a lot of the scaling initially.

35:05And now we're into the third generation. And what that allows you to do is to really look at how do I create layers and how do I get platform effect? So if you think about like a single robot, you're just designing for that robot to work. When we talk about fleets and when we talk about making new robots and expanding quickly, like the different styles of robots and types of robots, you really want to try to reuse a lot of those pieces to get efficiency, right? So one of our main goals as a business is how quickly can we bring up and develop a new robot form factor and a new application together with a partner.

35:46And that all comes down to really efficient, really good decisions and efficient use of platform. So one easy example I can give you is safety level firmware running within a lot of our robots. It's making that kind of a common block that you don't have to recertify every time and you get to reuse it. So that refactoring of both the software and the hardware to allow for these layers really allows you to keep going to scale faster and faster. Honestly, there's all this talk nowadays about physical AI. And I think Brian can claim that you were doing AI, physical AI, before it was cool or before it was a buzzword.

36:35what would you say were some of the milestones in the last couple of years whether it's your own or in the space in general and how is that changing the landscape for what's going on with autonomous mobile robots or AMRs today because it's we've seen this development in parallel for a long time but now there's this understanding it's like it's not just a matter of scale it's not just a matter of complexity but you had this level of, you mentioned that feedback loop, both from customers, but then from the technology itself. How has that advanced from your point of view? First, I realized I have to start using the word physical AI probably rather than robots.

37:16So I'm not the hype guy, but I'm practicing. So yeah, like you said, we've been doing physical AI before it was cool, before it was hype. It was called robots. It was called AMRs. You know, really AI, boil it down, right? It's an independent perception and decision-making in action, which is exactly what we have kind of always done. As we look at, you know, really the new Gen AI models and the energetic AI, really an accelerant. I mean, incredible capabilities here, right, that, you know, we're taking advantage of. And that will lead into what is, you know, let's call it for us, the fourth generation of robotics.

38:03But really applying those within safe structured architectures, where, you know, again, it has to have high reliability, it has to have high productivity, and it has to be guarded. Right. And so where a lot of my energy is, you know, I'll say away from the hype is like, what is the boring infrastructure and the guardrails around all of those capabilities? And how do we leverage all of those capabilities for the application like the application need? So what I'll argue is don't solve the problem in a way that's harder than necessary. And so we really, takes a lot of work to get to simple. And what we want to do for these applications is optimize each of these levels that are necessary.

38:56So where we need good deterministic firmware and safety systems, we leverage that. Where we can allow more autonomy, you know, again, I'll say before it was cool, right? We've been using, you know, machine learning, deep learning, reinforcement learning, the latest models, like basically the day they're out or the day we help invent them. In the same way, we're leveraging the new Gen AI models. And going forward, we'll be leveraging the agentic AI as part of that overall ecosystem. And it always comes back to, if I'm going to deploy this at scale in a public space, even in a controlled space, I have to structure that with the right safety levels and guardrails around it and with the right level of productivity.

39:50That's a lot of our focus, I would say, is taking the important... The hype is interesting. The hype is great. It brings a lot of investment into the robotics or physical AI world. Taking advantage of the advancements within that hype and really applying those to commercial applications, that's where the business is. And that's really what we're about. So I love I love the new models. I love the new degrees of freedom for like manipulation, things like that. I really am excited about applying those into real applications. And, you know, a lot of the work we're doing is is is building out the infrastructure for that.

40:35Like we will just continue to do that as we as we go into those new applications.

40:40Gene Demaitre: Well, you know, it's interesting. You guys operate in a unique place, which is public spaces around public sometimes, although you're operating at night, sometimes off hours, you're still in those public spaces. And with AI in the loop, you said it yourself also, you know, the deterministic aspects of controls as well as, you know, the supervisory parts, I assume, are AI. What are some of the things that are challenges of putting AI into that control loop that you've learned? You don't have to share all your secret sauce here, but certainly everybody's sort of dealing with this sort of black box aspect of AI and the decisions it makes versus the classic controls and safety architectures.

41:22Yeah. So maybe to start that is I'll say we went after kind of intentionally the hardest problem first, which is the public space. So I always like to start there and just to ground people on like the the extra challenge that that that brings. So within the public space, you you can't imply any responsibility on to people that that are in and around and kind of interacting like with your with your autonomous robot. Right. And that bar is is incredibly high. then. You know, one advantage of the industrial back in the day was, you know, an industrial arm with a cage around it. Then it was industrial arm with, you know, laser protection, like the IR beams moving to collaborative.

42:08But even in like AMRs and collaborative robots within a, you know, like a warehouse or in a business, right, you can train your employees and you can include that as part of your overall like application HARA. When you're in the public space, you really have to assume essentially no implied responsibility for people in and around the robot. So what that means is, you know, your safety level has to be really high to protect, you know, both, I would say, you know, somebody who's unaware of it, as well as somebody who's intentionally maybe even trying to test the system, which ironically happens a lot, right?

42:47So that's really the most challenging case. And so as we think about that, it really comes down to, you know, guardrails related to degrees of freedom of the robot. Right. And kind of matching those. Right. So to the extent that I can create a, you know, a safety curtain and a safety zone like around a robot, I can now allow a model to have more degrees of freedom because I can protect that zone. as I as I get into spaces where you know the public is really close to the robot I'm weeding in and out you know I have to have a pretty tight safety curtain there in order to work within those spaces in order to be productive right I could always have a large safety curtain and that's great now I have a robot that doesn't move so totally safe but not very productive you know so So as we think about, say, like navigating through a space, like, you know, take an airport is a great example.

43:46I've got a robot running in an airport terminal and the plane gets in, everybody gets off. We affectionately call it kind of a zombie apocalypse from the data standpoint, right? You just, it's just like all of a sudden, you know, hundreds of people are just coming like at your environment. And, you know, so navigating safely there and having, I'll say, a tight controlled safety zone within the robots, having good firmware control on high level to like high level kind of keep you out of the situation and low level to protect you when it's sort of impossible to stay out of that situation. You know, that's like navigation becomes kind of the base use case.

44:27And then as we move up, you know, I think where this really starts to get interesting, sort of, again, is like with the advent of like manipulation and more capable models. I get to applications, say, like picking or placing in a retail store or any kind of manipulation of, you know, say like in a construction, something like that. you need those models with highest degree of freedom in order to actually solve those problems. Right. So we're never going to discreetly program the picking of every single product with an actuator. Right. It's the problem is just too large. Now, discreetly and kind of with our systems, we can we can get to the space.

45:11We can control the safety envelope within that space. And now we can apply a generalized model in order to do like the picking applications. Right. So this this is where, you know, you hear like VLA is really powerful for that. And they're necessary to solve that application. But it can't be like the one thing that that, you know, it's not the answer to the whole problem. Right. It's the answer to a very specific part of the problem that you now need to, you know, I'll go back to like kind of the grounding and the guardrails in order to do that safely. Well, and you've been talking about scale.

45:53BrainCorp recently surpassed, you said, 50 ,000 brain OS powered robots and 5.3 million autonomous operating hours in the first half of 2026. Are there different types of problems or lessons that emerge that smaller robot fleets never encounter? You know, when you now have this installed base, are you seeing new insights or new behaviors at that level? Yeah, I would say thousands and millions probably. So we, you know, in total, we're, well, I'll say well over 25 million hours of safe public operation now, probably close to 30. You know, we take upwards of 25 million images a day within like the retail and the scanning.

46:41So when you think about the data set and the, you know, the number of corner cases that get exposed in there, I'll say almost on any daily run, our robot runs into, you know, some type of corner case, right? It's just everything is dynamic. So really all of those hours aggregating all of that, it just allows you to, you know, allows you to operate in all of those edge cases. And like, I'll give you a simple ones where we're in. These aren't, these aren't sort of the challenge anymore, but, you know, ones that have been solved this way, you know, black tiles within a white floor. They have a completely, you know, a completely different return for a lot of sensors.

47:27You know, and if you operate a small fleet and if you operate, say, in some environments, you may never actually run across like a checkerboard floor. Right. Or like a simple inlay in the floor. You would never know that that's a problem. Now you start deploying cross fleets and all of a sudden, hey, I've got, you know, I've got 100 robots that are seeing this strange behavior or the, you know, it's getting blocked somewhere. And it's because, you know, in fact, there's a new type of black floor, right? Or a new type of tile. So those are just like simple, like simple cases where, you know, every sensor, there's no perfect sensor, right?

48:05So cameras, lidars, you know, like everyone has strengths and weaknesses. even within say like lidars you know every version of the lidar has a different strength and weakness different firmware running you know different returns and so there's there's almost an endless supply of like these corner cases and so that's where like at scale we really you know as we push out updates to the robots you know we can really monitor an incredible amount of the kind of metrics to show like, hey, I'm not having any regressions here. So like you can test a lot of that in the lab, but you really, you don't know for sure until you start rolling out across like the whole fleet.

48:46And that's where, you know, they'll say very often you find like one more small case that you need to be able to address. And that's like kind of on the backend of the development. And then I think where a lot of the advancement's going that is again, really interesting and accelerating is, Robots, the AI running is able to, I'll say, adapt and detect to more of those corner cases, more on the edge and more real time, kind of all the time as you go forward. So we pull a lot of that back. We use a lot of that for our models internally. But more and more, we're looking for adaptation even at the robot level to, I'll say, solve their own corner cases, right?

49:32That's really where you get, you know, kind of a next level of advancement. Well, and I think that's interesting because for a long time, everyone's like, oh, more is better, right? And you just want scale. But I think you're bringing up a great point that you want to also have the ability for the individual robot to sort it out rather than have to like try and draw some aggregate lessons. How big is a typical fleet for a given customer? Like how many robots are we talking. And do you know offhand what your biggest single deployment might be? Yeah. So, so fleet size, it's, it's kind of a, I'll say it, it matches our range of like robot sizes even.

50:09You know, we have, we have a really long tail of customers that have one, one robot right now. Right. All the way up to like our largest enterprise customer is, is Walmart. Walmart, you know, we're full scale across all of their stores, four or 5 ,000 units, you know, and going back to kind of the previous question is having, you know, having good success and a delightful customer experience at both of those extremes is really important. Right. So a customer about one, that robot is just as important to them as somebody who is, you know, managing the whole fleet of the Walmart robots, right? So making sure that both of those customers have good ROI, have good experience, like with the robots is also kind of one of the most interesting challenges of a large fleet.

51:01So if I talk about like, hey, we've made an advancement and we push out an upgrade to the fleet, maybe overall, you know, the numbers look good and the numbers didn't go down, but maybe that update, you know, would negatively impact like a certain subset of robots, right? Like you actually have to be able to understand both of those aspects and make sure that one, you have overall fleet improvement, but two, you know, you're advancing, I would say for like every individual case as well. And that's where like feedback for that long tail of customers with the one, the tens, the twenties, the hundreds.

51:42You know, it's really important because all of those are really unique in a lot of cases. But that from a business standpoint, you know, all of those smaller fleets represent your future scaling because that's where that's where a lot of the expansion is. Right. So most customers don't start out with like a hundred at once. They start out with one robot with 10 robots and they move to a hundred. And then from a hundred, they move to a second application and then to a third application. And then you get to sort of, you know, what, what will be the scaling of this, which is like the, the overall, um, like facility automation where robots are simply kind of the action, um, like the, the action device within like the facility automation.

52:29Gene Demaitre: Yeah. It's, it's interesting to think about the scale of keeping track of all those single robot deployments still have to be connected in your following what's going on there. Let's talk about what you just came back to saying with releases, new features. I think one of the significant features that you've released in the last year or two with BrainOS Clean 2.0, you introduced SelfPath AI. I know the early days, you said it yourself. You started out just trying to do the basics. You did teach and repeat for the robot and then let it do that autonomously. And now, as I understand, I want you to tell us a little bit more about the self path and how that can actually define the path in a space, optimal path for the robot.

53:15Gene Demaitre: So that even makes it simpler now for someone to deploy and manage the operation of a cleaning machine, for example. Yeah. So, so brain, we, we pioneered the teach and repeat. And the reason we did that is that was key to initial adoption from the market. So really straightforward, really simple, like no technician required, as intuitive as any operator could basically set it up, use it, run it efficiently. So really important at the time was teach and repeat. And it was also the first time where we were applying, call it an autonomy kit to a manual machine. And this is a lot like you see the progress in autonomous driving.

54:05So you go along, you have cruise control, you have adaptive braking, you have lane following, you have autonomy options. And eventually you get to full autonomy with no steering wheel. And same kind of true like within these applications is the first robots teach and repeat with an autonomy kit applied to a manual machine was a very comfortable starting point and a very intuitive starting point. That said, that is not optimized for automation. So those robots, they need space for a person. They need a steering wheel. They have cost. They have complexity associated with the manual operation. They have a certain size and form factor that really, really started from this was designed to be a manually operated piece of equipment.

54:50That's not optimized for a robot. So as we look at the next generation robots, both form factor and autonomy need to change in order to get new levels of productivity and advancement. So robots become more designed to be robots or the equipment starts to be more designed to be robots. So you're talking about shapes that are more round. You're talking about the differential drives more so than tricycle drives, you know, robots that can turn in place. Right. So the perfect robot, right, a round robot that can turn in place never gets stuck. And so form factor is an important one. But then you have to match these software capabilities to those form factors.

55:34And that's really, you know, that's our focus. Right. So we're guiding our OEM partners for form factor. And then from an autonomy standpoint, our customers are looking for more and more and like full automation. So they're really looking for robots that just do their job. You know, the idea of the robot comes out of the doghouse, does its job and goes back. That's that's really what you know, what customers looking for, because any time you have to interact with it, that's that's an efficiency loss. So if I have to do care and feeding, if I have to manually set things up, that's an efficiency loss for that investment.

56:19That's true all the way to deployment and starting and setup. So this is where we get to self-path, which is, look, our goal here is lowest effort, lowest cost deployment, or fastest deployment, and a robot that really understands its space and can most effectively complete the job. So one interesting Achilles heel of teach and repeat is that it does a great job of following and doing what the operator taught it, both in the good ways and the bad ways. So if sort of the operator, you know, taught it inefficiently or taught it poorly, those robots are really good at continuing to represent that, right?

57:04Now, with SelfPath, we're saying, hey, now that we've got enough data, now that we've got enough automated capabilities for route generation, we can essentially calculate a better path than, I would say, the average or even the advanced operator can teach. Just because we've got the whole map and we know the problem we're trying to solve. So one, being able to easily deploy a robot, have that robot set up its own activities and its own route to clean the area is both an efficiency gain on the deployment side and on the operation side. And then the second key part of our self-path is we're auto-generating those routes on a per-run basis, which is a really important aspect as opposed to, I would say, prior technologies of ours and competitors where you look at what was called area fill.

58:08So teach and repeat was first, area fill was second, where I drive a perimeter and I'm going to calculate a route within that perimeter. Here, we're calculating routes essentially real-time on a run basis, which allows us to really work within changes within that environment. So the example I will give in a retail store, on occasion, retail stores reset their aisles, call it seasonal, or they're just going to change the aisles from north-south to east-west. prior models would have gotten kind of broke with those larger changes, and you'd have to then redeploy, which is an inefficiency, it's a bad experience, it's another cost.

58:53The ability for us to basically scan the environment, set up routes real-time within those environments, really makes the system now more robust to change, and therefore lower cost and more efficient without like those deployment aspects. And that's really been like one of the underlying advancements with SelfPath is approaching, I would say, a really good solution to what's called lifelong mapping, which is when do I know a change or something in the space is permanent versus temporary and adjusting and evolving like over time, right? And so that's kind of the big advancement right now, which is, hey, I can get something set up quickly and I can adapt over time within that environment to keep high performance without having to do like a redeployment or retraining.

59:50Gene Demaitre: What about more intelligent features that like, you know, classic cleanup on aisle five? So there's something something's falling off the shelf. You need to go clean that up and then get out the way as quickly as possible. Or even, you know, it's a fall day and it's snowing or raining and people are tracking mud at the entrance. So I need to hit that a couple of times a day to keep that clean. And just, you know, the robot just goes out and sort of figures, finds the area that it needs to clean. Less about features, more about what's on the floor kind of operation, right? Yeah. And so what you're getting at is what is called orchestration.

1:00:29I'll say I think that's the new buzzword, the new physical eye. So we take physical eye and we have orchestration. This is the new, you know, robot and fleet management. But orchestration is really what's getting unlocked with the new Gen AI models and like a Gen Tech AI. And it is right at our fingertips. So, you know, you think about like easy robot, like coordination between robots or between a robot and an app. Okay, that's sort of what you have today, right? So if I have an app, I could hit a button. I could say, hey, come clean this area, right? And that app could signal back to the system.

1:01:11You know, robot knows, like the phone knows, or the app knows the location roughly of the device. The robot has the map, and the robot can come to that area and do like a cleaning activity. Where that's moving towards is with integration of like LLMs within the maps and within the systems. You know, now you can actually have systems that trigger those events and can apply logic to those. So I can have a, say, a spill detector or I can have a, you know, something connected to the Internet to say it's snowing outside in New Jersey. Right. And when it snows, I know I'm going to be needing to do like the racetrack cleanup more often in the store.

1:01:56Like that type of logic is really unlocked with, you know, connecting these systems into some of those new models. And that's that's a lot of the work. I think leading into our work with UCSD and a lot of research on semantic maps and embedded maps, really this is the infrastructure to allow for that. And what in essence becomes, it's interesting because Brain, our whole history as a robotics company, You know, really, we're becoming and moving towards a, I'll say an IoT and an AI company from that aspect, which is, yep, the robot is an important part of this ecosystem. But that's what it is. It's one component of an ecosystem that has like a lot of other signals.

1:02:46And it has, you know, really impressive models over the top of all this orchestrating it. and the ability for our customers to interact with our entire system, that is really kind of the big future unlock. So I'll say cleanup IO5, pretty straightforward now, and getting really intelligent to where maybe the building can say cleanup IO5 all by itself. Very cool. Sean, you mentioned the importance of your OEM partnerships. and obviously, you know, we talk about the stack. We're looking at the hardware and the software. And this is certainly one of, I think, BrainCorp's big differentiators compared with some of the other AMR companies that are out there.

1:03:33But what makes your partnerships work well? Because, you know, you have the autonomy stack. Other companies are making the cleaning robots or the mobile robots. And there's, you know, there's the whole notion of, okay, how do you manage service and customer support and all those things? What goes into that to make those relationships work well for you? I think it's really looking at where the strength of the partnerships are. So, you know, it's a bit of the divide and conquer and leveraging, like, what are the strengths of each company? And finding, you know, so one, it's really finding partners that have really complementary skill sets, right?

1:04:17And the second is, is I would say partners that really match kind of your ambition and your impedance. So you don't want to be like sort of outsized one way or the other also. So from a skill set standpoint, this is where we really invest heavily on the software, on the AI, on the automation aspect of things. We rely heavily on our partners to manage the manufacturing, the design, the physical robots, and all of the care and feeding that go with those, right? So all the field service and things like that. And so finding partners, I would say, that are really capable in a complementary area allows you to scale quickly and kind of reach new levels of scale.

1:05:01that you, you know, it's not impossible, but more challenging, I would say, in a pure vertical, you know, a vertical integrated approach. Well, you mentioned Tenant as one of the companies with whom you have a relationship. And I know that BrainCorp has extended that partnership with plans for 10 new robotic cleaning products within 24 months or two years. How do you manage the pace of innovation and maintain that product alignment? Yeah, I'll go back to probably platform is the key word. So in order to do that, you really have to have a strong platform effect because you can't make 10 snowflakes.

1:05:48You have to be able to leverage infrastructure. You have to be able to leverage different tools at kind of every level of the stack in order to achieve the goal of each application. But from an engineering perspective, almost make it as derivative as possible. So really, anything that is common really needs to be efficient and needs to be common so that you can put your investment and your energy towards the things that need to be unique. And that reuse, that platform efficiency, Like that's what unlocks us to be able to, you know, expand, I'll say the product offering, you know, very quickly here.

1:06:31And of course, on top of that, you have all of, I'll say we're using all of the new AI as well, like even, you know, in our internal development. So I don't think anybody kind of, you know, writes a line of code anymore without at least thinking about it and how it works together. So the other side of it is just accelerating the pace of development and leveraging all of the new tools in order to do that. Well, and as you said, you're extending capabilities. I know we recently covered BrainCorp has expanded from floor care to inventory intelligence and shelf scanning. What was some of the lessons you were able to apply from your existing fleet into these new areas?

1:07:17Yeah, cleaning was a great entry point and still is for a lot of companies. It is a clear kind of ROI and a clear task. You know, floors need to be cleaned and most businesses have a floor. So it's a really good entry point for automation for our customer and for a business. What that allows us is that actually helps us build the infrastructure for that customer, like of their facility. So with the cleaning robot running within a space, you create a really rich digital twin of that environment that now allows you to up level into different applications. And one thing we learned really early on is the amount of data and the insights you get from the environment just to navigate safely really have a lot higher value in other applications.

1:08:15Right. So the fact that, you know, robots are running with with full 3D LIDARs and high definition cameras, you know, again, they make an incredibly rich environment for automation. And the very next question you, of course, ask is like, oh, if you if you have the ability to capture the images within the space, like, can you give me more information about what you're seeing? And, you know, so one is a very specific application use case, which is, hey, let's not only do the images, but let's let's even get better cameras and more high resolution, you know, and point them really at the shelves in order to provide insights at a retail level for inventory.

1:08:55So out of stocks, product availability, price compliance, very specific application of kind of this general ability, which is, hey, you have information about the environment. What value can you provide with that? So one is very specific application. So we're scanning and retail analytics and like warehouse analytics. like where are my palettes? RFID really unlocks a lot of even higher resolution to that with product count and product location. And then on the second side, going back to, we were talking about kind of the future, being able to actually query that information and work with that information.

1:09:45So you're talking about like clean up IL-5, right? Those images also work to feed those general models with sort of a piece of information and context within the application.

1:09:59Gene Demaitre: Yeah, I love that. I love the pragmatic approach you've taken to extending your capabilities step-by-step building on what you've had before. Before we let you go, so the last question really is the next interesting innovation is in Project Braun, which is extending now to what I think has been one of the holy grail applications for retail, which is shelf stocking. It sounds easy to us as engineers, but it's really a difficult task. But again, it builds on everything you've given us so far today. So just briefly, introduce us to Project Braun. I know it's still in the planning early deployment stages.

1:10:39Gene Demaitre: So tell us what's going on there. Yeah, so Project Braun was really our first proof of technology of integrating manipulation with the VLA within our system. So going all the way back to kind of the first part of our conversation, you know, having the ability to integrate these large models and these new models into the system, I think is really key. And so Project Braun, we did almost two years ago now, I'll say a good year plus ago, to really prove out the technical feasibility of integration of these two things. And, you know, this is where, as we look going forward, certainly moving, you know, AMRs are really good at moving things around, right?

1:11:27And now manipulation is really good and getting really good at, I'll say, taking things that have been moved around and more precisely move them around. You know, so I do, I definitely see the future applications are coming within that space.

1:11:46It's, you know, our goal, I would say, is to really be able to leverage the latest advancements within like an application. So where the second part of this is, I think, the comfort of the customers within those environments. And it goes to like the change management and the adoption as well. So probably the lenses that we're looking at this through is one is the technical, which is, okay, what is the best form factor to integrate like manipulation, right, for the task? The second is what are the best models tuned for those environments? This is where a lot of our back-end data, scanning data, things like that will feed, will, you know, very rich, I'll say data source to feed into these models, right?

1:12:32And then, you know, how do you implement that within customer environments, right, where you gain high confidence and adoption? And I think this is where very specifically, you know, I'll say to call out sort of the humanoid hype versus the, quote, boring robots. You know, the technical problem of humanoid robots, really complex, like really being solved well can't be, right? These are incredible, you know, dynamic devices that have really, really good controls to them. the problem may not actually be the technical of solving that. The problem is maybe the adoption of that from like a customer and a public perspective, right?

1:13:23So I think there's, you know, we live in the world together, right? So there's very much the human aspect of it that you have to take into account, like with any of the new technologies. So it's not just like, can we do it? That's one. Over around that is like, hey, can we do this safely? I'm going to argue that is an absolute requirement. And then two, can I do this in a way that is really adopted and accepted and therefore highly productive? So when I think about inflection points of the industry, to me, this inflection point of an industry is when robots become boring or when the solution becomes boring.

1:14:09If you think about today, you know, most people go into, say, a building or environment and they won't blink or think twice about a automated cleaning robot going past, right? It is furniture, right? You know, so it needs a good ROI, it needs to be productive, and it fades into the background. And to me, that's kind of what we're looking for as you apply the new agentic AI capabilities, new manipulation capabilities, VLA's in these form factors is when do you reach a point with the right form factor, the right application, the right implementation that it becomes boring and kind of fades into the background?

1:14:53And that's really what, you know, that's what we're looking for. So I think you mentioned in the beginning where maybe BrainCorp, we don't tout our stats loudly enough. You know, boring is good in that regard, right? We're building the real scale and the real business. So I'm really comfortable, you know, in this space, I'll just say happily scaling in the background.

1:15:16Gene Demaitre: Absolutely. Well, John, it's been a great conversation. I want to thank you for sharing all that with us. It's been great to peek under the hood, so to speak, with Brain and really get a good understanding of where you are. So thanks for joining us today. Yeah. Thanks, Mike. I really appreciate the opportunity and the good questions and discussions. And Gene, thank you.

1:15:40Gene Demaitre: So Gene, that was a great interview with John. And one of the things I loved is that he talked about how robots should be boring, especially in the commercial space where consumers and the public are interacting with them in some form. And that certainly happens for some sub-segment of the brain corp deployment. And really his measure of success will be when the public views them just as a normal part of the built environment, right? They're just there. No one stops to take a picture of it, maybe gets out of the way of it when it comes by if it needs to. But in the most part, that's what his measure of success will be.

1:16:21Gene Demaitre: And I think they're on a path towards that endgame. What were some of the things that you found interesting from that conversation? Well, first, I agree with you that when robots become appliances or part of the infrastructure, they'll have been successful. They may not be as sexy anymore. I remember when the first robot was in my grocery store, I took selfie with it as much as anyone else did. But now it's just there and that's what you want. I think one of the other hurdles that's worth tackling is linking workflows right so we talked with john a little bit about how different types of robots and different types of functions will coexist whether it's an airport or a hospital or a warehouse and you know when we're talking about doing a range of things like cleaning up spills or reacting to weather conditions it's more than just navigating around a pillar or a person And it's really getting to that point of extending the capabilities of robots and then doing it at scale.

1:17:23And I like how brain has been very practical and pragmatic and how it's adding those functionalities, those capabilities like inventory. We already know they're cleaning. Having done some of those tasks early in my working life, I appreciate the ability for robots to take that on. and, again, free up people for the supervisory aspect.

1:17:48Gene Demaitre: Yeah, it's definitely been fun to watch them grow. You know, it's been probably a decade now since, you know, they first came on the market and very focused on what they've done. But, you know, just one step after another, one hurdle after another, they have, in many regards, some of the biggest fleets of robots deployed, you know, around the world, and not many folks recognize that. But indeed, not large numbers of fleets on any given site, but certainly individual robots and individual stores all over the world, which is really amazing and fascinating to comprehend. So I want to thank John for coming on the show and sharing that with us.

1:18:32Yeah, there is a theme to this episode of our podcast that's definitely scale, right? I think the size of the deals is getting bigger. The size of robot fleets is getting bigger. The number of robots in warehouses and other facilities is getting bigger. And I think that's encouraging. I think that is a good antinope to some of the wackiness that we see in other coverage of the technology space. But before we go, here's another reminder about our upcoming robo business event in October. It's October 20th and 21st in Santa Clara, California. It's our 20th anniversary show. And as Mike said earlier, he, Steve, and I have worked on it for probably more than half of that time.

1:19:15And so we're really looking forward to celebrating and sharing insights from our headlining keynoters. But we're also expecting over 100 exhibitors, more than 2 ,000 attendees. And if you register between now and Monday the 31st, you can save money with early bird discounts.

1:19:34Gene Demaitre: To learn more, go to robobusiness.com. Well, Gene, that's going to do it for episode 259. Thanks again to everyone for tuning in. And thanks to John Black, CTU of BrainCorp, for joining us today. Gene, thanks for joining me as well today. New episodes of the Robot Report podcast drop each week. You can find us everywhere you get your podcasts. Please subscribe, leave us a rating, give us a review, but most importantly, tell your friends about us. This episode was produced by Mike Weitzman and edited by Tyler Kubisti. Our executive producer is Amanda Buhner. Thanks for listening.