theCUBE + NYSE Wired: Physical AI & Robotics Leaders QA2 | Sankaet Pathak, Foundation Future Industries

Clips
More from theCUBE + NYSE Wired: Physical AI & Robotics Leaders QA2

Sankaet Pathak

Co-founder & CEO

Foundation Robotics

play_circle_outline Envisioning a Future: Deploying Robots for Earthly and Extraterrestrial Tasks to Revolutionize Data Collection and Commercial Use

play_circle_outline Robots perform repetitive tasks in auto manufacturing, addressing high turnover labor rates.

play_circle_outline Current production model features a bipedal robot improving task cycle speed and flexibility.

play_circle_outline The team is growing in San Francisco and Germany, currently seeking additional talent.

play_circle_outline Customer feedback is positive with deployments in auto and consumer goods manufacturing.

Info
Transcript

Sankaet Pathak, Foundation Future Industries

Sankaet Pathak

Co-founder & CEO Foundation Robotics

Sankaet Pathak, co-founder and chief executive officer at Foundation Robotics Inc., joins theCUBE’s John Furrier during theCUBE + NYSE Wired: Robotics & AI Infrastructure Leaders 2025 event to explore what it takes to scale humanoid robotics in the real world. The conversation dives into Foundation Robotics’ mission to unlock “infinite labor” through AI-native automation.

Pathak unpacks the company’s unique approach to autonomous mobility, industrial deployment and proprietary AI model development. From manufacturing efficiency to adaptive task execu... Read more

explore Keep Exploring

What initial use cases are being targeted for deploying robots in various environments? add

What types of jobs have been the focus of automation efforts in various industries over the past 13 months? add

What advancements have been made in the robotic production model and how do they impact task efficiency? add

What is the current distribution of the team and future plans regarding team size and locations? add

What are the current customers and use cases for the robots being deployed? add

bolt Powered by CUBE AI

Sankaet Pathak, Foundation Future Industries

search

John Furrier

>> Welcome back everyone to theCUBE, we are here in the Palo Alto studios. I'm John Furrier, host of theCUBE. We're here for the Robotics, AI Leaders Series, three days of coverage and a big face-to-face event at Rosewood on Wednesday. Got a great 170 people showing up for face-to-face. Of course, it's our digital twin. We are in the studio breaking all down all the action in the future of AI infrastructure leaders who are driving the future. Of course, robotics is the physical AI aspect of it and of course robotics is taking the world by storm. You're real use cases, real practical use and an accelerated entrepreneurial culture as well as just technologies. Sankaet's here, co-founder and CEO of Foundation Future Industries. They deployed the fastest humanoid yet to the market. Congratulations. I've got that scoop here before we came on camera. Thanks for coming on the program, being part of our show.

Sankaet Pathak

>> Yeah, thanks for having me.

John Furrier

>> So I had to kind of say the stat, so it's true you're the fastest deployed.

Sankaet Pathak

>> Yeah, we're about a 13-month-old company with a production robot deployed. So our robots are currently doing actual field work on an auto manufacturing plant and our goal is to ship about 100 plus of them this year and then try to ramp up to thousands next year.

John Furrier

>> That's awesome. I love the humanoid side of it. Love the industrial impact. Robots have been around for a while, but you're seeing a lot of culture shift around the real world use cases gone mainstream because of the use cases. There's a lot of demand. So take me through the origination. 13 months, not a lot of time. You guys ramped up pretty quick. Take me through the origination and the progression of growth.

Sankaet Pathak

>> Yeah, I mean we spent a lot of time trying to figure out what are the first initial use cases we should really go after. The long-term vision for Foundation is we want to deploy a massive fleet of robots in cities on different planets, wherever, and then you can just call them on demand like Waymo or Uber and they'll come and do whatever you want them to do. To be able to get there you need to be able to deploy a pretty large fleet that can collect a lot of data. The best place to be able to do that is in factories, warehouses, and defense use cases because a lot of them are structured, data is more easily available and they're kind of repetitive and that would enable you to be able to deploy large fleets. So a lot of our focus in the last 13 months has been to specifically do pick place, move, type jobs in car manufacturing, consumer goods manufacturing, some defense logistics work as well. Primarily to be able to replace all of this repetitive labor that also has pretty high turnover rates. So like close to 100% turnover rates for most of these roles. And the goal is to have these robots replace those and do three shifts pretty much nonstop.

John Furrier

>> So take me through, you're sitting around, you say, hey, I'm going to start a robotics company, go get some funding. Obviously that's not what happened. But take us through the process. Were you scratching an itch? Did you have a vision? And then the funding needs to be there, safety, all the software. It's not trivial.

Sankaet Pathak

>> It's definitely hard. Yeah, when I was thinking of starting Foundation, I kind of just more so thought about what are the technologies that would have a profound impact on humanity in my lifetime? And AI was one of the biggest ones. And in AI, I believe kind of real world autonomy or what I call infinite labor is probably one of the biggest impacts that can have on humanity for multiple reasons. The birth rates are declining and right now we already have a labor shortage that will only get worse in next 30 years. You'd still have about the same population because people are living longer, so you'd have about the same demand. But you'd have less capacity to be able to manufacture. Japan, Greece, all of these are good example countries where you already are seeing the impacts of essentially a very low young population and what that really does to an economy and eventually to the civilization itself. So this seemed like one of the most important things to work on. So on the pessimistic side, you do want to be able to avert some kind of big recession that could come because you don't have enough labor. On the positive side you want to be able to expand beyond earth. And one of the biggest constraints outside of actually building the rocket ship is labor because human astronauts cannot go and build infrastructure in space. So you do need robots. So I felt like with-

John Furrier

>> We got to get the Mars somehow and, yeah, that's the path.

Sankaet Pathak

>> Yeah, I think with the limited time I have I thought this would be probably the most impactful thing I'd be working on.

John Furrier

>> Okay, so great mission, love the mission. Now aligning with investors, how does that go?

Sankaet Pathak

>> I mean you have to be able to demonstrate that, one, you can build something that's useful and second people want to buy the stuff you build that's useful. So last 13 months have been pretty much that. Substantiation of the company, being able to demonstrate that we can actually build a robot that is production-ready and their customer is willing to buy tons of them. You can say whatever you want to investors, it really comes down to can you demonstrate, one, you can execute on the technology, second, can you really execute on distribution? And I believe the team's done a great job in the last 13 months and we'll just continue to do that. I believe we're going to do that well we'll have investors willing to invest and if we don't then we deserve to die.

John Furrier

>> So I love the mission of having a fleet, that's great, you've got more volume. But also it makes me think of that Star Wars episode Attack of the Clones when you see all those robots being built and then they find out that they had a master robot and then they cloned it, basically the clones. But all kidding aside, that's in my vision of what a fleet would look like. Thousands of robots being manufactured and then shipped. Is there central software? I'm just kind of nerding out here. Interesting. How would that lay out? Take me through the design side of it.

Sankaet Pathak

>> Yeah, if you really want to make a very giant fleet useful, obviously you need to make sure that the robots autonomously can do tasks safely and can do what's called zero-shot tasks, tasks that is never seen before and it's seeing for the first time. So a lot of the work that we're doing right now is around that. Being able to build reliable hardware, enable it to do very specific jobs well, that would enable us to collect more diverse data to then be able to zero-shot tasks. But then the final frontier for massive fleet to actually be useful for large-scale projects is something I call fleet coherence, which is every single robot is aware of what the other robot's doing in its fleet. And with that, for instance, if you want to deploy 1,000 robots to be able to build a city, every single robot knows what part of the infrastructure is built, what needs to be built, where should they be allocated and deployed. And that's the only way you can also avert this one central large AI. So this would be pretty distributed like how human intelligence is except better connected, which we're kind of connected with the internet, but it's like having internet your brain, so humans plus neural link. But I think that's the eventual goal, which is if you have a fleet that's fully coherent, understands exactly what the other robot's doing in the fleet, then you can do much larger high context window type operations and build, for instance, cities.

John Furrier

>> I like this idea and people calling it collaborative, it's a category collaborative robotics. Is that the right wording? What category would you call this? I mean is it collaborative? Because if you're building a city you're building something together, it's collaborative.

Sankaet Pathak

>> Yeah. I think it's collaborative robot with robot. Usually when people use collaborative robots, they mean robot and human. We would much rather have humans not do any labor. It's clear we don't enjoy doing it, so might as well try to automate all of it. But yeah, the robots would have to collaborate with each other for this to be successful.

John Furrier

>> All right, so take me through where you guys are at now. Obviously you deployed. The demos are awesome. You have a rolling version. You have legs now. Where are you on the progress? Can you share some of the product specific advances?

Sankaet Pathak

>> Yeah, for sure. So currently the production model that we have is a humanoid upper torso on a wheel base. So that's the one we've deployed at the auto manufacturing facility. We have quickly iterated on it and now we have legs. Now we have a bipedal robot. For a lot of the use cases that we're going after, a bipedal would have a faster cycle time than a wheeled robot because we don't have to do long distances, we have to work within cells.

John Furrier

>> When you say cycle time, what do you mean by cycle time? Manufacturing cycle or life cycle?

Sankaet Pathak

>> Cycle times mean the speed at which you do a task.

John Furrier

>> Okay. -

Sankaet Pathak

>> So what's your turnaround time on a task. Because if you have wheels you have to reverse, forward, do a bunch of these different things, it's slightly slow. You have legs, you can just turn around, it's much more nimble. So now we're in the process of decommissioning almost all of the AMRs by end of this year, aka the wheeled robots, and primarily going forward we are doubling down on legged, bipedal robots.

John Furrier

>> Okay. Obviously we heard safety first, that's been a theme. So we don't need to drill down unless you want to. But I want to ask you, what are you optimizing for right now? Are you optimizing for better manufacturing, better software, go to market, all the above? What's your focus?

Sankaet Pathak

>> Currently our absolute focus is get to a milestone where these robots are doing full shifts fully autonomously.

John Furrier

>> Give an example.

Sankaet Pathak

>> For instance, one of the tasks we're doing is called cladding in which a part comes down the conveyor belt, you pick it up, you put a label on it, you QC the part, and then you essentially rack the part. Then the rack fills up, somebody comes and takes the rack to final assembly. So that whole task just needs to be done again and again and again and again. It needs to be done for eight hours three times a day. So everybody's just working through that task. Well six hours, three times a day. Our goal is to make sure throughout those six hours there is no human operator that had to intervene to make sure the robot did that task accurately. So far we've crossed about two hours and our goal is to increase that from two to four to six. Technically there's no big change between doing one shift to three shifts, but in the next few weeks our goal is to be able to do full shifts fully autonomously. One thing that's different about Foundation to other companies is we're not deploying these robots with a tele-operator behind the scenes. We're deploying these robots to be autonomous from the get-go. So it's very important to be able to achieve that. Before you start scaling manufacturing, the most important thing is that the product actually works. The hardware is quite reliable, so we actually don't have a lot of hardware breaks, so now it's a function of with AI and software being able to go from two hours to six hours.

John Furrier

>> So software's critical?

Sankaet Pathak

>> It's now primarily a software problem.

John Furrier

>> Take me through the hiring process. What kind of candidates roll through? I mean robotics does attract diverse, technical alphas coming in, they want to work on some hard problems. Plenty of hard problems. Lay out some of the hard problems. And what are some of the candidates that you guys are interviewing, certain disciplines? Software obviously big one.

Sankaet Pathak

>> Yeah, pretty much hiring a cross-board from electrical mechanical to AI controlled software, cloud. You have to build the whole system so it's on this one feature, you can have to really put the whole system together. As you said, there's a lot of talented people interested in solving this as a problem, which does make it much easier to be able to find highly motivated people. Last 13 months it's been pretty much weekends and nobody took Christmas off. So by definition we were working pretty much nonstop and that's how it's going to be for the foreseeable future. So we're very honest with candidates that it's definitely going to be long working hours.

John Furrier

>> It's going to be a grind.

Sankaet Pathak

>> It's definitely a grind, it's a race. We love talent that's already worked on robots, if you worked on humanoids even better. But even people who've worked on very hard technical challenges. We have folks from SpaceX. We have folks from Tesla. We have folks from Boston Dynamics, 1X, bunch of these different companies. By the time you're a really good engineer who's highly technical and likes to work on very tough problems and have high endurance and high stamina, we love to attract those people.

John Furrier

>> You mentioned the use cases you're going after, well-structured, obviously low-hanging fruit market entry, you knock down use cases you can do, attainability. Getting attained-

Sankaet Pathak

>> Yeah. There are pretty much millions of jobs you can just do by doing this.

John Furrier

>> Yeah, it's a safe bet and it's also a great starting point.

Sankaet Pathak

>> Yeah.

John Furrier

>> Why is that important and where does that change? When does the tide turn?

Sankaet Pathak

>> Yeah, I thought it would be... Well two things. One, it's going to be very difficult to deploy a large fleet unless it's useful and it's very difficult to make robots useful unless either the task is constrained or you have a lot of data. So a lot of our focus has been, tactically speaking for us to be able to get to a place where we have very large fleet deployed so that we can collect a lot of diverse data we need to be able to do specific tasks really well so that people are willing to buy more and more and more of the robots. The second piece of this is I don't think anyone would fund Foundation if I went and said, I just want to build a civilization on Mars. So you have to be able to build-

John Furrier

>> Hold that back a little bit during the fundraising.

Sankaet Pathak

>> Yeah. Exactly.

John Furrier

>> You can say the north star. What's your north star? It's not the North Star, it's planet.

Sankaet Pathak

>> Yeah, it's technically on our website. But for us to be able to get there, we just have to be a high profit earning business, it has to be big and we're going to use almost all of those profits for furthering that goal. And that's a big part of also doing the defense and industrial use cases because the demand is massive, few customers get you large purchase orders and that enables you to be able to scale and deploy much faster on tasks that are much more structured and contained versus home, it's totally chaotic and it's going to be some more time before a robot's safe to be around your kids, for instance.

John Furrier

>> When do we cross over from a population standpoint robots to humans in San Francisco? How many years do you think it's going to take?

Sankaet Pathak

>> Where we have more robots in San Francisco than humans?

John Furrier

>> Yeah, population. I mean, Waymo's a good indicator, are you seeing-

Sankaet Pathak

>> I think we'll have more robots around the world than the population in San Francisco in the next five years. But in San Francisco alone, I don't know, I think we're probably couple of decades away.

John Furrier

>> Think so? Okay.

Sankaet Pathak

>> And I hope so too. I don't really want to see a world where there are fewer people and more robots. I want to see a world where humanity is flourishing, they're doing things they enjoy doing, kind of like the Star Trek-type future where everyone's pursuing their curiosity, they're not having to work for money. So that's the bigger motivation. But I do think in the next five years we'll have millions of these robots deployed.

John Furrier

>> And one of the things about AI that I love in the conversation is there is a human and humanity element. I think the robotics job piece is a great mission. Like I said, sometimes mission is greater than money, but money will follow if you hit your mission. So it's kind of tied together. Is that the human productivity is a big discussion in work, but we're getting at a different conversation here, this is about human productivity for themselves. Creativity, curiosity, being a good fellow human, being chill, do whatever you want.

Sankaet Pathak

>> Yeah. I think the definition of what it means to be a human is going to change drastically. Since the beginning of civilization, it has been really important that we participated in manufacturing goods and services. But we're soon going to hit a place where you don't need to work for those things to exist, which means for capitalism, the incentive would not be human labor, which opens up hopefully-

John Furrier

>> ...

Sankaet Pathak

>> for humans to do things that they enjoy doing.

John Furrier

>> One of the big trends we follow also, not necessarily this theme, but we'll do in next week in New York, is decentralization. If you have transparency, democratization of technologies, you can almost get capitalism done right. If it's open, you can see everything, there's no hiding. I mean capitalism isn't bad. It could be bad.

Sankaet Pathak

>> It's not perfect, but it is probably the best system.

John Furrier

>> You can get better at it by knowing what's going on.

Sankaet Pathak

>> Yeah.

John Furrier

>> All right. What's your dream scenario? Steady state, shoot the arrow forward five years, where are you? What's happening?

Sankaet Pathak

>> Yeah. This year, again, the goal is to deploy 150 some robots. Next year we aspiration you want to deploy 10 to 20,000 robots and then try to 10X production for the foreseeable future. And then after that, probably 2X or something like that. In next five years we want to be at a place where we deploy maybe a million robots or something like that and be a large profitable company. Currently it's just heads down executing on that.

John Furrier

>> And you guys working hard, so really grateful what you do. Talk about the company. How many people? Give us some of the numbers. Funding, you guys feel good about that? Flush full of cash? Where's the locations? You mentioned US, here in San Francisco, you got Germany.

Sankaet Pathak

>> Yeah.

John Furrier

>> Talk about the company, what it looks like.

Sankaet Pathak

>> Yeah, so currently the team is in San Francisco and in Germany. All of the manufacturing software hardware team is based in San Francisco, our AI research team primarily is based in Munich and I kind of split my time between both. We're close to about 40 people now. We have about 30 plus openings, so soon we're going to be close to 70, but the core R&D team will probably not be over 100 to 150 even in the long term for the humanoid program. Again, I don't want to comment on fundraising yet, but we have capital to be able to execute on what we're working on.

John Furrier

>> Awesome. What's the secret sauce? What's the breakthrough? How would you describe that?

Sankaet Pathak

>> I mean it's primarily, we pretty much have a very proprietary AI model that is able to encode into a model architecture the physics of the real world. So being able to understand objects, their dynamics, kinematics, relationships so that then you can reason through the action space. Which is very different to what other people are doing, other people are using imitation learning, which is more so trying to imitate the motions you see in the training data. We're more so trying to understand the task space, the robot inhabits, and based on that, be able to dynamically reason what needs to happen next and have that predictive ability. So the core of it, that is our strength. We also have cheaper, more efficient actuators, which are essentially the muscles of the robots, the motors that enable you to move the robots. We have the lowest friction gearbox, which essentially means the least wear and tear and highest torque throughput, which means it's more powerful and it can last longer. It's much, much simpler. So we think that's a better technology that enables us to scale the fleet faster. And also same with AI, having this deep variational base filter model, which is proprietary to Foundation, enables us to be able to train these AI models with very little data compared to the alternatives, which enables us to be able to get to full autonomy.

John Furrier

>> You mentioned that earlier that you guys want to deploy data, observability some data. A lot of people use synthetic data, they try to train. Similar approach? A little bit different?

Sankaet Pathak

>> Currently we're only using real world data, but not for any kind of, we don't have a religious association with one the other, it just happens to be that the amount of real-world data we have is sufficient and there are some gaps in synthetic data around physics. So those gaps have to be bridged and maybe for more complicated tasks we'll need more data. But currently we have enough data to be able to do this -

John Furrier

>> And you're in a use case where you have some good available data? Safe?

Sankaet Pathak

>> Yeah, customers are very willing to be able to share data If the outcome of that is they get an autonomous fleet. So it's much easier to collect data. And again, there's not a lot of diversity in the task. For instance, the cladding use case, the part always comes down the conveyor belt. It might stop a little sooner or a little bit later, but it always comes down the conveyor belt. It might come in a different orientation, but always comes down the conveyor belt. You pick it up, you put a label on it, you QC the part. If the QC fails, you discard the part. If it passes, you rack it. And you have to do that again and again, which is not what you do at home. So by definition here the problem statement is easier in terms of being able to build something reliable that works repeatedly.

John Furrier

>> Yeah. It's interesting too because now the coding risk goes down because you now have... Okay, it's coming down, that's taken out. So now it's what size? Computer vision, multimodal is always big in the proprietary and open models. You're leveraging computer vision, obviously.

Sankaet Pathak

>> Yeah. I mean you cannot code your way out of this. There's no scenario in which just like writing a bunch of code will enable you to be able to do these things autonomously. So it has to be through some kind of a deep learning architecture. We use techniques like localization, perception, being able to understand the environment, like trajectory planning, execution with control. So it's a mix of all of it. And the great thing about viewpoint multimodal models, a lot of this stuff is getting consolidated into singular policies, which was not possible a few years ago, which again is the big focus for us. Now. You have perception localization, all of this get consolidated down to SLAM model, you have a high-level reasoning model and you have pretty much an action token prediction model, and those three are kind of just deep learning models.

John Furrier

>> Is there a technique and technology that you see coming that you're excited about that will move the needle?

Sankaet Pathak

>> You have all of the material constituents from a technology perspective to be able to pull this off. Like some breakthroughs in reinforcement learning and transformer models were like the big holding pieces which have been now unlocked in the last two, three years. So you have all the constituents to be able to pull this off now. The only big limitation is data and obviously that limitation is a big gap or a small gap depending on the tasks you're really trying to do. So for home use, a lot of diverse data is your big gating factor. For factory warehouse use cases that are pretty repetitive, not as much, you can collect enough data to be able to train those policies.

John Furrier

>> Got it. All right, customers, what's the customer feedback you have now? You said you got some happy customers. Can you share the names or can you talk about the use case?

Sankaet Pathak

>> Yeah, we're currently keeping the names confidential, but the use cases, we have about three customers. One of them we're actually deployed with. Two of them we're hoping to be deployed with sometime this year as well. One in auto manufacturing. Two in consumer goods manufacturing. Then we've done some small dollar contracts for the US Air Force and Navy. We intend to do more of those, so more to come on that soon. Working with a couple of defense manufacturing companies as well. Right now the robots that are deployed are working on auto manufacturing in a task called cladding, which is the racking and unracking, label stamping

John Furrier

>> Repetitive tasks.

Sankaet Pathak

>> Repetitive tasks.

John Furrier

>> That's a good use case. Thank you for coming on theCUBE. Really grateful for your time. I know you're super busy now you got to go back and get some coding done. Get the team, keep working, you guys working around the clock, burn the midnight oil as they say. Congratulations.

Sankaet Pathak

>> Yeah, but it's fun.

John Furrier

>> Yeah. It's fun.

Sankaet Pathak

>> It's sci-fi, so it's great.

John Furrier

>> I mean, how can you not love the industry right now? It's really a special time.

Sankaet Pathak

>> Yeah. Thank you so much for having me.

John Furrier

>> I'm John Furrier. We are here inside theCUBE for our Robotics and AI Leaders. The future's unfolding in front of us. We're just doing our best to keep up and sharing that with you. Thanks for watching.