speaker-0: Hello, welcome to the Robotics Stack Podcast. Today we're joined by Matt Trossen, founder of Trossen Robotics. For more than two decades, he has been in the robotics industry. If you're a researcher, engineer, tinkerer, developer, you might even be using his products already. We will talk about what he's learned from being in the space for so long, the latest products from Trossen, of course, and where robotics is going next. This is episode 21. My name is Oscar.
speaker-1: My name is Leo. Matt, thank you for being here. We are super happy to have you with us today. Thank you very much. And you have a very nice setup on your end here. Yeah, we built in our podcast room right now. We've shot I think about ten episodes. We haven't released any yet. Are you planning to release regularly or is it more of a one off thing?
speaker-2: No, we'd like to have a regular cadence. I believe that it's kind of our duty and everyone else's duty in the stages of physical AI evangelize and promote what's happening and help have conversations going so there's more understanding. Think about industry you think back to that jobs would evangelize.
speaker-1: understanding of it. Like we think about it like the early PC industry. And if you and you have like Steve's right and let people know what is this new technology, what can it do? You know, for us it's marketing obviously to get the work. Also to educate, 'cause a lot of people are wondering
speaker-2: Obviously, but I think it's all looking at this next wave of robotics and what
speaker-1: what is it? What what can it do? And sometimes in depth conversations help keep that going. Yeah, this is exactly the insight we are looking for also when we started the robotics tech. We are going to dive into your background and the journey into building trust and robotics, obviously. But first I would like to ask you, you started your company more than twenty years ago. And if today you were at this stage, you are just getting started, would you still go into physical AI, into robotics if you were starting your company in twenty twenty six? Yeah, I think for sure. I think it's
speaker-2: at its most fascinating it's ever been. Waiting at this point. So if I was to start all over
speaker-1: point that it's and I've been waiting for it to get to this point all along. Or it'd be kind of crazy to go do something else, I think. The company two
speaker-2: When we started the twenty twenty
speaker-1: years ago, which is just wild to even say at this point. The idea PC based robotics. It was the idea that the PC would be the brains and the nervous system,
speaker-2: idea was basically piece of it was become the center or you know brain of the nervous
speaker-1: if you will.
speaker-2: we started selling components on a store at the timer called Fitz or control electronics.
speaker-1: digits USA they were USB components that allow you to control and take input from sensors. and that was the concept rifted towards robotics. At first it was just sort of a tinkering set and then we and
speaker-2: And just trended more towards things. Added the motors and the servos and
speaker-1: it just logically was like robotics is what needs this, right? And so that's
speaker-2: So in those twenty two years that
speaker-1: what's happened is you've had, you know, places like Wheel of Ross, which sort of established the ability to have the computer Linux. And it
speaker-2: Garage creep staff have in the center on a system is grown from there and now we wouldn't even think of building a road. I didn't know we were waiting. I didn't know that in my lifetime.
speaker-1: robot without a PC in it, right? what I was waiting for was AI. AI would exist, right? And so three years ago when Stan demonstrated how you can do robot learning through VLAs, everything realized,
speaker-2: Exploded and you know we're
speaker-1: my gosh, this is what we've been waiting for. This is what allows the robots to actually do something now, rather than just exist in a lab. From what I understood, you quit your job and decide to start. robotics company in two thousand and four. I'm guessing some people might have thought you were crazy at the time to start such an endeavor. What was your vision? Okay. I didn't quit my job of the gate. I quit my job about a year and a half later. I had a boss at the time who was nice enough to let me do didn't
speaker-2: So right out of the later. 'cause I have who do this on the side and
speaker-1: fire me as soon as I told him. And I actually grew his
speaker-2: The company in his
speaker-1: his office space, interestingly enough. But
speaker-2: The original idea actually wasn't robotics. It was those IO components that I was telling you about. They actually were called fidgets. It was spelled P H H I E T S. Physical
speaker-1: Called f which D A D G which stands for
speaker-2: widgets. And I had the idea for those two in the morning sit up and my gosh, this is a great idea.
speaker-1: In the morning bad girl.
speaker-2: Because back then I was a programmer, but I wanted to be able to build physical didn't know how to do electrical engineering.
speaker-1: things, right? But I didn't do electricity.
speaker-2: And so the concept was, well, what if I had a USB component that would plug in and then I would have
speaker-1: You know, through
speaker-2: Code and back then I'm a dinosaur. Back then I was coding
speaker-1: And you know, V B six and V B dot net and I th
speaker-2: thought well if I could program on a computer net and then control the real world around me really
speaker-1: computer in and interactive and like art pieces or like
speaker-2: virtual reality thing where you're sitting in a chair and it's vibrating and the fans are blowing faster and you're piloting a car or a sp like that.
speaker-1: spacecraft or something like that. Or I could do like really trippy vision of like an immersive music video or something
speaker-2: visuals and stuff kind of video or
speaker-1: of that nature.
speaker-2: that's what I was trying to problem solve. And so I can't of these, you know, modular components where you can control motors like fans and vibrating motors, servos
speaker-1: And then
speaker-2: to operate things and take input from things like sensors, sensors, distance sensors, switches, light sensors, water sensors. And sort of having a modular system to build you want while programming.
speaker-1: So like pressure is lowers, you name it, right? And build whatever you that was the genesis of the idea of the
speaker-2: I found another company in Canada called
speaker-1: fidgets, which actually already made these. They had the same idea about eighteen months earlier and a day or two. And then I called them and said, Hey, do you have a a distributor in America? I convinced them to give me a thousand dollars in I'll
speaker-2: So I was crestfallen for a called them and they said they didn't dollars free stuff and I said I
speaker-1: become your distributor.
speaker-2: and so once you put all those in the store, well if you sell
speaker-1: motor controllers they gotta sell the motors, right? If you sell servo controllers, you gotta sell the servos. If you sell you gotta and then now it's like, well we need chassis on top of this, right? And then what else can we do that's fun? Right. And it just drifted robotics. I was just sort of
speaker-2: Well, Pan and Tilts and Robot just naturally drift towards robotic, pulled in that direction.
speaker-1: just from following the market.
speaker-2: And so I built the website, I started promoting this stuff. I built some forums around it. I started buying every robot I could get my hands on from iRobasmotion, you know, rec kits for robots to crust crawler back in the day to Parallax components. This is all pre Spark Fun and Adafruit and
speaker-1: Amazon days, right? When you could build a little if you will, and to survive. That was the point.
speaker-2: And after that it moved towards robotics and PC as a thing. I didn't I was literally
speaker-1: based robotics when I realized this is this is the future. What I didn't know is I'm really twenty years too early to the game, right? I wanna accept ready
speaker-2: I was trying to pitch a concept that the world wasn't
speaker-1: for yet. And so Q2 decades of cloud and
speaker-2: Plawing my way survival, trying to find out where is the market going, products through that.
speaker-1: Delivering pro that
speaker-2: So we got the original company survived on selling to highlight at home, they're making crawlers, pan and tilt. And then the arm started selling in
speaker-1: arms things like that and selling to academia. So people started using them in their robots labs
speaker-2: Par extra.
speaker-1: because we sold low cost arms.
speaker-2: Every time we move a bigger arm or a or more professional arm, it got adapted up closer and closer to like high level.
speaker-1: It had got and we started like scaling up close to research and masters and PhD level. And specifically company
speaker-2: So that's basically the first nineteen years of the
speaker-1: right there in a nutshell. And so Sanford made the aloha using our hardware. Aloha stands for low cost.
speaker-2: Then Stan Pocket uses which you know a low cost hardware. So NASA, you know, one which was robotic manipulators. And that kicked off.
speaker-1: the whole AI space and, you know, physical AI at the time it wasn't even called physical. So we were buy this new wave and
speaker-2: model training sort of adopted by
speaker-1: immediately refocused on it. So I'm guessing the product line you currently have then is if you consider the entire lifespan of the company pretty recent, isn't it? Yeah, I mean if you look at the overall twenty two year history, yes, it's very recent. So it's over the last now for physical AI. We very
speaker-2: Three years we're on our third generation. We quickly adapted off the Robotus servos servos physical. It stands for quasi-direct drive. What that does is large motors inside, and then as little
speaker-1: Does is it has
speaker-2: gear reduction as possible. So that you have torque transparency and
speaker-1: Yeah.
speaker-2: compliance. If the arm hits something, it's not going to through it. So those three aspects.
speaker-1: And then in lower costs because they're standard planetary motors. Attributes are what basically everybody in research and physical AI
speaker-2: wanted so we had to adapt to that because that was the first generation of the arms. Now third generation we're hardening the product to making a more robust controlers are toughening
speaker-1: was the first or second generation made that leap and products are more robust. We're putting in industrial level control. We are tough up things like cables, our bracket designs, even the connectors that are on there, we're moving everything away from USB. It's all Ethernet based now. Things that are necessary for
speaker-2: all the things for that. And then we've also listened to the market to figure out the mobile is physically sized be the right not too not too small. The same kind of vertical lift on it so it can shelving up and down. So a lot of people do that. And then the payloads on them.
speaker-1: What is it the market wants? So ribb, which is a manipulator we sell, is to be the right size vers not too large and not well so it can do the payloads that people are asking for. It's got a vertical reach wanna do that. product because that's too dangerous in a market where everyone needs
speaker-2: To tweak the product a little bit or adjust it some. Technology is coming out so you're literally remote six months. So to overpolish
speaker-1: Or new technologies come so quickly that you're building your product line every single right.
speaker-2: stuff and then charge people for the polished product. So we like to say our robots are a little bit ugly because they're a little bit bare brackets and bars and things, but we're finding that that's what people have functional
speaker-1: They would rather have something funny.
speaker-2: that they can deploy right now and test in the field whether actually a deployment or whether
speaker-1: Whether it's you know the deployment or it's in a in research itself because they want to add or change the compute.
speaker-2: a sensor or change a grip or or you know, swap up the camera. Everyone's gonna different. So to us the Goldilocks zone is malleable.
speaker-1: cameras everyone's got a request for something slightly different is it's still to an extent even as a finished product, right?
speaker-0: You just mentioned the rivet and that is part of the new line that you have recently introduced. Is this new product line aimed to replace the old one? Maybe you can also take us to the core products right now, rivet being one and then the arms of course and and the other major products that you have. Sure.
speaker-2: So the new line that's being rolled out as name manipulated and
speaker-1: We speak has three main operators in it.
speaker-2: they're all under the Widow X name. We actually were names. Models of the Widow X. So these the ones you have in turn. The
speaker-1: come up with different for each of them and realize that the widow X line that it's easier just to make three different type of like marketing decisions internally. The
speaker-2: main one is the Widow X Pro. And so it's similar to existing Widow X AI that's out right now, except it has servo on the shoulder. And from the ground up, it's a rebuild.
speaker-1: Yeah.
speaker-2: construction of it. So it's got a lower cost brackets in it. So we can get the stuff, new shelling design on it. It's got an all new controller in the base. It's got control
speaker-1: Ethernet.
speaker-2: on it, industrial connectors and modular grippers and a compact gripper that doesn't have the big bow ears on it anymore. So all those improvements have made. We have a version of it called the core. It has the traditional smaller motor in the shoulder that's a lower cost. That's basically aimed at a budget conscious loads.
speaker-1: Which has the traditional basically education, people that are very that don't feel like they need to move higher payload.
speaker-2: So people that want to save a little money. So those are the two bottom line ones. And then we have the heavy. And the heavy has four large motors so that entire R
speaker-1: in the base and the elbow that the entire arm is beefed up and it has a one per meter reach and so you've got you know extreme reach and you've got up to six kilometer load on that
speaker-2: kilograms of payload arm. So a lot of people want to move larger things for and a lot of that's trying to move
speaker-1: It's because people are trying to
speaker-2: into deployments. So they're trying to come out of the lab and in fact
speaker-1: actually put these things factories or warehouses or small businesses to actually perform work. And so you need more payload and reach for that arms
speaker-2: Form. So that's the three.
speaker-1: that we've made.
speaker-2: We also have something replacing the stationary. So the old Aloha stationary
speaker-1: The is this big cage
speaker-2: product that sits on top of a table and has arms pointing inward. So that's done in the day. It has four arms and four arms. They're all actually that's now the workbench. Two arms that are forward facing.
speaker-1: Arms pointing that's the original way it was that's so as two leader, two followers, but they're servo based arms. That's being replaced so that we call them and so that is two facing because the whole market has moved to forward facing arms. It's a better embodiment that everyone's training us and
speaker-2: arms on and then it's just got two handles on it and computers and your arms all together. So it's no longer you have to table and then work
speaker-1: Your computer and you can just move it around really quickly. Set up like an entire work only within the confines of that table. Now instead of having this,
speaker-2: having actual robot arms for the leaders.
speaker-1: we have what we call the glide.
speaker-2: So this solves a lot of problems. It's called stiction training.
speaker-1: in the industry because you have what when you're trying to do your training
speaker-2: So when you're using a robot arm as a leader arm, because it's heavy and it's like servos cop
speaker-1: Because got big in it, you have to have gravity c turned on so it can hold it itself in space. you're actually sort of the arm. But the problem is is the know what gravity and what's the
speaker-2: And then when you're moving it around, sort of guiding the arm doesn't know what's gonna human pushing it. So when you slow down and then pause or do really s movement fighting the servo itself, trying to
speaker-1: slow movements you start what's safe.
speaker-2: do finite, granular, dexterous tasks.
speaker-1: It's very difficult when using Gravity Comp on a servo based arm. The
speaker-2: So you'll end up overshooting
speaker-1: thing you're trying to grab, knocking things over. It's it's really hard to be articulate with it.
speaker-2: Really hard to be the day that I got really fed up with watching this, I was at two years ago and we were watching people do demos on the arms and struggling over or over
speaker-1: At ICRA people and they were struggling to move blocks around and they were knocking the stack over overshooting and knocking them around. And I was sitting there going, terrible.
speaker-2: This is
speaker-1: We have to solve this for people. You know, there's the old saying of bad data in, bad data out in programming, right? Well, in robotics, if you are training the models having difficulty learning, right? You want smooth, dextrate trained.
speaker-2: So we went through a lot of different operations.
speaker-1: We tried to use smaller motors. We tried based gravity camp. We tried to put the motors in the base and have belt drives and all sorts of crazy things we were trying to do. And we realized that there's been
speaker-2: Just no hope.
speaker-1: doing it with using servos. So we just threw it all out and from the air glass just
speaker-2: Throw ground up, redesign the entire slide.
speaker-1: to be encoder based and lightweight. It's like a featherweight is this genius design that one of my engineers came up with where they use cams and springs in the joints so that it actually does spring-based gravity comp where you need it and it it works brilliantly. So you're holding these and they're very light and so you can use them all day long. And since you don't anymore, you don't have stiction. So now you can just
speaker-2: Really smoothly move the arms in very finite and training.
speaker-1: Movements data it should be trained. So that's what we decided for learning.
speaker-2: Leader arm, since we've released them, they've been a smash hit. They're the number one seller out of clamoring for them and we're
speaker-1: of everything that we've released. People are rushing to produce as many as we can because it's jurisdiction issue for everybody. You've been using arms leaders as well the size of your fleet because you can just your followers. That is the new leader arm. If we do
speaker-2: it has solved this they're also lower costing it in the IVs they work with not just our army and so if you have a fleet and you robot as well you can now double spike glide arms and then take old leaders and turn them into a foul. So we also have that if you really must have gravity comp, you have a version of the core that is how you can that the new products because some people that
speaker-1: is still a leader, so can still purchase in the people do want it still. It covers those and then we have Rivet.
speaker-2: which is the mobile manipulator. And so that has the
speaker-1: As the
speaker-2: base is about six hundred by seven hundred millimeters. We've got eight hundred and ninety millimeters of travel on the Z axis T
speaker-1: We I believe it's eight meters of there's a sturdy
speaker-2: bar on it.
speaker-1: and it can hold the two pro arms on it as well. And so supposed to hit the sweet spot.
speaker-2: So that's the goalie lack zone of what most people have asked for. On a larger version, which will probably that it will be able to handle the headline. It is a lot. It's been a lot of work. We're also working on seven duffs.
speaker-1: in a rover and we are working come out Q1 next year that heavy arms as well. And that's the new project. We're as
speaker-0: That's a lot of product.
speaker-1: well, but I don't want to bring them up because we haven't released them yet. Yeah, I want to react quickly on the the first part of what your answer about the gravity compensation because it feels like this is like one of the blind spots that the research community has. And since you are very close to the hardware, you realize it was like a hardware issue and not something you would solve with a model. And it's fascinating because I feel like sometimes everybody right now is overly specialized into one tiny part of the process and is having a hard time taking a step back and understanding that maybe there are other limitations. And especially in hardware. I feel like hardware is kind of neglected in many ways. So I'm curious right now, ever since you ship this product, have competitors copied you? Is your way of doing things becoming the new normal?
speaker-2: So we haven't shipped the new product line.
speaker-1: Yeah, it's in production. We have the prototype run that we made that and that's what we used when we went to California to show
speaker-2: gonna answer that more of a general way is really interesting leading up technology.
speaker-1: Because it is in, you know, blue branch technology how ideas spread, right? And so that deeply we're
speaker-2: So you said that 'cause we're integrated.
speaker-1: seeing things and we we see a s a problem and we say, Let's see if we can solve this. Benefit of being here for twenty two years and being as deep in the ecosystem as that we just serve research based. We serve research, we serve data clear, we serve and we serve deployment, which anywhere from startup, right? From all of those places having here's the things we want.
speaker-2: collection or teleoperation solutions and we sort of deployment can be from a fortune one thousand company to a stock. So we're getting signals from the market all over places saying here's the problems we're having we want to solve. And so sometimes you can just glean from what everyone's telling you. Sometimes they don't even know.
speaker-1: that they have that problem, right? And they want it solved and so sort of re tea
speaker-2: So as a company you can be the
speaker-1: leaves and make something, right? But
speaker-2: A lot of times you're seeing stuff come out or solutions from other sort of an iterative design that everyone's sort of sharing amongst each other.
speaker-1: Right.
speaker-2: So when we originally made our arms robotic service loans of those one Ali button.
speaker-1: From the servos back in the day, we had clones left and right. Like you go to the bottom there would be clones of our arms being in China out there, right? Like that's normal. They would even just have our picture and
speaker-2: Sometimes they're arm and then and I would just for fun I would pop
speaker-1: Email them and say, like, Hey, take that it's not yours and sometimes they would apologize and claim an intern did it or something. and other times they would just you because you can't sue them over there, right? So that would certainly and then when
speaker-2: happened. And then when the QDD motors came out, you had a handful of companies come out with different design center around
speaker-1: designs that all started to on the basic design that you see everyone have now, right? And
speaker-2: You can argue about who was first and who borrowed from whom until the cows come home. But
speaker-1: So the truth is everyone's borrowing from each other, right? One company
speaker-2: So one company
speaker-1: comes out with a a servo layout that may be adopted because a lot of people collect data on it do that, but then they'll change the gripper and
speaker-2: So another company will they'll change for
speaker-1: they'll make the gripper a little bit better with a different, you know, rack and pinion gearing system or something else. And then another company comes out with fingers that are a little better, right? So like is people are starting to use component grippers.
speaker-2: Like right now in evolution it's happening compliant grip that are half rubber, half stiff material. And so generalists are some of the first suddenly over the place.
speaker-1: companies like Genesis are the first that came out with that and then suddenly you see it popping up all over the place, right? And so people are mimicking each other from each other. technology like this because everyone's trying to figure out you know stuff factors for things like foras it sort of land something that works everybody starts to converge in the world is the smartphone. Right? 'Cause think about before
speaker-2: How does this work? What are the best form factors things? And then you have, I think it's called the form factor convergence. When the market lands on something that way. The best example iPhone for the Apple iPhone came out, how many hundreds of different variations of in the world?
speaker-1: phones exist and then all of a sudden nailed something that really really worked and everything within years, right? Space is that once you know job that proliferate,
speaker-2: Everyone merged on it. So that's gonna happen in the robotic. Once you figure out a form factor that works for job A or job B or job C will play
speaker-1: right? And nobody owns it, right? Unless somebody races out and gets a patent, which is highly doubtful because you can patent you do something, but you can't patent that concept. It's a shared idea in the space, right?
speaker-2: So yes, we've been copied as well.
speaker-1: We've borrowed from other people's designs as well. It's the only way to basically build products and anybody who claims any difference, like being disingenuous.
speaker-0: That absolutely makes sense. With these experiences in mind, are you now one hundred percent producing in the US or how is your supply chain looking? How are you making these products?
speaker-2: Yeah, so we've been building in the US warehouse
speaker-1: day one, right? So we have two houses here in Chicago right now where we build all of our products and then we're looking at a third to move into that's the size of a football field. I strongly believe that the quality has to be
speaker-2: Strong control has to be
speaker-1: here under the same roof. So I intend to keep it that way for a long time. We can go overseas with some production we would only do
speaker-2: When scale demands that we can production do so when we can have our own people on the ground there, keeping an eye on making but until then Chicago. So parts are made all around the world, most of them are in China.
speaker-1: Just like everybody, because they are the ones that can make things that the most cost and they've actually gotten good at making things in quantity with decent quality, right? So bring them and then
speaker-2: So we still purchase like everyone else parts in and
speaker-1: build them here in house.
speaker-0: As to your product lineup, you also introduced some software parts, the S DK for example. Can you talk a little bit about this and why do you think this is important? Sure.
speaker-2: So we've had SDKs for a very long time for all of our products. So that's nothing there. Trying to push the announcement about it sound how robust specifically is sort of the new part of that. Also
speaker-1: I think us to push announcement so that people understand how cost it is for physical AI specifically.
speaker-2: have Trost and Cloud now. So that is in alpha at the moment. But we believe that that because you need full stack solutions for people, right? So in very early adopters, people will build their own Frankenstein tech stack solutions, right? And they may be just working on hard
speaker-1: Will Stein solution right be hardware but need to build player, right? They're
speaker-2: So they do the software, right? Most people applications later and then
speaker-1: have to build their own SDKs because they don't have any SDKs out.
speaker-2: there that they want. So they would build that. They've got to build their own cloud solution because there's no cloud solution on the market. They build their own encoders and filtering algorithms and yada yada yada.
speaker-1: Yeah.
speaker-2: So they people build their own. But as technology matures and you have more late stage adopters a bunch of they'd rather buy it out of a box and just work on application, right? So we also want to raise our you can do that by
speaker-1: value as a company and that
speaker-2: offering more of a stack. Right, and not ch hardware. So we think that the it's basically the next largest
speaker-1: by AI space wave of automation and humanity.
speaker-2: is what's gonna happen. So if you think about what physical AI is, it is the merger industrial age.
speaker-1: of the induction age and the information age together, right?
speaker-2: It is physical labor and intellectual labor combined. And so modern robotics are bolted from the factory floor. They're eyeballing reactive.
speaker-1: training basically getting unballs, they're becoming and they can be trained like humans to do human like right change the world, right? Change of humanity, physically changed the face of the world.
speaker-2: Tasks. And if you think about how much the industrial changed the face
speaker-1: I brought a whole new phase in humanity, right? And then the information yet again, right? So we invented the computer.
speaker-2: And so intellectual automation.
speaker-1: now AI, right? And we're seeing a wave of information that we never even thought would happen in our lifetimes, right? And that if you have the AI, the brain, doing intellectual and physically being able to move the world. So it's going to change the world yet a third time
speaker-2: times. Put these together, then you can have actual tasks around it. World third time.
speaker-1: in ways we can barely even conceive of. It's just in the in the sci-fi movies that we can imagine what this is going to look like, right?
speaker-2: So the idea behind Trostin is that we don't just want to be a hardware, have the full serve.
speaker-1: company we want to have solution that can serve
speaker-2: All we call them prospectors, right? People that are looking for We don't really want to be a prospect or a picks and axes and shovel company. Sell the components and hardware to the
speaker-1: ourselves. We are so we want to companies that are doing this.
speaker-2: to get them up and running, they need more than just the hardware. They also need the SDK, which is attached to a cloud that has the tools and training. That's a full product. That's a full solution. And that is what the market needs. So we have been in a unique position more hardware in more places than company who's in this space. Here the longest. And we decided to add the software layers to that so that we have a robust product that serves the needs of the market all.
speaker-1: Serves needs as it evolves.
speaker-2: Right. So now the concept is when you buy an arm, you take it out of the box, you can log into Truss and Cloud, you can start collecting your data on the same
speaker-1: Day.
speaker-2: collect the data for whatever need it is that you're trying to learn. You can pick the model that you want to use for
speaker-1: It is
speaker-2: people that might be using a model that's already out there, an open source model like Pi. They might be actually the model builder themselves who want to collect data and keep improving them.
speaker-1: It could be a company for the
speaker-2: It's forking a model and tuning specifically themselves, all the above, right? But what they want is as quickly as possible. They don't want to buy the hardware and then figure out where do I store my data on what cloud. Now I gotta put my software stack together for my teleop. I've got to put my software in intake of the data, processing. Like it takes time to build
speaker-1: that does my data, my labeling and cleaning, my inferencing.
speaker-2: all that stuff. And so we've added that to our hardware. So now if you purchase a trust or you can do all of that ecosystem bringing it up.
speaker-1: arm you can from bringing to training to cleaning and then go believe that we're that value
speaker-2: back down onto the robot itself and go into deployment. And that's how we believe we're making a company that can go the distance and has
speaker-1: in this space
speaker-2: and we believe you have to own that entire stack to do.
speaker-1: Do so.
speaker-0: It really feels like your company has been kind of looking what's going on and now is the time you're like activating the boost with all these interesting learnings and experiences in the structure that you've built up over this long time. And now kinda all falls into place.
speaker-2: say you're absolutely right. Like so our first investment round now.
speaker-1: we're taking outside investment in and it's an angel raise with eyes on growing especially have to move to survive.
speaker-2: fast as we can because you fast arrive at all in this space, but this is the time to
speaker-1: basically pounce on the market and build the companies that'll last organically grow. Our belief is that
speaker-2: And we can't organically grow through this phase.
speaker-1: it needs investment and money to grow this. And so we've chosen to basically change the way we do things.
speaker-0: Yeah, it's interesting because I already had a segment about you in one of our podcasts and then I was researching the investments and I was wondering like they are completely bootstrapped. Like and then you look at all these other companies and like why are they not raising? They must be raising. What the hell? And it kind of makes sense now. You also brought a CEO on, Anis Shah. How did that happen? So
speaker-2: Part of my realizing that we need to grow faster and bring on investment was thinking about team building and that we needed to bring more sea level. At that point it was just little over
speaker-1: people into the company because up until it was just me running the whole company. And it starts growing
speaker-2: beyond a point where one person can do it or in the technical design product and
speaker-1: Do it all, right? And my strengths are nine of the products.
speaker-2: then watching the market and reading the tea leaves of the market. Those are the two areas that I've been honed to do.
speaker-1: do for twenty two years of experience, right? We kinda think about like wax on wax. Like you don't even realize learning these skills along the way.
speaker-2: Off the old karate crud metaphor about realize you're learning skills, right? So being a small bootstrap company and learning to survive in bleeding edge markets like robotics, sort of watch the market, see, you know, inflection point is that what's the next line of products that it's asking for, what different pieces of software need to be bundled together to serve that market, right? That's like what Tronson does essentially. And if you look at physical AI now. It's just more of the same. We used to do it for hobbyists. I'm gonna grab
speaker-1: Where it would be
speaker-2: you these servos, this controller, this battery pack, these set of brackets, and this example code. And that's enough to get you up and run your desktop, for instance, starting to work with Raw Ross packages in right this evolution of just
speaker-1: With a robot arm on your so we started adding robot. It's just a
speaker-2: layering in more and more tech as it evolved. So in the physical AI space we're really it's just now
speaker-1: And they're doing the same thing.
speaker-2: it's a cloud that has modern AI models in it that can be
speaker-1: My trained up it just gets bigger and bigger, right?
speaker-2: on the hardware. So as the company gets larger, we need more people and to run the different departments.
speaker-1: Right.
speaker-2: So we brought Anish on. So Anish, he's got a very storied career as well. You should probably have him on his own podcast. So he has worked in startups, he's worked at different businesses, different sizes. He had a stint as a stand-up comedian
speaker-1: Comedian.
speaker-2: for seven years. He worked at McKinsey and he's worked at Moderna during the vaccines. He's kind of done all sorts And he's had a passion for AI and robotics for years. He's been recently wondering if he could get into this utilizing the skill set he has. Well, it turns out he's in our backyard. He's been a couple towns over from us. And so he said, Yeah, I'm this guy who's interested in this. Here's
speaker-1: And once he's doing research he discovered so he just messaged me and said, Hey
speaker-2: my skill set. You know, let's have lunch. And I thought, it sounds interesting, sure. And so we went and had lunch and that turned into a four
speaker-1: Four.
speaker-2: hour conversation and showing him the shop and discussing concepts and where we went.
speaker-1: go with and as we get married together. Meeting to when he joined so we basically got to know
speaker-2: This he likes to say we fell in love with each other and decided to go and work on this company. So from that first meeting the team, it was nine months to know each other more and worked on some projects together. And
speaker-1: And it just became really clear to me
speaker-2: that the strengths of strategic has at this
speaker-1: thinking that he has is exactly what the company needed this time.
speaker-2: right. So we brought him on as COO for a while and recently decided that it's time for him to C EO role. So I could constantly
speaker-1: Then it was time to move into the C concentrate on the C T O roll. Because
speaker-2: so this dynamic works really well because he can work on the raises, work on strategic partnerships, work on growth of the company while get to
speaker-1: Well I guess
speaker-2: focus on building the actual products inside the company. And so that's building the C so
speaker-1: Yeah.
speaker-2: that as we bring investments and grow from there there's more people at the helm trying.
speaker-1: There they're trying to you know, hold the reins of what's basically a bucking
speaker-2: You know.
speaker-1: rhino, if you will.
speaker-0: And speaking of partnerships, are there different extents of how you work with people together? Do you have like specific partnerships or is it more like you buy our product, we have a certain type of warranty and service that we provide after it, or is there some differences there? So
speaker-2: Yeah. So you always want to sell your core products as much as you can and then alter off of your core products. So we're not like a pure consulting shop where we come in and say
speaker-1: And what kind of robot do you need when we build one from scratch?
speaker-2: What I like to say is the invention isn't just the robot kits who make the invention the company that can react and adapt to rapidly. Like that's the challenge. And so
speaker-1: to a changing market rather than challenge, right?
speaker-2: there's gonna be an embodiment war or
speaker-1: So like if I
speaker-2: Darwinian war amongst all the different form people are gonna dream up over the next ten years and everyone needs to try different
speaker-1: factors that people are going to dreamers on things to see what works for different use cases, right?
speaker-2: So what we go back to like the prospectors, which are people that are trying to do deployments and and find applications for
speaker-1: to this new thing called physical AI. Each of them wants something
speaker-2: Each of them is slightly different, right? And so we're building a company that can react to that. And if you want fifty robots with a different end effector on it, we rivet with a taller Z access and more payload. So we have to enlarge the base a little bit.
speaker-1: do that. If you need maybe then we'll work with you to do that. And those are happening now to look at deployments with fifty a hundred units.
speaker-2: sorts of conversations are now as people are starting to 100, 200 units, right? The cloud part of the services feeds in that where people need different tooling in the clouds. So we'll work with and the different technology layers necessary to achieve goals that and then that way the the market is informing
speaker-1: achieve your goal there, whether you're a model maker or an end user.
speaker-2: us on how to evolve the products. So we look for a sweet spot with partners and clients who are near where we already are. But just need to slightly improve it a little bit more in one way or another. Improve the product. And then we hope to sell that same product to more people. Right. So if somebody says we need bigger arms on a mole manipulator, we'll
speaker-1: Well,
speaker-2: build that and sell it to all the customers out there, right? So we're kind of looking to all swim in the same direction because if we try and that don't overlap in too many areas, that's a good to come. You'll get spread too thin, keep
speaker-1: make too many things are a good way to die as a company, right? Because if then your resources just can't
speaker-2: up with it. So you have to be selective about it and say no nine times out of ten and they're directly. But we're
speaker-1: And then yes to the ones that make sense because they're heading in the same direction you are. But
speaker-2: open to all sorts of different types of clients. We have clients that are startups right now that are trying to deploy
speaker-1: Or deploy and
speaker-2: to medical deployments or grocery stores
speaker-1: Plus
speaker-2: for picking things off the shelf or packing house as well as like Fortune One Hundred companies that are trying to protection put into robotics so that they become part of the
speaker-1: trying to do data collection or get their products, they can become physical AI game.
speaker-2: And so the challenge is finding the right partners that of
speaker-1: Yeah.
speaker-2: you who want maybe something a little bit different that your competitors can't, but not so different than
speaker-1: That you can answer competitors, that it changes your business model. I don't know. Did I answer your question?
speaker-2: Yeah.
speaker-0: Absolutely. I think if you spend time and obviously work with a lot of clients, you get almost like a sixth sense about what is needed, where the deployments are going and obviously you can use that for your research. Do you have a specific research team or was your team setup looking like right now? So a lot of it is just
speaker-2: Feel at the moment. So the team isn't large enough to have an entire division that's taking all signals and then reports that I hope that we'll
speaker-1: There's no emailing us and talking to us and
speaker-2: That should go through a matrix to determine what it is that you're going to create next. So no shortage of signals coming that are calling us to us at conference. Can you do
speaker-1: saying
speaker-2: this? Can you do that? Through just hundreds of conversations, you're like you said, where you get that
speaker-1: You're getting a feel like sense of this is where it's going.
speaker-2: that's the broad brush strokes are are easier to figure out. I call it the Little Birds program. So if you guys ever watch Game of Thrones, one of the characters talked about it.
speaker-1: Little birds which are vessels that would feed the information. We basically
speaker-2: were his spies that he had and all the different castles back. And so I sort of stole that little program. So
speaker-1: got our ears to the grindstone trying to hear what is it that people want in that literally does that twenty four market going, what are we developing next? And
speaker-2: need and I do want a division or seven and so there is the smart starts preparing for it like twelve minutes out, right? Right now we're very reactive. And that's
speaker-1: We're basically being pulled by where the latest thing is that people are asking for.
speaker-2: perfectly natural. And and as you get larger start to level off for
speaker-1: in a brand new market that's bleeding edge and then things start to level off when you start to hear
speaker-2: more and more of just
speaker-1: we need a bigger arm, right? Being pointing you in just the bigger arm direction. Or
speaker-2: Or
speaker-1: it's become clear that these compliant grippers are what everyone's gonna use. And so you have to just
speaker-2: Listen to that. But the normal process in a more mature large company is this sort of product development cycle that can be twelve or twenty-four.
speaker-1: goes through a four month cycle. If you think about something mature like refrigerators or automobiles or something, the
speaker-2: Their time from concept to
speaker-1: creating it is much, much longer we live in. There was also a common problem that we have been not directly running into, but with that we have observed in robotics that there are many companies now that deployed humanoids, teleoperated humanoids to do basic tasks like cleaning, cooking, pick and place into real homes in order to acquire teleop data at scale. The main issue is that the people who teleoperate these robots, they basically wear a VR headset and a handheld controller and they cannot do much more than forty five minutes to one hour at a time because it is extremely exhausting to wear the VR headset. It's heavy, it will make you dizzy and so on. So we have been thinking lately about is there maybe a better way to provide a telehop interface as well.
speaker-2: Yeah, a lot of people have
speaker-1: started doing egocentric data collection, right? Where you're wearing a headband that has cameras on it and then you're either just using your bare hands which are being recorded or you might be using a Umi like device where
speaker-2: Yeah.
speaker-1: you have ripper that you're holding and it's triggered or you're actually wearing it like a glove and you can
speaker-2: pinch this way. So you're seeing people try all sorts of different and some of those are getting down,
speaker-1: things and and getting the cost
speaker-2: right? So that you're not buying four arms in order to collection.
speaker-1: Order to do the data collecting.
speaker-2: The debate is is that data good enough to use yet or not, right? What you lack when you use that
speaker-1: Because we stuff is the actual telemetry data, right? From your matics.
speaker-2: kinematics. And so just egocentric alone, you know, for the feedback we've got from most researchers, it's like you'll never
speaker-1: Is not enough.
speaker-2: You have to have someone opp physical robot arms the task. And so it might be that you'll do ninety percent or something can be done from egocentric action, but then the last mile, the last ten percent you have to have the can
speaker-1: The first can be done vision data collecting percent. You make the data of the arm actually physically moving or the no
speaker-2: Model just hasn't
speaker-1: idea what it's being asked to do, right? Remains to be seen, but this is the research learning as it goes along, right? Now
speaker-2: And what that blend is, you know, market learning. So the cost is coming that we're
speaker-1: it's You know, like what we call the workbench light, just
speaker-2: That's
speaker-1: two arms, the three cameras on a metal plate, your power
speaker-2: Late in your
speaker-1: solution, a networking solution solved for you, dramatically brings down the cost from what used to be a big stationary box that was just, you know, in your resource. And then if you choose ego trick to add to that data, then those headsets can be anywhere from a five
speaker-2: Five hundred dollar phone.
speaker-1: to a two or three thousand dollar headset, right? So that's bringing that cost down.
speaker-2: And the mobil up of what you can collect. and the other thing is customers they're hiring in the SDK and the cloud and all that.
speaker-1: We service this market, right? So people buy our and
speaker-2: Our kits.
speaker-1: then can use the cloud to upload the data and lay and and
speaker-2: and clean it, label it.
speaker-1: do everything they need to. The hope is that we'll eventually have a full marketplace where we even help them resell that data and isn't the egress of data being moved around.
speaker-2: And so there is and all of that as well. That helps lower the cost.
speaker-1: And and make it quicker and easier for these companies to get spun up, right? 'Cause we're solution like a franchise at that point where it's here's
speaker-2: selling them the whole it's almost like it's here's the hardware, here's the software you need, some tool. So
speaker-1: don't we have the data collection, we have the data cleaning tools, we even help you sell it. So people can think of different areas where they think they can add value to the by collecting data that other people have instance or service customers in different ways.
speaker-2: add value to the system but haven't for more empowered to do so because the technology's just
speaker-1: evolved right whereas for the last few years they would have had to figure out that whole data stack themselves in terms of like cleaning around we've seen different ways that well seen offering to come clean your free if you just let somebody come in with an egocentric headset
speaker-2: themselves as well. In terms of collecting complex data in a house or something that have been done as well, we've seen like the gig economy offer their house for centric assets.
speaker-1: which is a really interesting way to approach it right emails from data
speaker-2: I daily get even firms all over the world that are just putting headsets on top of factory workers and they're just
speaker-1: Just pointing out just trying to resell the data.
speaker-2: like there's this mass rush of people trying to get rich.
speaker-1: on this new thing that they've discovered, right? And that's gonna drive the prices down because the competition's gonna make it cheaper and cheaper and cheaper on a data cloud.
speaker-2: collection. So obviously everyone knows this with LLMs, you had the entire knowledge of the human internet crawl between your
speaker-1: race on the internet. And so you could just go crawl that and trade
speaker-2: LLMs.
speaker-1: when it comes to robotics, you actually need physical world data and that doesn't really exist yet. So we're just figuring out how to do that. When you bring up humanoids, you bring up a whole different can of which
speaker-2: Worms which
speaker-1: is is this even the right form factor to be collecting data? Right? Is there even a market for this? And if you ask companies like it's
speaker-2: Myself.
speaker-1: yeah, the pragmatic answer is absolutely not. This is it's way too early for something like this. And no one's gonna be buying humanoids and putting them in in homes anytime soon or even
speaker-2: And every
speaker-1: And there's a hype bubble around it that's going to pop. Everyone inside the industry believes this and knows this. Even people who work at the humanoid companies believe this. But there's
speaker-2: They know that
speaker-1: no market for their product right now. And so when
speaker-2: the investment dollars dry up, all those things are gonna pop.
speaker-1: Right.
speaker-2: Should we be collecting
speaker-1: You know, data on on what it is to clean a house?
speaker-2: Or something.
speaker-1: Probably, yeah, 'cause we still want to train how to do that. It just must be a humanoid that does it. You know.
speaker-2: That may not be when you want to think about pragmatic ways to roll out this technology, it'll start in industrial space. Right.
speaker-1: Like you'll That's bolted to a podium leaning over a a sink or a torso, for instance, you wouldn't have the legs on it. That's just a waste of technology.
speaker-2: If you want to have arms that are assembling something on a factory floor or an assembly line, it makes sense to dangle those arms on the gantry system.
speaker-1: Right and have them slide going up and down and rotating, and remove at the edge of a table.
speaker-2: So data collection will probably move in the direction of the way this stuff deployed. But currently the only people that can afford to go
speaker-1: do mass data collection are these companies that have billions of dollars in investment.
speaker-2: Well,
speaker-1: At the moment, those are humanoid companies, right? So you're hearing the buzz of them.
speaker-2: Everyone else is doing data collection.
speaker-1: Right. And then everyone's doing egocentric action because it's cheaper and they're trying to resell it. And so you're just seeing kind of just different areas to figure out what works and what doesn't work.
speaker-2: Yeah.
speaker-0: I just try to wrap my head around these numbers that I'm reading with regards to humanoid production. Why are they planning to produce so many if they kind of know, well, we're just producing them for I don't know, for who are they producing them?
speaker-1: That's very good question. I'd like to know the answer myself because I don't think anyone's gonna buy them right now or because the company's paid to put the robots in those places, right? See people please
speaker-2: Any of those deployments that are out there themselves pay prices. They're asking to please try this product out for free, and those test cases don't last very long, right? Take it back. It doesn't it would make sense some
speaker-1: Then they're like back, don't do anything, right?
speaker-2: more market research and found out if people actually wanted to buy the product first. But to date, I don't know anybody who actually wants to buy a humanoid to deploy a useful scenario be an ROI.
speaker-0: Right.
speaker-2: Why are billions of dollars going into this industry factories to build tens of thousands of units of a product that no has raised their hand and said they want to purchase yet? It's backward. That's why a lot of people in the industry
speaker-1: The industry believed that when the bubble
speaker-2: pops all of this stuff will be chopped up for parts, right? So it's almost like human rights belong in research.
speaker-1: Long in research.
speaker-2: We should be trying to solve these problems in research. And it's like using the goal of going to the moon, we can learn how to make
speaker-1: It's like choosing so that we boosters and computers, right? To do something very hard.
speaker-2: Very challenging and very hard. It's good to have goals, right? I like robots trying to play soccer. Awesome, because it's like here's a goal that's really hard to solve it. That have excelled really well at making robots balance and difficult of sitting karate
speaker-1: And let's you know, and you have companies like Unitry making can be in the most situations. They can do
speaker-2: and ballet and pull buses and they do all these.
speaker-1: Yeah.
speaker-2: But it's a research challenge at the moment. There is a humanoid to pull a bus,
speaker-1: Isn't a use case for
speaker-2: right? And they're not going to be in homes looking at the things anytime soon.
speaker-1: Cool plus A soon, right?
speaker-2: So there's kind of a wrong application of the technology at the moment. Should we be trying to do these hard challenges? Absolutely. Should we be trying to figure out how to make better actuators, how to make better algorithms? All of these are
speaker-1: Walking gates things, sure.
speaker-2: we should. Should we be telling the world that we're gonna make a factory with ten thousand of them buy them? Probably not. And so a lot of people are gonna lose their ass too
speaker-1: Because people are gonna buy and so lose their passes in these investments. 'Cause it's
speaker-2: early for this technology to be deployed. Someday We'll have legged robots doing something useful
speaker-1: And humanoids doing use.
speaker-2: for us. So the argument now is I make legs when we flat,
speaker-1: Why made the world?
speaker-2: right? So legs are useful because robotics is the art of reverse engineering nature, I like to say.
speaker-1: We like to say it so much we marked that phrase.
speaker-2: So if you look at anything we try and develop in robotics, we're basically looking at reverse and reverse engineering. So for instance, we've used LIDAR and sonar to do navigation for decades now. And it's almost like we're cheating.
speaker-1: Look at nature.
speaker-2: Nature has a few animals that use that. You know, we have bats, we have dolphins, but by and large nature never efficient and it works in almost right so when you look at our robots navigating now we're starting to use vision more than we're using
speaker-1: using LiDARs. Like we'll probably always use LiDARs since it's a great tech
speaker-2: technology that supports it, but vision is getting more and more important, right? And if physical layout
speaker-1: If live in eye, we're putting cameras on manipulators so we can see what we're doing. Not just talking about like slam navigation in an open space. Of us mimicking nature, nature do this and let's start mimicking it. How the brain thinks. So we're starting to make intelligence by
speaker-2: So that's an instance of learning how to do AI is us reverse engineering artificial intelligence.
speaker-1: learning how to do that.
speaker-2: We have yet to figure out how to make skin, right? Very difficult, yet to make a muscle, an elastic muscle that can electricity sort of compliant,
speaker-1: Right. There's a lot of researchers on that. It's a very difficult and we've yet must restrict with electricity and be sort of
speaker-2: right? So if you look at locomotion, wheels,
speaker-1: We use
speaker-2: because we've made the world flat and they're incredibly efficient. And it's because the main motor we know how to make rotates 60
speaker-1: Three six.
speaker-2: and it's with heavy magnets and and right.
speaker-1: Wires.
speaker-2: Something like a car makes a lot of sense with a current internal combustion engine or an electric motor driving on a flat surface. So if you have robotics, we have MRs, it's already a solved problem. They all use
speaker-1: in a warehouse and use wheels, right?
speaker-2: so why would we use legs inside a flat environment? The wrong
speaker-1: It's the
speaker-2: tool for the job, right? If you're gonna make a robot that needs to
speaker-1: It's
speaker-2: scamper through the forest, now you don't want wheels. Now
speaker-1: And
speaker-2: you want to reverse engineer nature.
speaker-1: nature did it with legs.
speaker-2: And it makes sense there, right? So the argument is we'll have sta
speaker-1: is how you handle tears maybe? And I don't know what the answer to that'll be. It might be some sort of wheel that can make its way upstairs or
speaker-2: Or not.
speaker-1: they might be arguing that you have to have legs for that. Who who knows?
speaker-2: But as long as you're staying in a flat.
speaker-1: Area, wheels get the job done, right? People wondering why we're making humanoid legged humanoid robots other than love
speaker-2: Then for the
speaker-1: of making them. No, this is always super insightful. I have a question back on data. You mentioned egocentric video data and of course data robot data as well. We've seen recently approaches where UMI data becomes more and more important and becomes like an entire step of training between egocentric data and and tele op data. Is it something that you also observe through what your partners might be asking you or not? And what's your take on it overall?
speaker-2: Yeah, yeah. So that's a third third way of doing it. And we have what we'll call the Trumi, Tross and Robotic Trumi product
speaker-1: So we actually have a problem.
speaker-2: we sell for that. And that just has the GoPros on it. And the next version of that eccentric
speaker-1: We're gonna have our own egos.
speaker-2: headband with matching cameras that the hand grip. So do the gen two of the True synced so you
speaker-1: And then it'll all be data so that
speaker-2: have the hardware synced together. So then people can go collect that data when they want that. And so that's us reacting to the market saying
speaker-1: saying that they want that data and that if it's very valuable to have
speaker-2: So we're
speaker-1: some early prototypes of that being made right now. Well, we've seen a lot of value in that. It's gonna be a bree.
speaker-2: So that is us reacting to that as well. So to answer your question, yeah. And I think it's the blend of all three.
speaker-1: Like I don't think anyone is gonna really
speaker-2: truly win in the end. And I think that eventually AI will get so smart that you can just put cameras sort of around somebody they may not even have to wear them on their head anymore. And it'll just go, Yeah, I understand that task now. But
speaker-1: I understand
speaker-2: once the AI is getting smart enough to operate any type of embodiment or to look at something
speaker-1: Something
speaker-2: from afar and then f put themselves in its reverse sort of engineer how they're doing it, which is how we it
speaker-1: Either way.
speaker-2: right. So you have you can watch somebody how to do a task and then jump in and start doing it yourself, right?
speaker-1: Do attack start doing that smart ability as well.
speaker-2: And robots will eventually get that ability as well. So I think most people we talk three
speaker-1: That we have to think it'll be a blend of all three.
speaker-2: and
speaker-0: Back to deployments, what do people that nail deployments do write if it works?
speaker-2: Focus on a small task thing.
speaker-1: One thing.
speaker-2: So this idea that you're gonna have additional models that has so far proven to be very difficult and hard to so the people that are having the most successful tasks and just trying to perfect that task, collecting
speaker-1: foundational model can do everything is to hold off most are narrowing down on very specific tasks tasks.
speaker-2: data around that task, sticking to one embodiment and that will get really close to a deployment that is actually useful.
speaker-1: That's where we're seeing people release our OIs to like useful. And we still have
speaker-2: You know, the world at large is still trying to figure out what's gonna be
speaker-1: And I hope they're using trust.
speaker-2: of the Roomba of physical AI super awesome case that just takes off. And somebody's gonna find it and hardware when they do, but everybody's hunting now and figure that
speaker-1: Right. I'm trying to figure out.
speaker-2: out. But narrowing down your scope is absolutely the most important is can you do it with simple grasping techniques?
speaker-1: Minks.
speaker-2: If you can do it with a parallel gripper and a parallel gripper only without having sensing ears yet, even better.
speaker-1: Great. And if you can do it without being on the finger, right?
speaker-2: So it's gotta be a simplified task that uses a parallel gripper that's very focused. So our friends over at Yandu have deployment that they have, which is package boxes into
speaker-1: released small bugs into plastic bags and
speaker-2: call a polybagging machine. And so there's a use case where it fits all of those criteria. And so they're already offering solutions purchasing.
speaker-1: So you don't believe at all in physical intelligence approach of cross embodiment?
speaker-2: I think it's a long way off. It's sort of shooting for the moon right out of the
speaker-1: Date.
speaker-2: and it's difficult to do. And it's not just Pi that's doing it to people
speaker-1: There's a bunch of
speaker-2: that have that sort of goal, but
speaker-1: You know, we've seen
speaker-2: so far that it's it's a hard problem to solve. And so if you narrow it down, it's easier to get higher and higher success rates. And if you asked any of those companies, they'd tell you the same thing, right? But the dream is that it can do anything under
speaker-1: That anything.
speaker-2: the sun. What'll be interesting is to see how the ecosystem evolves
speaker-1: Bulbs?
speaker-2: and it might be that Models can be sold to other companies that work in and they hone
speaker-1: And then
speaker-2: the model. And so it's like, well, this one's really good at digging holes. This one's good at cleaning toilets. This one's good at stuffing. And so
speaker-1: And
speaker-2: you have what I call prospectors, right? Who go out and take a model and then learn how to do something specific with it and have success. And then the model is learning from all of those people. the sun, but also if you're deploying a robot in let's say a grocery store and it's supposed to stock in shelves and package people's orders off the shelves, for instance. Why would that robot toilet like a whole right? So you wouldn't have its brain holding
speaker-1: Need to know how to clean a toilet or day hole.
speaker-2: all onto all of that information when it's focused on this one task. You know, so you can separate that stuff off because it's more efficient in terms of how much compute are you using this problem, right?
speaker-1: For it to think through the right.
speaker-2: So will that be at the model level or will that be at a deployment level of somebody layer of a model? That remains.
speaker-1: be a somebody who is doing an application to be seen, right?
speaker-2: So the model maker is out of the gate kind of shooting for the moon, but here we are two years later. None of them can. So you're
speaker-1: You see people.
speaker-2: starting to focus in on more niche.
speaker-1: solutions and we seem better
speaker-2: Accessory etc.
speaker-0: If you would start something right now, maybe tackle a certain problem that you're seeing, any ideas on that?
speaker-2: Well, if I knew that answer, I would be tackling it, right? So that's why we're serving all the people that are out there prospecting and who's successful. It's probably one of the number one questions I'm asked is where do I if I had the answer my billion dollars.
speaker-1: Or do you think physical will succeed first? And the answer company would already be worth a because I would be tackling that.
speaker-2: I don't have that answer. So that's why I'm trying to sell my shovels to as many prospectors.
speaker-1: as possible so they can go find the answers worldwide problem.
speaker-2: Because it is sort of like a world we need to attack it by just trial and error, trial and error, trial and error until we figure out where is this technology useful. And all new technologies go through that phase, right? When we steam engine, they didn't know what to do with it. They had to do
speaker-1: First came out with the
speaker-2: trial and error until they figured out, we can run a loom or when we figured out that the water wheel turn a stream into energy.
speaker-1: was a great way to energy. It's like what do we do with this until
speaker-2: They figured out, well, if we put this cog on it, it can lift this big heavy hammer that that smashes apple all day or something, right? Every new technology goes through that phase and physical AI is now. I think that the opportunities, a big one, and you're already seeing some success here is anything that automates,
speaker-1: As to a feeding humans?
speaker-2: right? So anything that automates our growing of crops, our processing our creating our meals, our serving of our meals and our cleaning and washing of our dishes afterwards. I think there's great opportunities in places like that.
speaker-1: Yeah.
speaker-2: Because that's stuff that's made by the hundreds of millions. So I'm curious to see if that happens. I've mentioned grocery stores a few times. I know that people are attacking that space trying to solve food delivery, but that's part of this right. And then in factories
speaker-1: We're looking for things in the
speaker-2: that traditional robotics can't do that humans are still doing. So like machine tending to
speaker-1: is a big one, right?
speaker-2: So how do you put a pull a part out of a machine, next part in, make sure it's situated right, and then turn the machine back on and next
speaker-1: Then put the next one and run them.
speaker-2: part? A lot of people are trying to work on that sort of areas where
speaker-1: Or trying to sort of thing. Things are floppy and soft.
speaker-2: traditional robots have a hard time dealing with, so anything that's in t instance a popular example
speaker-1: Textiles for instance.
speaker-2: That is something a traditional robot can't still have to do So you wanna find things like that because physical AI is not meant to replace traditional robotics. Traditional
speaker-1: And then you've got to put everything together and close, right? Can't do and a human do, and it's a mind number job that you do all day long, right?
speaker-2: Very good at what it does, right? So if you have a robot you want it to be extremely accurate, I can do something like put a rivet in a car
speaker-1: But you want to be extremely active to some card.
speaker-2: door in the same spot all day long, very
speaker-1: Very, very quickly traditional robots answer that need already. And so physical AI set that it's the work
speaker-2: is not here to upset work in concert with that, doing the next layer it's not good at by dealing with chaotic environments and things flexible or flex out of a box.
speaker-1: things that are floppy or might show up being poor where you need to do some thinking to to do it, right? I wish
speaker-2: I wish
speaker-1: I knew exactly where those answers were because I would just go
speaker-2: Pick some partners and
speaker-1: right after them and solve them. With as many partners as we can and learning as we can figure out
speaker-2: working is to go to sort of figure out what those are. But I think we're gonna have our first few breakouts in the next twelve months. That's my guess. We're all waiting for the models to get better as well. Now we're starting to see some throughs just kind of mind blowing and how good somebody at learning to do something much, much less training them. We still
speaker-1: And some breakthroughs that are these models are getting something within a few minutes with less data.
speaker-2: have to sort of tie those together to the hardware and then find somebody put the money in to go try and deploy it. business model around selling it as well. Like there's a lot of things to sell even
speaker-1: Lot of things to solve in business.
speaker-2: around the fact of can the robot do it? Right? So there are a sales team the customers to invest in buying
speaker-1: Right. So you've got to find out is there that can go find the groups and convince them to buying the product.
speaker-2: is there a deployment team that can go put it in place? Everyone has a slightly different setup in their environment that's using this robot. how are you doing maintenance robots? Maintenance what are the logistics financials
speaker-1: you know it's using maintenance on these so you have to be able to put together a maintenance plan that they're willing to pay for whatistics and the financial
speaker-2: solutions to all of these right.
speaker-1: of these
speaker-2: How do you spin them up in volume that all price point works? Because it's really expensive. And so how does a company purchase that and
speaker-1: Robots can be deployed at a price point that is, let's be honest, this shit is really right now, right? Robot.
speaker-2: resell it to someone else as a solution that's lower cost than the actual workers doing it at the moment?
speaker-1: wages of a work or doing right.
speaker-2: The only time a company will adopt a new technology
speaker-1: technology is when it saves them money. They don't do it for the love of the game. They will continue to use whatever the lowest cost is until it's too painful and the new technology,
speaker-2: option is helpful, then they will adopt new technology.
speaker-1: right? Clearly. Now I have a question about the company overall when you mentioned a funding round, I don't know how much you can say about that, about how much revenue you have been making before raising, like bootstrapping, and how much you are aiming for after raising capital money capital money.
speaker-2: Venture Yeah, we're not public with those numbers, so it is to honestly,
speaker-1: To raise the revenue, obviously.
speaker-2: but right now we're doing a small raise to finish sort of finish the
speaker-1: So it's the finish
speaker-2: product line, start scaling the production numbers of it, and then finish rolling out the cloud, right? The cloud at
speaker-1: It's just like in super alpha.
speaker-2: the moment, and then prove that we can convince people to use our cloud with
speaker-1: with our hardware, right, which is something we have to do. The
speaker-2: And that they'll allow us to collect
speaker-1: data and train the data in certain instances as well. Right. So we have to
speaker-2: prove that model out. And so the first round, which is a small round, I'll say it's much less than dollars,
speaker-1: Ten million dollars.
speaker-2: is the goal for that, right? So to prove it. And then in that process of us proving that and showing we have successfully deployed an entire technology stack with this latest hardware and we're doing deployments where we're selling batches of 50 and 100 of these, then we can go and make our case to do a much larger round at a larger evaluation,
speaker-1: Value.
speaker-2: right? Because what we don't want to do is do too large of a raise right now and sell off too much equity before it's necessary.
speaker-1: So you don't
speaker-2: And so it's sort of like raise, build value, raise and try and pace this correctly. You don't like burn up your company in the meantime,
speaker-1: like
speaker-2: right? And over promise and underdeliver sort of scenario and have a down round, right? There's a tricky balance of that when trying to grow a company to decide when, right? Everyone is we're insane bootstrapped as long as we have and I raised debt to get there, right? So it's
speaker-1: because we b as long as and we probably are right. So it's me risking
speaker-2: a lot of debt in order to get there. But it was because we're trying to build a a certain amount of value before we
speaker-1: go and do a raise, right?
speaker-2: So these are the things deciding in a company as you go. If we were a brand new company that had never done anything before and we were just two Stanford grads, then you do the biggest
speaker-1: Do a
speaker-2: raise you possibly can out of the gate, you give off a big chunk of your company, you take a shot at it, and you see how it goes, right? But we already have revenue. We're already selling, right? And so the way that people look at your business is completely different. Pre revenue
speaker-1: Right.
speaker-2: and post revenue are completely different valuations. And oddly enough, you get punched in the teeth for having revenue. People
speaker-1: Avenue, right?
speaker-2: just don't value you the same way as when it's Which we think crazy because we've already built the product, we already have a strong international brand, we're already working with all of the top fortune ones. We already
speaker-1: We're
speaker-2: have built out the teams and we already have a tech stack built, right? So that should in theory.
speaker-1: In order to get the valuation to why
speaker-0: balance between all of them. Yeah, because people have the reference point when there is some existing transaction going on and as we all know, you know, venture capital likes to dream. They w want to dream of the big numbers.
speaker-2: Yeah, a question always is is well where
speaker-1: You making more money, right? So why aren't you already making all this money? And it's like, well, if we were, we wouldn't be asking you for an investment, right? Because we'd already be at that place, right? We need investment to get some. the unique company we are, we've been around for twenty two years, the answer is well, we're
speaker-2: Well,
speaker-1: waiting for AI to show up.
speaker-2: like AI is only three years old. Basically the first nineteen years of of the company you kind of just have to throw out the window. Company's hobby and education kits.
speaker-1: window because it was a company serving with small like hobby level
speaker-2: Really that's all it was. And then once AI arrived, everything changed. And
speaker-1: and we started serving the AI and market changed, right?
speaker-2: so the company itself, we went from a less than twelve people to fifty people.
speaker-1: fifty in three years, right?
speaker-2: And so there's explosive growth. And you might as well just start the timeline as three years ago. But even then we're still actually bootstrapping and that threes models off don't
speaker-1: And we're actually making money throws people where they don't understand how to value the company or think about it. And a lot of PCs I think are waking up to the companies
speaker-2: And a lot of E's are just now quit the fact that hardware companies
speaker-1: are now extremely valuable again because it's right investing in software companies and thinking through software. And so they're starting to reach understand this new
speaker-2: Half the name. They're used to software companies, formulate their models to understand the new world that we all have just found ourselves in. And I don't blame them for it, right? Like told them this was coming and learn it and adapt.
speaker-1: Right. No one told me and and we all learn as we go and right.
speaker-2: But it's quickly becoming a physical world again. And our argument is we love the fact that we're an AI company that's deeply vested in our hardware, because guess what? AI cannot replace hardware, it only makes the hardware more valuable. And so if you invest in pure A companies or software companies right now, dangerous play because it's very easy to make software easier and faster with every passing month. So you're playing in dangerous purely software companies.
speaker-1: game by investing in companies that don't have a hook in something like like hardware or data c
speaker-2: Collection, right? We want
speaker-1: AI to get smarter and more powerful. 'Cause
speaker-2: every time it does, our arms get smarter and they get more valuable. And we truly believe that you have to have the whole stack. If you
speaker-1: Like if
speaker-2: were just a pure hardware company and you didn't have the other parts have to sell the whole system, I would small organic company with the sales of the hardware. But that isn't like we're basically building an entire flywheel and we're in a unique position to do it because we have more hardware in more deployments and more research labs than any other company in the world that's in
speaker-1: world and physical AI right now. We
speaker-2: And because of that, we have the little birds that are telling us what to build and how to build and we're out ahead and people are doing deployment with trust and robotics. And we're also giving the cloud services
speaker-1: deployments with hardware giving them the
speaker-2: so that they can build the models and collect data, right? So like we're also going to be serving the teleoperation sectors of the market, which is essentially labor arbitrage where they're going to put the robots in place and just have them teleoperated to do labor tasks that are having trouble hiring locally for.
speaker-1: Right.
speaker-2: Right. So instead of someone locally thirty dollars an hour to do a task, you can now put a robot arm in place and have somebody in another country that for eleven dollars.
speaker-1: And it's doing it an hour, right?
speaker-2: Well, guess what you're doing all along while you're doing that? You're collecting the data to train models so that you can eventually automate it and then make it cheaper, right? So the companies that are first deployments with the hardware and collecting information are the ones that build the most value through these different phases
speaker-1: Phase
speaker-2: that the market is moving towards while it figures out how do we get the full
speaker-1: automation
speaker-2: Can we follow?
speaker-1: right it's not gonna be fully automated out of the gate. There's gonna be humans in the loop.
speaker-2: using teleop for it. The new product line that has workbench and the pivot and the glides which sit on top of what we call the cockpit. Telly operated from anywhere in the world. So we're serving two industries
speaker-1: Can be world, right?
speaker-2: for that. We're doing the the current data collection industry
speaker-1: Yeah.
speaker-2: and the emerging teleop industry, which works on
speaker-1: Which wrong labor arbitrage.
speaker-2: and we have a cloud service so that we're collecting all that data so that now the model makers can train on that. Right it's an entire ecosystem and that is what person are about
speaker-1: It's a trust robotics.
speaker-2: now and so we're sort of slow image that we're just a hardware company. We are now fully played and there's a message
speaker-1: Nothing off this old company. Now a platform based company's value on that.
speaker-2: because that allows us to
speaker-1: To play in the physical AI space and ride the wave while companies
speaker-2: Thousands of these companies.
speaker-1: arrive and say we need solutions and we need them yesterday to deploy. And currently we're the only company that has a full package and we're the only company in the United States that has that.
speaker-0: I think that's an epic summary. Please consider Truss Robotics if you wanna play around with an arm or something like that. If you have a use case, deployment, anything like that, check them out. This was the Robotics Stack Podcast, episode twenty-one with Matt Troston. My name is Oscar.
speaker-1: My name is Eli. Thank you.
speaker-2: So much for having me on. Bye bye.
speaker-1: My name is Leo. Matt, thank you for being here. We are super happy to have you with us today. Thank you very much. And you have a very nice setup on your end here. Yeah, we built in our podcast room right now. We've shot I think about ten episodes. We haven't released any yet. Are you planning to release regularly or is it more of a one off thing?
speaker-2: No, we'd like to have a regular cadence. I believe that it's kind of our duty and everyone else's duty in the stages of physical AI evangelize and promote what's happening and help have conversations going so there's more understanding. Think about industry you think back to that jobs would evangelize.
speaker-1: understanding of it. Like we think about it like the early PC industry. And if you and you have like Steve's right and let people know what is this new technology, what can it do? You know, for us it's marketing obviously to get the work. Also to educate, 'cause a lot of people are wondering
speaker-2: Obviously, but I think it's all looking at this next wave of robotics and what
speaker-1: what is it? What what can it do? And sometimes in depth conversations help keep that going. Yeah, this is exactly the insight we are looking for also when we started the robotics tech. We are going to dive into your background and the journey into building trust and robotics, obviously. But first I would like to ask you, you started your company more than twenty years ago. And if today you were at this stage, you are just getting started, would you still go into physical AI, into robotics if you were starting your company in twenty twenty six? Yeah, I think for sure. I think it's
speaker-2: at its most fascinating it's ever been. Waiting at this point. So if I was to start all over
speaker-1: point that it's and I've been waiting for it to get to this point all along. Or it'd be kind of crazy to go do something else, I think. The company two
speaker-2: When we started the twenty twenty
speaker-1: years ago, which is just wild to even say at this point. The idea PC based robotics. It was the idea that the PC would be the brains and the nervous system,
speaker-2: idea was basically piece of it was become the center or you know brain of the nervous
speaker-1: if you will.
speaker-2: we started selling components on a store at the timer called Fitz or control electronics.
speaker-1: digits USA they were USB components that allow you to control and take input from sensors. and that was the concept rifted towards robotics. At first it was just sort of a tinkering set and then we and
speaker-2: And just trended more towards things. Added the motors and the servos and
speaker-1: it just logically was like robotics is what needs this, right? And so that's
speaker-2: So in those twenty two years that
speaker-1: what's happened is you've had, you know, places like Wheel of Ross, which sort of established the ability to have the computer Linux. And it
speaker-2: Garage creep staff have in the center on a system is grown from there and now we wouldn't even think of building a road. I didn't know we were waiting. I didn't know that in my lifetime.
speaker-1: robot without a PC in it, right? what I was waiting for was AI. AI would exist, right? And so three years ago when Stan demonstrated how you can do robot learning through VLAs, everything realized,
speaker-2: Exploded and you know we're
speaker-1: my gosh, this is what we've been waiting for. This is what allows the robots to actually do something now, rather than just exist in a lab. From what I understood, you quit your job and decide to start. robotics company in two thousand and four. I'm guessing some people might have thought you were crazy at the time to start such an endeavor. What was your vision? Okay. I didn't quit my job of the gate. I quit my job about a year and a half later. I had a boss at the time who was nice enough to let me do didn't
speaker-2: So right out of the later. 'cause I have who do this on the side and
speaker-1: fire me as soon as I told him. And I actually grew his
speaker-2: The company in his
speaker-1: his office space, interestingly enough. But
speaker-2: The original idea actually wasn't robotics. It was those IO components that I was telling you about. They actually were called fidgets. It was spelled P H H I E T S. Physical
speaker-1: Called f which D A D G which stands for
speaker-2: widgets. And I had the idea for those two in the morning sit up and my gosh, this is a great idea.
speaker-1: In the morning bad girl.
speaker-2: Because back then I was a programmer, but I wanted to be able to build physical didn't know how to do electrical engineering.
speaker-1: things, right? But I didn't do electricity.
speaker-2: And so the concept was, well, what if I had a USB component that would plug in and then I would have
speaker-1: You know, through
speaker-2: Code and back then I'm a dinosaur. Back then I was coding
speaker-1: And you know, V B six and V B dot net and I th
speaker-2: thought well if I could program on a computer net and then control the real world around me really
speaker-1: computer in and interactive and like art pieces or like
speaker-2: virtual reality thing where you're sitting in a chair and it's vibrating and the fans are blowing faster and you're piloting a car or a sp like that.
speaker-1: spacecraft or something like that. Or I could do like really trippy vision of like an immersive music video or something
speaker-2: visuals and stuff kind of video or
speaker-1: of that nature.
speaker-2: that's what I was trying to problem solve. And so I can't of these, you know, modular components where you can control motors like fans and vibrating motors, servos
speaker-1: And then
speaker-2: to operate things and take input from things like sensors, sensors, distance sensors, switches, light sensors, water sensors. And sort of having a modular system to build you want while programming.
speaker-1: So like pressure is lowers, you name it, right? And build whatever you that was the genesis of the idea of the
speaker-2: I found another company in Canada called
speaker-1: fidgets, which actually already made these. They had the same idea about eighteen months earlier and a day or two. And then I called them and said, Hey, do you have a a distributor in America? I convinced them to give me a thousand dollars in I'll
speaker-2: So I was crestfallen for a called them and they said they didn't dollars free stuff and I said I
speaker-1: become your distributor.
speaker-2: and so once you put all those in the store, well if you sell
speaker-1: motor controllers they gotta sell the motors, right? If you sell servo controllers, you gotta sell the servos. If you sell you gotta and then now it's like, well we need chassis on top of this, right? And then what else can we do that's fun? Right. And it just drifted robotics. I was just sort of
speaker-2: Well, Pan and Tilts and Robot just naturally drift towards robotic, pulled in that direction.
speaker-1: just from following the market.
speaker-2: And so I built the website, I started promoting this stuff. I built some forums around it. I started buying every robot I could get my hands on from iRobasmotion, you know, rec kits for robots to crust crawler back in the day to Parallax components. This is all pre Spark Fun and Adafruit and
speaker-1: Amazon days, right? When you could build a little if you will, and to survive. That was the point.
speaker-2: And after that it moved towards robotics and PC as a thing. I didn't I was literally
speaker-1: based robotics when I realized this is this is the future. What I didn't know is I'm really twenty years too early to the game, right? I wanna accept ready
speaker-2: I was trying to pitch a concept that the world wasn't
speaker-1: for yet. And so Q2 decades of cloud and
speaker-2: Plawing my way survival, trying to find out where is the market going, products through that.
speaker-1: Delivering pro that
speaker-2: So we got the original company survived on selling to highlight at home, they're making crawlers, pan and tilt. And then the arm started selling in
speaker-1: arms things like that and selling to academia. So people started using them in their robots labs
speaker-2: Par extra.
speaker-1: because we sold low cost arms.
speaker-2: Every time we move a bigger arm or a or more professional arm, it got adapted up closer and closer to like high level.
speaker-1: It had got and we started like scaling up close to research and masters and PhD level. And specifically company
speaker-2: So that's basically the first nineteen years of the
speaker-1: right there in a nutshell. And so Sanford made the aloha using our hardware. Aloha stands for low cost.
speaker-2: Then Stan Pocket uses which you know a low cost hardware. So NASA, you know, one which was robotic manipulators. And that kicked off.
speaker-1: the whole AI space and, you know, physical AI at the time it wasn't even called physical. So we were buy this new wave and
speaker-2: model training sort of adopted by
speaker-1: immediately refocused on it. So I'm guessing the product line you currently have then is if you consider the entire lifespan of the company pretty recent, isn't it? Yeah, I mean if you look at the overall twenty two year history, yes, it's very recent. So it's over the last now for physical AI. We very
speaker-2: Three years we're on our third generation. We quickly adapted off the Robotus servos servos physical. It stands for quasi-direct drive. What that does is large motors inside, and then as little
speaker-1: Does is it has
speaker-2: gear reduction as possible. So that you have torque transparency and
speaker-1: Yeah.
speaker-2: compliance. If the arm hits something, it's not going to through it. So those three aspects.
speaker-1: And then in lower costs because they're standard planetary motors. Attributes are what basically everybody in research and physical AI
speaker-2: wanted so we had to adapt to that because that was the first generation of the arms. Now third generation we're hardening the product to making a more robust controlers are toughening
speaker-1: was the first or second generation made that leap and products are more robust. We're putting in industrial level control. We are tough up things like cables, our bracket designs, even the connectors that are on there, we're moving everything away from USB. It's all Ethernet based now. Things that are necessary for
speaker-2: all the things for that. And then we've also listened to the market to figure out the mobile is physically sized be the right not too not too small. The same kind of vertical lift on it so it can shelving up and down. So a lot of people do that. And then the payloads on them.
speaker-1: What is it the market wants? So ribb, which is a manipulator we sell, is to be the right size vers not too large and not well so it can do the payloads that people are asking for. It's got a vertical reach wanna do that. product because that's too dangerous in a market where everyone needs
speaker-2: To tweak the product a little bit or adjust it some. Technology is coming out so you're literally remote six months. So to overpolish
speaker-1: Or new technologies come so quickly that you're building your product line every single right.
speaker-2: stuff and then charge people for the polished product. So we like to say our robots are a little bit ugly because they're a little bit bare brackets and bars and things, but we're finding that that's what people have functional
speaker-1: They would rather have something funny.
speaker-2: that they can deploy right now and test in the field whether actually a deployment or whether
speaker-1: Whether it's you know the deployment or it's in a in research itself because they want to add or change the compute.
speaker-2: a sensor or change a grip or or you know, swap up the camera. Everyone's gonna different. So to us the Goldilocks zone is malleable.
speaker-1: cameras everyone's got a request for something slightly different is it's still to an extent even as a finished product, right?
speaker-0: You just mentioned the rivet and that is part of the new line that you have recently introduced. Is this new product line aimed to replace the old one? Maybe you can also take us to the core products right now, rivet being one and then the arms of course and and the other major products that you have. Sure.
speaker-2: So the new line that's being rolled out as name manipulated and
speaker-1: We speak has three main operators in it.
speaker-2: they're all under the Widow X name. We actually were names. Models of the Widow X. So these the ones you have in turn. The
speaker-1: come up with different for each of them and realize that the widow X line that it's easier just to make three different type of like marketing decisions internally. The
speaker-2: main one is the Widow X Pro. And so it's similar to existing Widow X AI that's out right now, except it has servo on the shoulder. And from the ground up, it's a rebuild.
speaker-1: Yeah.
speaker-2: construction of it. So it's got a lower cost brackets in it. So we can get the stuff, new shelling design on it. It's got an all new controller in the base. It's got control
speaker-1: Ethernet.
speaker-2: on it, industrial connectors and modular grippers and a compact gripper that doesn't have the big bow ears on it anymore. So all those improvements have made. We have a version of it called the core. It has the traditional smaller motor in the shoulder that's a lower cost. That's basically aimed at a budget conscious loads.
speaker-1: Which has the traditional basically education, people that are very that don't feel like they need to move higher payload.
speaker-2: So people that want to save a little money. So those are the two bottom line ones. And then we have the heavy. And the heavy has four large motors so that entire R
speaker-1: in the base and the elbow that the entire arm is beefed up and it has a one per meter reach and so you've got you know extreme reach and you've got up to six kilometer load on that
speaker-2: kilograms of payload arm. So a lot of people want to move larger things for and a lot of that's trying to move
speaker-1: It's because people are trying to
speaker-2: into deployments. So they're trying to come out of the lab and in fact
speaker-1: actually put these things factories or warehouses or small businesses to actually perform work. And so you need more payload and reach for that arms
speaker-2: Form. So that's the three.
speaker-1: that we've made.
speaker-2: We also have something replacing the stationary. So the old Aloha stationary
speaker-1: The is this big cage
speaker-2: product that sits on top of a table and has arms pointing inward. So that's done in the day. It has four arms and four arms. They're all actually that's now the workbench. Two arms that are forward facing.
speaker-1: Arms pointing that's the original way it was that's so as two leader, two followers, but they're servo based arms. That's being replaced so that we call them and so that is two facing because the whole market has moved to forward facing arms. It's a better embodiment that everyone's training us and
speaker-2: arms on and then it's just got two handles on it and computers and your arms all together. So it's no longer you have to table and then work
speaker-1: Your computer and you can just move it around really quickly. Set up like an entire work only within the confines of that table. Now instead of having this,
speaker-2: having actual robot arms for the leaders.
speaker-1: we have what we call the glide.
speaker-2: So this solves a lot of problems. It's called stiction training.
speaker-1: in the industry because you have what when you're trying to do your training
speaker-2: So when you're using a robot arm as a leader arm, because it's heavy and it's like servos cop
speaker-1: Because got big in it, you have to have gravity c turned on so it can hold it itself in space. you're actually sort of the arm. But the problem is is the know what gravity and what's the
speaker-2: And then when you're moving it around, sort of guiding the arm doesn't know what's gonna human pushing it. So when you slow down and then pause or do really s movement fighting the servo itself, trying to
speaker-1: slow movements you start what's safe.
speaker-2: do finite, granular, dexterous tasks.
speaker-1: It's very difficult when using Gravity Comp on a servo based arm. The
speaker-2: So you'll end up overshooting
speaker-1: thing you're trying to grab, knocking things over. It's it's really hard to be articulate with it.
speaker-2: Really hard to be the day that I got really fed up with watching this, I was at two years ago and we were watching people do demos on the arms and struggling over or over
speaker-1: At ICRA people and they were struggling to move blocks around and they were knocking the stack over overshooting and knocking them around. And I was sitting there going, terrible.
speaker-2: This is
speaker-1: We have to solve this for people. You know, there's the old saying of bad data in, bad data out in programming, right? Well, in robotics, if you are training the models having difficulty learning, right? You want smooth, dextrate trained.
speaker-2: So we went through a lot of different operations.
speaker-1: We tried to use smaller motors. We tried based gravity camp. We tried to put the motors in the base and have belt drives and all sorts of crazy things we were trying to do. And we realized that there's been
speaker-2: Just no hope.
speaker-1: doing it with using servos. So we just threw it all out and from the air glass just
speaker-2: Throw ground up, redesign the entire slide.
speaker-1: to be encoder based and lightweight. It's like a featherweight is this genius design that one of my engineers came up with where they use cams and springs in the joints so that it actually does spring-based gravity comp where you need it and it it works brilliantly. So you're holding these and they're very light and so you can use them all day long. And since you don't anymore, you don't have stiction. So now you can just
speaker-2: Really smoothly move the arms in very finite and training.
speaker-1: Movements data it should be trained. So that's what we decided for learning.
speaker-2: Leader arm, since we've released them, they've been a smash hit. They're the number one seller out of clamoring for them and we're
speaker-1: of everything that we've released. People are rushing to produce as many as we can because it's jurisdiction issue for everybody. You've been using arms leaders as well the size of your fleet because you can just your followers. That is the new leader arm. If we do
speaker-2: it has solved this they're also lower costing it in the IVs they work with not just our army and so if you have a fleet and you robot as well you can now double spike glide arms and then take old leaders and turn them into a foul. So we also have that if you really must have gravity comp, you have a version of the core that is how you can that the new products because some people that
speaker-1: is still a leader, so can still purchase in the people do want it still. It covers those and then we have Rivet.
speaker-2: which is the mobile manipulator. And so that has the
speaker-1: As the
speaker-2: base is about six hundred by seven hundred millimeters. We've got eight hundred and ninety millimeters of travel on the Z axis T
speaker-1: We I believe it's eight meters of there's a sturdy
speaker-2: bar on it.
speaker-1: and it can hold the two pro arms on it as well. And so supposed to hit the sweet spot.
speaker-2: So that's the goalie lack zone of what most people have asked for. On a larger version, which will probably that it will be able to handle the headline. It is a lot. It's been a lot of work. We're also working on seven duffs.
speaker-1: in a rover and we are working come out Q1 next year that heavy arms as well. And that's the new project. We're as
speaker-0: That's a lot of product.
speaker-1: well, but I don't want to bring them up because we haven't released them yet. Yeah, I want to react quickly on the the first part of what your answer about the gravity compensation because it feels like this is like one of the blind spots that the research community has. And since you are very close to the hardware, you realize it was like a hardware issue and not something you would solve with a model. And it's fascinating because I feel like sometimes everybody right now is overly specialized into one tiny part of the process and is having a hard time taking a step back and understanding that maybe there are other limitations. And especially in hardware. I feel like hardware is kind of neglected in many ways. So I'm curious right now, ever since you ship this product, have competitors copied you? Is your way of doing things becoming the new normal?
speaker-2: So we haven't shipped the new product line.
speaker-1: Yeah, it's in production. We have the prototype run that we made that and that's what we used when we went to California to show
speaker-2: gonna answer that more of a general way is really interesting leading up technology.
speaker-1: Because it is in, you know, blue branch technology how ideas spread, right? And so that deeply we're
speaker-2: So you said that 'cause we're integrated.
speaker-1: seeing things and we we see a s a problem and we say, Let's see if we can solve this. Benefit of being here for twenty two years and being as deep in the ecosystem as that we just serve research based. We serve research, we serve data clear, we serve and we serve deployment, which anywhere from startup, right? From all of those places having here's the things we want.
speaker-2: collection or teleoperation solutions and we sort of deployment can be from a fortune one thousand company to a stock. So we're getting signals from the market all over places saying here's the problems we're having we want to solve. And so sometimes you can just glean from what everyone's telling you. Sometimes they don't even know.
speaker-1: that they have that problem, right? And they want it solved and so sort of re tea
speaker-2: So as a company you can be the
speaker-1: leaves and make something, right? But
speaker-2: A lot of times you're seeing stuff come out or solutions from other sort of an iterative design that everyone's sort of sharing amongst each other.
speaker-1: Right.
speaker-2: So when we originally made our arms robotic service loans of those one Ali button.
speaker-1: From the servos back in the day, we had clones left and right. Like you go to the bottom there would be clones of our arms being in China out there, right? Like that's normal. They would even just have our picture and
speaker-2: Sometimes they're arm and then and I would just for fun I would pop
speaker-1: Email them and say, like, Hey, take that it's not yours and sometimes they would apologize and claim an intern did it or something. and other times they would just you because you can't sue them over there, right? So that would certainly and then when
speaker-2: happened. And then when the QDD motors came out, you had a handful of companies come out with different design center around
speaker-1: designs that all started to on the basic design that you see everyone have now, right? And
speaker-2: You can argue about who was first and who borrowed from whom until the cows come home. But
speaker-1: So the truth is everyone's borrowing from each other, right? One company
speaker-2: So one company
speaker-1: comes out with a a servo layout that may be adopted because a lot of people collect data on it do that, but then they'll change the gripper and
speaker-2: So another company will they'll change for
speaker-1: they'll make the gripper a little bit better with a different, you know, rack and pinion gearing system or something else. And then another company comes out with fingers that are a little better, right? So like is people are starting to use component grippers.
speaker-2: Like right now in evolution it's happening compliant grip that are half rubber, half stiff material. And so generalists are some of the first suddenly over the place.
speaker-1: companies like Genesis are the first that came out with that and then suddenly you see it popping up all over the place, right? And so people are mimicking each other from each other. technology like this because everyone's trying to figure out you know stuff factors for things like foras it sort of land something that works everybody starts to converge in the world is the smartphone. Right? 'Cause think about before
speaker-2: How does this work? What are the best form factors things? And then you have, I think it's called the form factor convergence. When the market lands on something that way. The best example iPhone for the Apple iPhone came out, how many hundreds of different variations of in the world?
speaker-1: phones exist and then all of a sudden nailed something that really really worked and everything within years, right? Space is that once you know job that proliferate,
speaker-2: Everyone merged on it. So that's gonna happen in the robotic. Once you figure out a form factor that works for job A or job B or job C will play
speaker-1: right? And nobody owns it, right? Unless somebody races out and gets a patent, which is highly doubtful because you can patent you do something, but you can't patent that concept. It's a shared idea in the space, right?
speaker-2: So yes, we've been copied as well.
speaker-1: We've borrowed from other people's designs as well. It's the only way to basically build products and anybody who claims any difference, like being disingenuous.
speaker-0: That absolutely makes sense. With these experiences in mind, are you now one hundred percent producing in the US or how is your supply chain looking? How are you making these products?
speaker-2: Yeah, so we've been building in the US warehouse
speaker-1: day one, right? So we have two houses here in Chicago right now where we build all of our products and then we're looking at a third to move into that's the size of a football field. I strongly believe that the quality has to be
speaker-2: Strong control has to be
speaker-1: here under the same roof. So I intend to keep it that way for a long time. We can go overseas with some production we would only do
speaker-2: When scale demands that we can production do so when we can have our own people on the ground there, keeping an eye on making but until then Chicago. So parts are made all around the world, most of them are in China.
speaker-1: Just like everybody, because they are the ones that can make things that the most cost and they've actually gotten good at making things in quantity with decent quality, right? So bring them and then
speaker-2: So we still purchase like everyone else parts in and
speaker-1: build them here in house.
speaker-0: As to your product lineup, you also introduced some software parts, the S DK for example. Can you talk a little bit about this and why do you think this is important? Sure.
speaker-2: So we've had SDKs for a very long time for all of our products. So that's nothing there. Trying to push the announcement about it sound how robust specifically is sort of the new part of that. Also
speaker-1: I think us to push announcement so that people understand how cost it is for physical AI specifically.
speaker-2: have Trost and Cloud now. So that is in alpha at the moment. But we believe that that because you need full stack solutions for people, right? So in very early adopters, people will build their own Frankenstein tech stack solutions, right? And they may be just working on hard
speaker-1: Will Stein solution right be hardware but need to build player, right? They're
speaker-2: So they do the software, right? Most people applications later and then
speaker-1: have to build their own SDKs because they don't have any SDKs out.
speaker-2: there that they want. So they would build that. They've got to build their own cloud solution because there's no cloud solution on the market. They build their own encoders and filtering algorithms and yada yada yada.
speaker-1: Yeah.
speaker-2: So they people build their own. But as technology matures and you have more late stage adopters a bunch of they'd rather buy it out of a box and just work on application, right? So we also want to raise our you can do that by
speaker-1: value as a company and that
speaker-2: offering more of a stack. Right, and not ch hardware. So we think that the it's basically the next largest
speaker-1: by AI space wave of automation and humanity.
speaker-2: is what's gonna happen. So if you think about what physical AI is, it is the merger industrial age.
speaker-1: of the induction age and the information age together, right?
speaker-2: It is physical labor and intellectual labor combined. And so modern robotics are bolted from the factory floor. They're eyeballing reactive.
speaker-1: training basically getting unballs, they're becoming and they can be trained like humans to do human like right change the world, right? Change of humanity, physically changed the face of the world.
speaker-2: Tasks. And if you think about how much the industrial changed the face
speaker-1: I brought a whole new phase in humanity, right? And then the information yet again, right? So we invented the computer.
speaker-2: And so intellectual automation.
speaker-1: now AI, right? And we're seeing a wave of information that we never even thought would happen in our lifetimes, right? And that if you have the AI, the brain, doing intellectual and physically being able to move the world. So it's going to change the world yet a third time
speaker-2: times. Put these together, then you can have actual tasks around it. World third time.
speaker-1: in ways we can barely even conceive of. It's just in the in the sci-fi movies that we can imagine what this is going to look like, right?
speaker-2: So the idea behind Trostin is that we don't just want to be a hardware, have the full serve.
speaker-1: company we want to have solution that can serve
speaker-2: All we call them prospectors, right? People that are looking for We don't really want to be a prospect or a picks and axes and shovel company. Sell the components and hardware to the
speaker-1: ourselves. We are so we want to companies that are doing this.
speaker-2: to get them up and running, they need more than just the hardware. They also need the SDK, which is attached to a cloud that has the tools and training. That's a full product. That's a full solution. And that is what the market needs. So we have been in a unique position more hardware in more places than company who's in this space. Here the longest. And we decided to add the software layers to that so that we have a robust product that serves the needs of the market all.
speaker-1: Serves needs as it evolves.
speaker-2: Right. So now the concept is when you buy an arm, you take it out of the box, you can log into Truss and Cloud, you can start collecting your data on the same
speaker-1: Day.
speaker-2: collect the data for whatever need it is that you're trying to learn. You can pick the model that you want to use for
speaker-1: It is
speaker-2: people that might be using a model that's already out there, an open source model like Pi. They might be actually the model builder themselves who want to collect data and keep improving them.
speaker-1: It could be a company for the
speaker-2: It's forking a model and tuning specifically themselves, all the above, right? But what they want is as quickly as possible. They don't want to buy the hardware and then figure out where do I store my data on what cloud. Now I gotta put my software stack together for my teleop. I've got to put my software in intake of the data, processing. Like it takes time to build
speaker-1: that does my data, my labeling and cleaning, my inferencing.
speaker-2: all that stuff. And so we've added that to our hardware. So now if you purchase a trust or you can do all of that ecosystem bringing it up.
speaker-1: arm you can from bringing to training to cleaning and then go believe that we're that value
speaker-2: back down onto the robot itself and go into deployment. And that's how we believe we're making a company that can go the distance and has
speaker-1: in this space
speaker-2: and we believe you have to own that entire stack to do.
speaker-1: Do so.
speaker-0: It really feels like your company has been kind of looking what's going on and now is the time you're like activating the boost with all these interesting learnings and experiences in the structure that you've built up over this long time. And now kinda all falls into place.
speaker-2: say you're absolutely right. Like so our first investment round now.
speaker-1: we're taking outside investment in and it's an angel raise with eyes on growing especially have to move to survive.
speaker-2: fast as we can because you fast arrive at all in this space, but this is the time to
speaker-1: basically pounce on the market and build the companies that'll last organically grow. Our belief is that
speaker-2: And we can't organically grow through this phase.
speaker-1: it needs investment and money to grow this. And so we've chosen to basically change the way we do things.
speaker-0: Yeah, it's interesting because I already had a segment about you in one of our podcasts and then I was researching the investments and I was wondering like they are completely bootstrapped. Like and then you look at all these other companies and like why are they not raising? They must be raising. What the hell? And it kind of makes sense now. You also brought a CEO on, Anis Shah. How did that happen? So
speaker-2: Part of my realizing that we need to grow faster and bring on investment was thinking about team building and that we needed to bring more sea level. At that point it was just little over
speaker-1: people into the company because up until it was just me running the whole company. And it starts growing
speaker-2: beyond a point where one person can do it or in the technical design product and
speaker-1: Do it all, right? And my strengths are nine of the products.
speaker-2: then watching the market and reading the tea leaves of the market. Those are the two areas that I've been honed to do.
speaker-1: do for twenty two years of experience, right? We kinda think about like wax on wax. Like you don't even realize learning these skills along the way.
speaker-2: Off the old karate crud metaphor about realize you're learning skills, right? So being a small bootstrap company and learning to survive in bleeding edge markets like robotics, sort of watch the market, see, you know, inflection point is that what's the next line of products that it's asking for, what different pieces of software need to be bundled together to serve that market, right? That's like what Tronson does essentially. And if you look at physical AI now. It's just more of the same. We used to do it for hobbyists. I'm gonna grab
speaker-1: Where it would be
speaker-2: you these servos, this controller, this battery pack, these set of brackets, and this example code. And that's enough to get you up and run your desktop, for instance, starting to work with Raw Ross packages in right this evolution of just
speaker-1: With a robot arm on your so we started adding robot. It's just a
speaker-2: layering in more and more tech as it evolved. So in the physical AI space we're really it's just now
speaker-1: And they're doing the same thing.
speaker-2: it's a cloud that has modern AI models in it that can be
speaker-1: My trained up it just gets bigger and bigger, right?
speaker-2: on the hardware. So as the company gets larger, we need more people and to run the different departments.
speaker-1: Right.
speaker-2: So we brought Anish on. So Anish, he's got a very storied career as well. You should probably have him on his own podcast. So he has worked in startups, he's worked at different businesses, different sizes. He had a stint as a stand-up comedian
speaker-1: Comedian.
speaker-2: for seven years. He worked at McKinsey and he's worked at Moderna during the vaccines. He's kind of done all sorts And he's had a passion for AI and robotics for years. He's been recently wondering if he could get into this utilizing the skill set he has. Well, it turns out he's in our backyard. He's been a couple towns over from us. And so he said, Yeah, I'm this guy who's interested in this. Here's
speaker-1: And once he's doing research he discovered so he just messaged me and said, Hey
speaker-2: my skill set. You know, let's have lunch. And I thought, it sounds interesting, sure. And so we went and had lunch and that turned into a four
speaker-1: Four.
speaker-2: hour conversation and showing him the shop and discussing concepts and where we went.
speaker-1: go with and as we get married together. Meeting to when he joined so we basically got to know
speaker-2: This he likes to say we fell in love with each other and decided to go and work on this company. So from that first meeting the team, it was nine months to know each other more and worked on some projects together. And
speaker-1: And it just became really clear to me
speaker-2: that the strengths of strategic has at this
speaker-1: thinking that he has is exactly what the company needed this time.
speaker-2: right. So we brought him on as COO for a while and recently decided that it's time for him to C EO role. So I could constantly
speaker-1: Then it was time to move into the C concentrate on the C T O roll. Because
speaker-2: so this dynamic works really well because he can work on the raises, work on strategic partnerships, work on growth of the company while get to
speaker-1: Well I guess
speaker-2: focus on building the actual products inside the company. And so that's building the C so
speaker-1: Yeah.
speaker-2: that as we bring investments and grow from there there's more people at the helm trying.
speaker-1: There they're trying to you know, hold the reins of what's basically a bucking
speaker-2: You know.
speaker-1: rhino, if you will.
speaker-0: And speaking of partnerships, are there different extents of how you work with people together? Do you have like specific partnerships or is it more like you buy our product, we have a certain type of warranty and service that we provide after it, or is there some differences there? So
speaker-2: Yeah. So you always want to sell your core products as much as you can and then alter off of your core products. So we're not like a pure consulting shop where we come in and say
speaker-1: And what kind of robot do you need when we build one from scratch?
speaker-2: What I like to say is the invention isn't just the robot kits who make the invention the company that can react and adapt to rapidly. Like that's the challenge. And so
speaker-1: to a changing market rather than challenge, right?
speaker-2: there's gonna be an embodiment war or
speaker-1: So like if I
speaker-2: Darwinian war amongst all the different form people are gonna dream up over the next ten years and everyone needs to try different
speaker-1: factors that people are going to dreamers on things to see what works for different use cases, right?
speaker-2: So what we go back to like the prospectors, which are people that are trying to do deployments and and find applications for
speaker-1: to this new thing called physical AI. Each of them wants something
speaker-2: Each of them is slightly different, right? And so we're building a company that can react to that. And if you want fifty robots with a different end effector on it, we rivet with a taller Z access and more payload. So we have to enlarge the base a little bit.
speaker-1: do that. If you need maybe then we'll work with you to do that. And those are happening now to look at deployments with fifty a hundred units.
speaker-2: sorts of conversations are now as people are starting to 100, 200 units, right? The cloud part of the services feeds in that where people need different tooling in the clouds. So we'll work with and the different technology layers necessary to achieve goals that and then that way the the market is informing
speaker-1: achieve your goal there, whether you're a model maker or an end user.
speaker-2: us on how to evolve the products. So we look for a sweet spot with partners and clients who are near where we already are. But just need to slightly improve it a little bit more in one way or another. Improve the product. And then we hope to sell that same product to more people. Right. So if somebody says we need bigger arms on a mole manipulator, we'll
speaker-1: Well,
speaker-2: build that and sell it to all the customers out there, right? So we're kind of looking to all swim in the same direction because if we try and that don't overlap in too many areas, that's a good to come. You'll get spread too thin, keep
speaker-1: make too many things are a good way to die as a company, right? Because if then your resources just can't
speaker-2: up with it. So you have to be selective about it and say no nine times out of ten and they're directly. But we're
speaker-1: And then yes to the ones that make sense because they're heading in the same direction you are. But
speaker-2: open to all sorts of different types of clients. We have clients that are startups right now that are trying to deploy
speaker-1: Or deploy and
speaker-2: to medical deployments or grocery stores
speaker-1: Plus
speaker-2: for picking things off the shelf or packing house as well as like Fortune One Hundred companies that are trying to protection put into robotics so that they become part of the
speaker-1: trying to do data collection or get their products, they can become physical AI game.
speaker-2: And so the challenge is finding the right partners that of
speaker-1: Yeah.
speaker-2: you who want maybe something a little bit different that your competitors can't, but not so different than
speaker-1: That you can answer competitors, that it changes your business model. I don't know. Did I answer your question?
speaker-2: Yeah.
speaker-0: Absolutely. I think if you spend time and obviously work with a lot of clients, you get almost like a sixth sense about what is needed, where the deployments are going and obviously you can use that for your research. Do you have a specific research team or was your team setup looking like right now? So a lot of it is just
speaker-2: Feel at the moment. So the team isn't large enough to have an entire division that's taking all signals and then reports that I hope that we'll
speaker-1: There's no emailing us and talking to us and
speaker-2: That should go through a matrix to determine what it is that you're going to create next. So no shortage of signals coming that are calling us to us at conference. Can you do
speaker-1: saying
speaker-2: this? Can you do that? Through just hundreds of conversations, you're like you said, where you get that
speaker-1: You're getting a feel like sense of this is where it's going.
speaker-2: that's the broad brush strokes are are easier to figure out. I call it the Little Birds program. So if you guys ever watch Game of Thrones, one of the characters talked about it.
speaker-1: Little birds which are vessels that would feed the information. We basically
speaker-2: were his spies that he had and all the different castles back. And so I sort of stole that little program. So
speaker-1: got our ears to the grindstone trying to hear what is it that people want in that literally does that twenty four market going, what are we developing next? And
speaker-2: need and I do want a division or seven and so there is the smart starts preparing for it like twelve minutes out, right? Right now we're very reactive. And that's
speaker-1: We're basically being pulled by where the latest thing is that people are asking for.
speaker-2: perfectly natural. And and as you get larger start to level off for
speaker-1: in a brand new market that's bleeding edge and then things start to level off when you start to hear
speaker-2: more and more of just
speaker-1: we need a bigger arm, right? Being pointing you in just the bigger arm direction. Or
speaker-2: Or
speaker-1: it's become clear that these compliant grippers are what everyone's gonna use. And so you have to just
speaker-2: Listen to that. But the normal process in a more mature large company is this sort of product development cycle that can be twelve or twenty-four.
speaker-1: goes through a four month cycle. If you think about something mature like refrigerators or automobiles or something, the
speaker-2: Their time from concept to
speaker-1: creating it is much, much longer we live in. There was also a common problem that we have been not directly running into, but with that we have observed in robotics that there are many companies now that deployed humanoids, teleoperated humanoids to do basic tasks like cleaning, cooking, pick and place into real homes in order to acquire teleop data at scale. The main issue is that the people who teleoperate these robots, they basically wear a VR headset and a handheld controller and they cannot do much more than forty five minutes to one hour at a time because it is extremely exhausting to wear the VR headset. It's heavy, it will make you dizzy and so on. So we have been thinking lately about is there maybe a better way to provide a telehop interface as well.
speaker-2: Yeah, a lot of people have
speaker-1: started doing egocentric data collection, right? Where you're wearing a headband that has cameras on it and then you're either just using your bare hands which are being recorded or you might be using a Umi like device where
speaker-2: Yeah.
speaker-1: you have ripper that you're holding and it's triggered or you're actually wearing it like a glove and you can
speaker-2: pinch this way. So you're seeing people try all sorts of different and some of those are getting down,
speaker-1: things and and getting the cost
speaker-2: right? So that you're not buying four arms in order to collection.
speaker-1: Order to do the data collecting.
speaker-2: The debate is is that data good enough to use yet or not, right? What you lack when you use that
speaker-1: Because we stuff is the actual telemetry data, right? From your matics.
speaker-2: kinematics. And so just egocentric alone, you know, for the feedback we've got from most researchers, it's like you'll never
speaker-1: Is not enough.
speaker-2: You have to have someone opp physical robot arms the task. And so it might be that you'll do ninety percent or something can be done from egocentric action, but then the last mile, the last ten percent you have to have the can
speaker-1: The first can be done vision data collecting percent. You make the data of the arm actually physically moving or the no
speaker-2: Model just hasn't
speaker-1: idea what it's being asked to do, right? Remains to be seen, but this is the research learning as it goes along, right? Now
speaker-2: And what that blend is, you know, market learning. So the cost is coming that we're
speaker-1: it's You know, like what we call the workbench light, just
speaker-2: That's
speaker-1: two arms, the three cameras on a metal plate, your power
speaker-2: Late in your
speaker-1: solution, a networking solution solved for you, dramatically brings down the cost from what used to be a big stationary box that was just, you know, in your resource. And then if you choose ego trick to add to that data, then those headsets can be anywhere from a five
speaker-2: Five hundred dollar phone.
speaker-1: to a two or three thousand dollar headset, right? So that's bringing that cost down.
speaker-2: And the mobil up of what you can collect. and the other thing is customers they're hiring in the SDK and the cloud and all that.
speaker-1: We service this market, right? So people buy our and
speaker-2: Our kits.
speaker-1: then can use the cloud to upload the data and lay and and
speaker-2: and clean it, label it.
speaker-1: do everything they need to. The hope is that we'll eventually have a full marketplace where we even help them resell that data and isn't the egress of data being moved around.
speaker-2: And so there is and all of that as well. That helps lower the cost.
speaker-1: And and make it quicker and easier for these companies to get spun up, right? 'Cause we're solution like a franchise at that point where it's here's
speaker-2: selling them the whole it's almost like it's here's the hardware, here's the software you need, some tool. So
speaker-1: don't we have the data collection, we have the data cleaning tools, we even help you sell it. So people can think of different areas where they think they can add value to the by collecting data that other people have instance or service customers in different ways.
speaker-2: add value to the system but haven't for more empowered to do so because the technology's just
speaker-1: evolved right whereas for the last few years they would have had to figure out that whole data stack themselves in terms of like cleaning around we've seen different ways that well seen offering to come clean your free if you just let somebody come in with an egocentric headset
speaker-2: themselves as well. In terms of collecting complex data in a house or something that have been done as well, we've seen like the gig economy offer their house for centric assets.
speaker-1: which is a really interesting way to approach it right emails from data
speaker-2: I daily get even firms all over the world that are just putting headsets on top of factory workers and they're just
speaker-1: Just pointing out just trying to resell the data.
speaker-2: like there's this mass rush of people trying to get rich.
speaker-1: on this new thing that they've discovered, right? And that's gonna drive the prices down because the competition's gonna make it cheaper and cheaper and cheaper on a data cloud.
speaker-2: collection. So obviously everyone knows this with LLMs, you had the entire knowledge of the human internet crawl between your
speaker-1: race on the internet. And so you could just go crawl that and trade
speaker-2: LLMs.
speaker-1: when it comes to robotics, you actually need physical world data and that doesn't really exist yet. So we're just figuring out how to do that. When you bring up humanoids, you bring up a whole different can of which
speaker-2: Worms which
speaker-1: is is this even the right form factor to be collecting data? Right? Is there even a market for this? And if you ask companies like it's
speaker-2: Myself.
speaker-1: yeah, the pragmatic answer is absolutely not. This is it's way too early for something like this. And no one's gonna be buying humanoids and putting them in in homes anytime soon or even
speaker-2: And every
speaker-1: And there's a hype bubble around it that's going to pop. Everyone inside the industry believes this and knows this. Even people who work at the humanoid companies believe this. But there's
speaker-2: They know that
speaker-1: no market for their product right now. And so when
speaker-2: the investment dollars dry up, all those things are gonna pop.
speaker-1: Right.
speaker-2: Should we be collecting
speaker-1: You know, data on on what it is to clean a house?
speaker-2: Or something.
speaker-1: Probably, yeah, 'cause we still want to train how to do that. It just must be a humanoid that does it. You know.
speaker-2: That may not be when you want to think about pragmatic ways to roll out this technology, it'll start in industrial space. Right.
speaker-1: Like you'll That's bolted to a podium leaning over a a sink or a torso, for instance, you wouldn't have the legs on it. That's just a waste of technology.
speaker-2: If you want to have arms that are assembling something on a factory floor or an assembly line, it makes sense to dangle those arms on the gantry system.
speaker-1: Right and have them slide going up and down and rotating, and remove at the edge of a table.
speaker-2: So data collection will probably move in the direction of the way this stuff deployed. But currently the only people that can afford to go
speaker-1: do mass data collection are these companies that have billions of dollars in investment.
speaker-2: Well,
speaker-1: At the moment, those are humanoid companies, right? So you're hearing the buzz of them.
speaker-2: Everyone else is doing data collection.
speaker-1: Right. And then everyone's doing egocentric action because it's cheaper and they're trying to resell it. And so you're just seeing kind of just different areas to figure out what works and what doesn't work.
speaker-2: Yeah.
speaker-0: I just try to wrap my head around these numbers that I'm reading with regards to humanoid production. Why are they planning to produce so many if they kind of know, well, we're just producing them for I don't know, for who are they producing them?
speaker-1: That's very good question. I'd like to know the answer myself because I don't think anyone's gonna buy them right now or because the company's paid to put the robots in those places, right? See people please
speaker-2: Any of those deployments that are out there themselves pay prices. They're asking to please try this product out for free, and those test cases don't last very long, right? Take it back. It doesn't it would make sense some
speaker-1: Then they're like back, don't do anything, right?
speaker-2: more market research and found out if people actually wanted to buy the product first. But to date, I don't know anybody who actually wants to buy a humanoid to deploy a useful scenario be an ROI.
speaker-0: Right.
speaker-2: Why are billions of dollars going into this industry factories to build tens of thousands of units of a product that no has raised their hand and said they want to purchase yet? It's backward. That's why a lot of people in the industry
speaker-1: The industry believed that when the bubble
speaker-2: pops all of this stuff will be chopped up for parts, right? So it's almost like human rights belong in research.
speaker-1: Long in research.
speaker-2: We should be trying to solve these problems in research. And it's like using the goal of going to the moon, we can learn how to make
speaker-1: It's like choosing so that we boosters and computers, right? To do something very hard.
speaker-2: Very challenging and very hard. It's good to have goals, right? I like robots trying to play soccer. Awesome, because it's like here's a goal that's really hard to solve it. That have excelled really well at making robots balance and difficult of sitting karate
speaker-1: And let's you know, and you have companies like Unitry making can be in the most situations. They can do
speaker-2: and ballet and pull buses and they do all these.
speaker-1: Yeah.
speaker-2: But it's a research challenge at the moment. There is a humanoid to pull a bus,
speaker-1: Isn't a use case for
speaker-2: right? And they're not going to be in homes looking at the things anytime soon.
speaker-1: Cool plus A soon, right?
speaker-2: So there's kind of a wrong application of the technology at the moment. Should we be trying to do these hard challenges? Absolutely. Should we be trying to figure out how to make better actuators, how to make better algorithms? All of these are
speaker-1: Walking gates things, sure.
speaker-2: we should. Should we be telling the world that we're gonna make a factory with ten thousand of them buy them? Probably not. And so a lot of people are gonna lose their ass too
speaker-1: Because people are gonna buy and so lose their passes in these investments. 'Cause it's
speaker-2: early for this technology to be deployed. Someday We'll have legged robots doing something useful
speaker-1: And humanoids doing use.
speaker-2: for us. So the argument now is I make legs when we flat,
speaker-1: Why made the world?
speaker-2: right? So legs are useful because robotics is the art of reverse engineering nature, I like to say.
speaker-1: We like to say it so much we marked that phrase.
speaker-2: So if you look at anything we try and develop in robotics, we're basically looking at reverse and reverse engineering. So for instance, we've used LIDAR and sonar to do navigation for decades now. And it's almost like we're cheating.
speaker-1: Look at nature.
speaker-2: Nature has a few animals that use that. You know, we have bats, we have dolphins, but by and large nature never efficient and it works in almost right so when you look at our robots navigating now we're starting to use vision more than we're using
speaker-1: using LiDARs. Like we'll probably always use LiDARs since it's a great tech
speaker-2: technology that supports it, but vision is getting more and more important, right? And if physical layout
speaker-1: If live in eye, we're putting cameras on manipulators so we can see what we're doing. Not just talking about like slam navigation in an open space. Of us mimicking nature, nature do this and let's start mimicking it. How the brain thinks. So we're starting to make intelligence by
speaker-2: So that's an instance of learning how to do AI is us reverse engineering artificial intelligence.
speaker-1: learning how to do that.
speaker-2: We have yet to figure out how to make skin, right? Very difficult, yet to make a muscle, an elastic muscle that can electricity sort of compliant,
speaker-1: Right. There's a lot of researchers on that. It's a very difficult and we've yet must restrict with electricity and be sort of
speaker-2: right? So if you look at locomotion, wheels,
speaker-1: We use
speaker-2: because we've made the world flat and they're incredibly efficient. And it's because the main motor we know how to make rotates 60
speaker-1: Three six.
speaker-2: and it's with heavy magnets and and right.
speaker-1: Wires.
speaker-2: Something like a car makes a lot of sense with a current internal combustion engine or an electric motor driving on a flat surface. So if you have robotics, we have MRs, it's already a solved problem. They all use
speaker-1: in a warehouse and use wheels, right?
speaker-2: so why would we use legs inside a flat environment? The wrong
speaker-1: It's the
speaker-2: tool for the job, right? If you're gonna make a robot that needs to
speaker-1: It's
speaker-2: scamper through the forest, now you don't want wheels. Now
speaker-1: And
speaker-2: you want to reverse engineer nature.
speaker-1: nature did it with legs.
speaker-2: And it makes sense there, right? So the argument is we'll have sta
speaker-1: is how you handle tears maybe? And I don't know what the answer to that'll be. It might be some sort of wheel that can make its way upstairs or
speaker-2: Or not.
speaker-1: they might be arguing that you have to have legs for that. Who who knows?
speaker-2: But as long as you're staying in a flat.
speaker-1: Area, wheels get the job done, right? People wondering why we're making humanoid legged humanoid robots other than love
speaker-2: Then for the
speaker-1: of making them. No, this is always super insightful. I have a question back on data. You mentioned egocentric video data and of course data robot data as well. We've seen recently approaches where UMI data becomes more and more important and becomes like an entire step of training between egocentric data and and tele op data. Is it something that you also observe through what your partners might be asking you or not? And what's your take on it overall?
speaker-2: Yeah, yeah. So that's a third third way of doing it. And we have what we'll call the Trumi, Tross and Robotic Trumi product
speaker-1: So we actually have a problem.
speaker-2: we sell for that. And that just has the GoPros on it. And the next version of that eccentric
speaker-1: We're gonna have our own egos.
speaker-2: headband with matching cameras that the hand grip. So do the gen two of the True synced so you
speaker-1: And then it'll all be data so that
speaker-2: have the hardware synced together. So then people can go collect that data when they want that. And so that's us reacting to the market saying
speaker-1: saying that they want that data and that if it's very valuable to have
speaker-2: So we're
speaker-1: some early prototypes of that being made right now. Well, we've seen a lot of value in that. It's gonna be a bree.
speaker-2: So that is us reacting to that as well. So to answer your question, yeah. And I think it's the blend of all three.
speaker-1: Like I don't think anyone is gonna really
speaker-2: truly win in the end. And I think that eventually AI will get so smart that you can just put cameras sort of around somebody they may not even have to wear them on their head anymore. And it'll just go, Yeah, I understand that task now. But
speaker-1: I understand
speaker-2: once the AI is getting smart enough to operate any type of embodiment or to look at something
speaker-1: Something
speaker-2: from afar and then f put themselves in its reverse sort of engineer how they're doing it, which is how we it
speaker-1: Either way.
speaker-2: right. So you have you can watch somebody how to do a task and then jump in and start doing it yourself, right?
speaker-1: Do attack start doing that smart ability as well.
speaker-2: And robots will eventually get that ability as well. So I think most people we talk three
speaker-1: That we have to think it'll be a blend of all three.
speaker-2: and
speaker-0: Back to deployments, what do people that nail deployments do write if it works?
speaker-2: Focus on a small task thing.
speaker-1: One thing.
speaker-2: So this idea that you're gonna have additional models that has so far proven to be very difficult and hard to so the people that are having the most successful tasks and just trying to perfect that task, collecting
speaker-1: foundational model can do everything is to hold off most are narrowing down on very specific tasks tasks.
speaker-2: data around that task, sticking to one embodiment and that will get really close to a deployment that is actually useful.
speaker-1: That's where we're seeing people release our OIs to like useful. And we still have
speaker-2: You know, the world at large is still trying to figure out what's gonna be
speaker-1: And I hope they're using trust.
speaker-2: of the Roomba of physical AI super awesome case that just takes off. And somebody's gonna find it and hardware when they do, but everybody's hunting now and figure that
speaker-1: Right. I'm trying to figure out.
speaker-2: out. But narrowing down your scope is absolutely the most important is can you do it with simple grasping techniques?
speaker-1: Minks.
speaker-2: If you can do it with a parallel gripper and a parallel gripper only without having sensing ears yet, even better.
speaker-1: Great. And if you can do it without being on the finger, right?
speaker-2: So it's gotta be a simplified task that uses a parallel gripper that's very focused. So our friends over at Yandu have deployment that they have, which is package boxes into
speaker-1: released small bugs into plastic bags and
speaker-2: call a polybagging machine. And so there's a use case where it fits all of those criteria. And so they're already offering solutions purchasing.
speaker-1: So you don't believe at all in physical intelligence approach of cross embodiment?
speaker-2: I think it's a long way off. It's sort of shooting for the moon right out of the
speaker-1: Date.
speaker-2: and it's difficult to do. And it's not just Pi that's doing it to people
speaker-1: There's a bunch of
speaker-2: that have that sort of goal, but
speaker-1: You know, we've seen
speaker-2: so far that it's it's a hard problem to solve. And so if you narrow it down, it's easier to get higher and higher success rates. And if you asked any of those companies, they'd tell you the same thing, right? But the dream is that it can do anything under
speaker-1: That anything.
speaker-2: the sun. What'll be interesting is to see how the ecosystem evolves
speaker-1: Bulbs?
speaker-2: and it might be that Models can be sold to other companies that work in and they hone
speaker-1: And then
speaker-2: the model. And so it's like, well, this one's really good at digging holes. This one's good at cleaning toilets. This one's good at stuffing. And so
speaker-1: And
speaker-2: you have what I call prospectors, right? Who go out and take a model and then learn how to do something specific with it and have success. And then the model is learning from all of those people. the sun, but also if you're deploying a robot in let's say a grocery store and it's supposed to stock in shelves and package people's orders off the shelves, for instance. Why would that robot toilet like a whole right? So you wouldn't have its brain holding
speaker-1: Need to know how to clean a toilet or day hole.
speaker-2: all onto all of that information when it's focused on this one task. You know, so you can separate that stuff off because it's more efficient in terms of how much compute are you using this problem, right?
speaker-1: For it to think through the right.
speaker-2: So will that be at the model level or will that be at a deployment level of somebody layer of a model? That remains.
speaker-1: be a somebody who is doing an application to be seen, right?
speaker-2: So the model maker is out of the gate kind of shooting for the moon, but here we are two years later. None of them can. So you're
speaker-1: You see people.
speaker-2: starting to focus in on more niche.
speaker-1: solutions and we seem better
speaker-2: Accessory etc.
speaker-0: If you would start something right now, maybe tackle a certain problem that you're seeing, any ideas on that?
speaker-2: Well, if I knew that answer, I would be tackling it, right? So that's why we're serving all the people that are out there prospecting and who's successful. It's probably one of the number one questions I'm asked is where do I if I had the answer my billion dollars.
speaker-1: Or do you think physical will succeed first? And the answer company would already be worth a because I would be tackling that.
speaker-2: I don't have that answer. So that's why I'm trying to sell my shovels to as many prospectors.
speaker-1: as possible so they can go find the answers worldwide problem.
speaker-2: Because it is sort of like a world we need to attack it by just trial and error, trial and error, trial and error until we figure out where is this technology useful. And all new technologies go through that phase, right? When we steam engine, they didn't know what to do with it. They had to do
speaker-1: First came out with the
speaker-2: trial and error until they figured out, we can run a loom or when we figured out that the water wheel turn a stream into energy.
speaker-1: was a great way to energy. It's like what do we do with this until
speaker-2: They figured out, well, if we put this cog on it, it can lift this big heavy hammer that that smashes apple all day or something, right? Every new technology goes through that phase and physical AI is now. I think that the opportunities, a big one, and you're already seeing some success here is anything that automates,
speaker-1: As to a feeding humans?
speaker-2: right? So anything that automates our growing of crops, our processing our creating our meals, our serving of our meals and our cleaning and washing of our dishes afterwards. I think there's great opportunities in places like that.
speaker-1: Yeah.
speaker-2: Because that's stuff that's made by the hundreds of millions. So I'm curious to see if that happens. I've mentioned grocery stores a few times. I know that people are attacking that space trying to solve food delivery, but that's part of this right. And then in factories
speaker-1: We're looking for things in the
speaker-2: that traditional robotics can't do that humans are still doing. So like machine tending to
speaker-1: is a big one, right?
speaker-2: So how do you put a pull a part out of a machine, next part in, make sure it's situated right, and then turn the machine back on and next
speaker-1: Then put the next one and run them.
speaker-2: part? A lot of people are trying to work on that sort of areas where
speaker-1: Or trying to sort of thing. Things are floppy and soft.
speaker-2: traditional robots have a hard time dealing with, so anything that's in t instance a popular example
speaker-1: Textiles for instance.
speaker-2: That is something a traditional robot can't still have to do So you wanna find things like that because physical AI is not meant to replace traditional robotics. Traditional
speaker-1: And then you've got to put everything together and close, right? Can't do and a human do, and it's a mind number job that you do all day long, right?
speaker-2: Very good at what it does, right? So if you have a robot you want it to be extremely accurate, I can do something like put a rivet in a car
speaker-1: But you want to be extremely active to some card.
speaker-2: door in the same spot all day long, very
speaker-1: Very, very quickly traditional robots answer that need already. And so physical AI set that it's the work
speaker-2: is not here to upset work in concert with that, doing the next layer it's not good at by dealing with chaotic environments and things flexible or flex out of a box.
speaker-1: things that are floppy or might show up being poor where you need to do some thinking to to do it, right? I wish
speaker-2: I wish
speaker-1: I knew exactly where those answers were because I would just go
speaker-2: Pick some partners and
speaker-1: right after them and solve them. With as many partners as we can and learning as we can figure out
speaker-2: working is to go to sort of figure out what those are. But I think we're gonna have our first few breakouts in the next twelve months. That's my guess. We're all waiting for the models to get better as well. Now we're starting to see some throughs just kind of mind blowing and how good somebody at learning to do something much, much less training them. We still
speaker-1: And some breakthroughs that are these models are getting something within a few minutes with less data.
speaker-2: have to sort of tie those together to the hardware and then find somebody put the money in to go try and deploy it. business model around selling it as well. Like there's a lot of things to sell even
speaker-1: Lot of things to solve in business.
speaker-2: around the fact of can the robot do it? Right? So there are a sales team the customers to invest in buying
speaker-1: Right. So you've got to find out is there that can go find the groups and convince them to buying the product.
speaker-2: is there a deployment team that can go put it in place? Everyone has a slightly different setup in their environment that's using this robot. how are you doing maintenance robots? Maintenance what are the logistics financials
speaker-1: you know it's using maintenance on these so you have to be able to put together a maintenance plan that they're willing to pay for whatistics and the financial
speaker-2: solutions to all of these right.
speaker-1: of these
speaker-2: How do you spin them up in volume that all price point works? Because it's really expensive. And so how does a company purchase that and
speaker-1: Robots can be deployed at a price point that is, let's be honest, this shit is really right now, right? Robot.
speaker-2: resell it to someone else as a solution that's lower cost than the actual workers doing it at the moment?
speaker-1: wages of a work or doing right.
speaker-2: The only time a company will adopt a new technology
speaker-1: technology is when it saves them money. They don't do it for the love of the game. They will continue to use whatever the lowest cost is until it's too painful and the new technology,
speaker-2: option is helpful, then they will adopt new technology.
speaker-1: right? Clearly. Now I have a question about the company overall when you mentioned a funding round, I don't know how much you can say about that, about how much revenue you have been making before raising, like bootstrapping, and how much you are aiming for after raising capital money capital money.
speaker-2: Venture Yeah, we're not public with those numbers, so it is to honestly,
speaker-1: To raise the revenue, obviously.
speaker-2: but right now we're doing a small raise to finish sort of finish the
speaker-1: So it's the finish
speaker-2: product line, start scaling the production numbers of it, and then finish rolling out the cloud, right? The cloud at
speaker-1: It's just like in super alpha.
speaker-2: the moment, and then prove that we can convince people to use our cloud with
speaker-1: with our hardware, right, which is something we have to do. The
speaker-2: And that they'll allow us to collect
speaker-1: data and train the data in certain instances as well. Right. So we have to
speaker-2: prove that model out. And so the first round, which is a small round, I'll say it's much less than dollars,
speaker-1: Ten million dollars.
speaker-2: is the goal for that, right? So to prove it. And then in that process of us proving that and showing we have successfully deployed an entire technology stack with this latest hardware and we're doing deployments where we're selling batches of 50 and 100 of these, then we can go and make our case to do a much larger round at a larger evaluation,
speaker-1: Value.
speaker-2: right? Because what we don't want to do is do too large of a raise right now and sell off too much equity before it's necessary.
speaker-1: So you don't
speaker-2: And so it's sort of like raise, build value, raise and try and pace this correctly. You don't like burn up your company in the meantime,
speaker-1: like
speaker-2: right? And over promise and underdeliver sort of scenario and have a down round, right? There's a tricky balance of that when trying to grow a company to decide when, right? Everyone is we're insane bootstrapped as long as we have and I raised debt to get there, right? So it's
speaker-1: because we b as long as and we probably are right. So it's me risking
speaker-2: a lot of debt in order to get there. But it was because we're trying to build a a certain amount of value before we
speaker-1: go and do a raise, right?
speaker-2: So these are the things deciding in a company as you go. If we were a brand new company that had never done anything before and we were just two Stanford grads, then you do the biggest
speaker-1: Do a
speaker-2: raise you possibly can out of the gate, you give off a big chunk of your company, you take a shot at it, and you see how it goes, right? But we already have revenue. We're already selling, right? And so the way that people look at your business is completely different. Pre revenue
speaker-1: Right.
speaker-2: and post revenue are completely different valuations. And oddly enough, you get punched in the teeth for having revenue. People
speaker-1: Avenue, right?
speaker-2: just don't value you the same way as when it's Which we think crazy because we've already built the product, we already have a strong international brand, we're already working with all of the top fortune ones. We already
speaker-1: We're
speaker-2: have built out the teams and we already have a tech stack built, right? So that should in theory.
speaker-1: In order to get the valuation to why
speaker-0: balance between all of them. Yeah, because people have the reference point when there is some existing transaction going on and as we all know, you know, venture capital likes to dream. They w want to dream of the big numbers.
speaker-2: Yeah, a question always is is well where
speaker-1: You making more money, right? So why aren't you already making all this money? And it's like, well, if we were, we wouldn't be asking you for an investment, right? Because we'd already be at that place, right? We need investment to get some. the unique company we are, we've been around for twenty two years, the answer is well, we're
speaker-2: Well,
speaker-1: waiting for AI to show up.
speaker-2: like AI is only three years old. Basically the first nineteen years of of the company you kind of just have to throw out the window. Company's hobby and education kits.
speaker-1: window because it was a company serving with small like hobby level
speaker-2: Really that's all it was. And then once AI arrived, everything changed. And
speaker-1: and we started serving the AI and market changed, right?
speaker-2: so the company itself, we went from a less than twelve people to fifty people.
speaker-1: fifty in three years, right?
speaker-2: And so there's explosive growth. And you might as well just start the timeline as three years ago. But even then we're still actually bootstrapping and that threes models off don't
speaker-1: And we're actually making money throws people where they don't understand how to value the company or think about it. And a lot of PCs I think are waking up to the companies
speaker-2: And a lot of E's are just now quit the fact that hardware companies
speaker-1: are now extremely valuable again because it's right investing in software companies and thinking through software. And so they're starting to reach understand this new
speaker-2: Half the name. They're used to software companies, formulate their models to understand the new world that we all have just found ourselves in. And I don't blame them for it, right? Like told them this was coming and learn it and adapt.
speaker-1: Right. No one told me and and we all learn as we go and right.
speaker-2: But it's quickly becoming a physical world again. And our argument is we love the fact that we're an AI company that's deeply vested in our hardware, because guess what? AI cannot replace hardware, it only makes the hardware more valuable. And so if you invest in pure A companies or software companies right now, dangerous play because it's very easy to make software easier and faster with every passing month. So you're playing in dangerous purely software companies.
speaker-1: game by investing in companies that don't have a hook in something like like hardware or data c
speaker-2: Collection, right? We want
speaker-1: AI to get smarter and more powerful. 'Cause
speaker-2: every time it does, our arms get smarter and they get more valuable. And we truly believe that you have to have the whole stack. If you
speaker-1: Like if
speaker-2: were just a pure hardware company and you didn't have the other parts have to sell the whole system, I would small organic company with the sales of the hardware. But that isn't like we're basically building an entire flywheel and we're in a unique position to do it because we have more hardware in more deployments and more research labs than any other company in the world that's in
speaker-1: world and physical AI right now. We
speaker-2: And because of that, we have the little birds that are telling us what to build and how to build and we're out ahead and people are doing deployment with trust and robotics. And we're also giving the cloud services
speaker-1: deployments with hardware giving them the
speaker-2: so that they can build the models and collect data, right? So like we're also going to be serving the teleoperation sectors of the market, which is essentially labor arbitrage where they're going to put the robots in place and just have them teleoperated to do labor tasks that are having trouble hiring locally for.
speaker-1: Right.
speaker-2: Right. So instead of someone locally thirty dollars an hour to do a task, you can now put a robot arm in place and have somebody in another country that for eleven dollars.
speaker-1: And it's doing it an hour, right?
speaker-2: Well, guess what you're doing all along while you're doing that? You're collecting the data to train models so that you can eventually automate it and then make it cheaper, right? So the companies that are first deployments with the hardware and collecting information are the ones that build the most value through these different phases
speaker-1: Phase
speaker-2: that the market is moving towards while it figures out how do we get the full
speaker-1: automation
speaker-2: Can we follow?
speaker-1: right it's not gonna be fully automated out of the gate. There's gonna be humans in the loop.
speaker-2: using teleop for it. The new product line that has workbench and the pivot and the glides which sit on top of what we call the cockpit. Telly operated from anywhere in the world. So we're serving two industries
speaker-1: Can be world, right?
speaker-2: for that. We're doing the the current data collection industry
speaker-1: Yeah.
speaker-2: and the emerging teleop industry, which works on
speaker-1: Which wrong labor arbitrage.
speaker-2: and we have a cloud service so that we're collecting all that data so that now the model makers can train on that. Right it's an entire ecosystem and that is what person are about
speaker-1: It's a trust robotics.
speaker-2: now and so we're sort of slow image that we're just a hardware company. We are now fully played and there's a message
speaker-1: Nothing off this old company. Now a platform based company's value on that.
speaker-2: because that allows us to
speaker-1: To play in the physical AI space and ride the wave while companies
speaker-2: Thousands of these companies.
speaker-1: arrive and say we need solutions and we need them yesterday to deploy. And currently we're the only company that has a full package and we're the only company in the United States that has that.
speaker-0: I think that's an epic summary. Please consider Truss Robotics if you wanna play around with an arm or something like that. If you have a use case, deployment, anything like that, check them out. This was the Robotics Stack Podcast, episode twenty-one with Matt Troston. My name is Oscar.
speaker-1: My name is Eli. Thank you.
speaker-2: So much for having me on. Bye bye.
