Physical AI and Robotics at GTC 2026 with Diana Wolf Torres

Jonathan Stephens: The worker is very interesting because they are fulfilling a need and their products are actually out there working right now. So their booth was showing moving tile for a company called Fire Clay Tile. The owner of that company couldn't get enough workers at like 40 or $50 an hour to lay tile for him. He just could not get the people. So he brought in Worker and Worker created a solution for him at the same rate, right, that he was willing to pay to lay the tile for him. And that's exactly what they were demonstrating. So what I found interesting about that, this is not that a robot robotic solution was taking work away from tile workers. This was fulfilling a labor gap, fulfilling a need. That was Diana Wolff-Torres. And that's one of the stories from GTC we're getting into on today's episode. Welcome back to the Thinking Machine podcast. I'm your host, Jonathan Stevens. Today's guest, Diana, writes a robotics-focused newsletter called DROIDs, and we're both creators invited to GTC by NVIDIA, the same conference that we met each other back in 2025 for the first time. In this episode, we're going to break down everything that we saw at GTC 2026, the various hardware announcements at the keynote, the OpenClaw, the DJX Spark, And then we're going to go deep into robotics side. If you ⁓ want to jump into a specific robotics company that we mentioned, I'll make sure I bookmark all of the different talking points from this video so you can jump around as you want. Thank you. And again, and as always, and as always, thanks to Lightwheel for making this episode possible. Now let's get to it. Diana, it's great to have you on the podcast today. I'm so excited about this episode because we're going to talk about GTC robotics, what trends we're seeing, what our reactions to going to GTC together this year was. But just to set up some context, we first met at GTC, but last year in 2025. And so we were part of an influencer team from Nvidia to get, I say, what do we get? An All Access Pass. Is that correct, Diana? Yeah, mean, Nvidia has the best creator influencer program of anywhere. It's the support they provide is so amazing to allow the creators to be able to just to do their jobs, to get the interviews that they need to get. And if you need anything or need to set up interviews, I mean, they just make it so easy throughout the week. the best part of being in that program is the connections you're able to make with your other creators. So I met you and I met some other creators who I would not have had a chance to meet otherwise. And it's so great. And so, and I've had some, ⁓ some companies when I'm interviewing, you know, like when I'm saying to you or other people, he's like, ⁓ you guys all know each other. And it is, it feels like we're a family and these connections are so beneficial because it feels like we all cover the industry from a different perspective. And it's just so nice to be able to work together. So. Yeah, it's just such a valuable week and I was sorry when it was over. Well, let's pivot. ⁓ So yeah, we're going to talk about physical AI, robotics. ⁓ We'll start with the keynote and kind of work through what we saw and what we thought was notable at GTC. Now, this isn't the largest conference. I think it was still 30-something thousand people. But compared to CS, it's not massive, but I still didn't see everything. I spent a lot of time, we had a booth, I was at a light wheel booth, so I there a lot and couldn't see everything, so was hoping maybe you point out some things I didn't see and we can kind of go through that. So I prepared some notes with what we can go through and let's just start with the keynote. I'll let you go first, what was the first thing at the keynote that you thought was notable that you wanted to talk about? Now before we get into the keynote, One note I would like to say is so for GTC, it's just this felt larger than last year. I felt, yeah, okay. I felt that as soon as I got into town on Monday, because the keynote was at 11. So I'm like, okay, I'll get into town by 7 30 to get parking. And it wasn't early enough. So, and last year it was, I got such sweet parking right at the convention center. So. There was a lot more people this year. 2025 Jensen said we're going to outgrow San Jose and now comes 2026 and I felt like we were at capacity. He keeps his promises because you know as I was walking in from like really far away parking lot I was like alright that man keeps his promises. Yeah so what what what at the keynote that you thought was notable I have a few things you can throw out what you think first. Okay, well the first thing that was interesting and you pointed this out was just this talk about the measurement of tokens per watt now. Yeah, I do remember 2025 and it had a whole monologue about tokens and the world is on tokens and how we'll be solving problems and tokens are kind of the backbone of how we are going to be computing in the future and that has not changed. If anything, we're getting deeper into that. But because of that, no longer are building data centers, we're building token factories, right? No, AI factories. He called them AI factories. basically an AI factory. And we're now looking at chips not buying necessarily like gigawatt of power or flops or ⁓ how many CUDA cores you have. We're looking at how many tokens per second or tokens per time and tokens per watt, because that's almost more important, right? We're going to be limited by how many watts of electricity you can get to a data center. And then you want to maximize your token output by speed and watts. That was an interesting change. But it makes sense because electricity is part of the limiting factor that you have to think about it like a car in terms of not just speed, how fast can the car go, but really it's about fuel efficiency. So that's why it's so important now is the efficiency. that you're getting. that's why there was this talk about this measurement of tokens per watt and not just talking about tokens or how fast everything is going to go. there was a whole lot of talk about efficiency this year. And along those lines, Vera Rubin, they were showing the, they talked about Vera Rubin. This is not the first time they brought it up, but how that's changing. The H100, if you have a bunch of H100s not still deployed, you may never deploy those. Because the cost of deploying those, even like the BH100 and 200s, the cost of running those is going to be much higher than the new chipset coming out that you may just decide to skip over them. You may have already purchased them, but you don't have the power yet to use them. It'll make more sense just to put the new Verubin chips in there. Which is just mind-bending that they can do that and compute to say, Hey, this new chip's so much more efficient, so much faster. The chips you haven't even installed, you'll just not need to do that. Call it a loss, you'll make more money. When I first heard that, I was like, hey, well, that's a good way to sell your new product. But then when I looked into the math of it, how much more efficient the Vera Rubin really is. If you look at the Blackwell like in iPhone 15, it's the previous generation that's already out. And now the Vera Rubin. is the iPhone 16, it's the next generation, but it's so many generations beyond, right? It's now coming in second quarter, right? I guess it's not perfect comparison because it's not quite out yet, but it's just so much more efficient that you're right. It actually does make sense economically to move on to the new system because it's going to cost you less. It's like if you went from an iPhone 3G to an iPhone 16. Maybe not quite that drastic. That's a better comparison. Thank you. Yeah. It's like, because they're showing like 4x efficiency. And 4x is sizable when we're talking about the cost of running these data centers and these AI factories. So yeah. I thought at some point they'd said 6x even. Was it 6x? I could be wrong. But they threw a lot of data at us in those two hours, so I could be remembering wrong. Yeah, there was a lot of charts. I'd say if you've never been to any keynotes or watched one of these. ⁓ It's tough at the start if you're not from that world. All right. Well, I want to switch gears to open claw because that took ⁓ word on the street that took part of art of keynote. It's still from my favorite part, but I also love open claw. They talked about, know, this is like the fastest growing open source project of all time. I have been following it, but I have not. I'm more of the controversy of it than the actual is it useful? And I went to GTC thinking I should probably learn about it, but it wasn't a goal. Then that's a main topic at GTC. They had a pavilion, which we'll talk about a build a cloud like build a bear for for us nerds. And that's the greatest thing I may have ever been to. ⁓ And I had no agenda to like, you start using OpenClaw. I moonlighted, I spent time with our creator group, but I also made sure I made to every single developer meetup I could. So I went to a bunch of those because I also like to work with the developers myself. And they were telling me all the interesting things that they've been experimenting with OpenClaw. Now, none of them had necessarily said, I'm using this to make a business or make a product. But seeing what they were doing inspired me to say, I need to actually understand this. I bought a DGX market though. I'm very impressed by that, by the way, when you told me that. And I started experimenting. And again, I feel like it's kind of like tokens in AI. If I don't start using agents now and sub-agents and realize how this works, there may be a day where all of a everyone that's doing meaningful things are using these kind of systems, and I am playing catch-up mode. So I always love to be... experimenting with new things and even though I may never ever take the spark and make a product out of it, the learning that I'm getting out of it is very, very valuable. Learning how an agent can talk to several other agents that can do tasks for me and ⁓ I know a lot of the tasks I'm having to do, it's not doing very well but they will get better. ⁓ I've seen that already happening so. And that's what's interesting, what I've heard from people who are now, let's say, a little more advanced in agents since it's ⁓ an open call, right? Since open call is still so new. But the key is to not have one agent, right? But the key is to have multiple agents. And so you have your lead agent, let's say the agent who runs your life, your butler agent that you're naming, you know, Bob the butler, right? Chief of staff, like to Your chief of staff, exactly. So your chief of staff. is keeping an eye on your other agents, each of which have a specific task. So you have one that's responsible for keeping of your editorial calendar, right, because you're a creator. And then you have another one for just being in charge of your work calendar, because you're also a chief evangelist and you have a lot of responsibilities at Light Wheel. But then you have another one that's called your dad, right, or your dude dad, right, or something, whatever you want to call him. But that's your other agent, because you're also a father. Right and so he keeps track of your family calendar and all your family obligations and and maybe the renovation you're trying to do on your bathroom or whatever it is you have going on in your house that has a lot of logistical things so you have multiple agents right rather than one agent trying to rule them all so Yeah, I've not tried to build a club. That's just that's just what I picked up. So What what what I've been learning for anyone who's interested on this podcast? ⁓ Well two things talk to someone who is high up in Omniverse world in engineering and he was talking to his open claw from his phone at our light wheel party. He's like, sorry. And then he's like, I was just, you know, sorry, I was just giving some instructions. And I was like, what is this guy who was pretty high up in Omniverse engineering world need an agent for? He's got employees that can do all the work. But I think, then I got my spark and I saw him the next day. and he's like, now you need two. One's not good, you need two sparks. I know he wasn't trying to sell you hardware, but he was running the Omniverse on one spark and the agentic layer on the other spark. So we had the full 128 gigabytes or a large, large as you can get on these sparks model and then have all the RAM on the other one to run like reinforcement learning on the other spark. And that's what he has been doing. He's been doing his own fun experiments and learning how agents are working with Omniverse. from his phone while he's out and about, while he's at work, he's chilling in. And it's just like, that's where my brain is. like, how am going to use this as an orchestration layer? And so he can have an agent that is an expert with Isaac Sim, an agent that is an expert with Isaac Labirina, an agent that is an expert with Cosmos, all the different like omniverse frameworks. You can have sub-agents that are, that like, you can load it up with all the instructions for just. running that thing, and it can be an expert, and always have the up-to-date reference material for running that. And then the other agent, you can make these models smaller, but just fine tune them. So I'm learning. And that is like what we're doing. We build an infrastructure layer for simulation, for physical AI. And so we have building a world, we have running simulation, we have evaluating results. Those are like the three main parts. But in the end of the day, We're going to put an agent in the middle because we want the agent to orchestrate for you. We don't want people moving code, moving code snippets around, moving data around. Just let someone else do that, someone else is an agent. So that's why I bought this thing. I want to learn how that's going to work. So yeah. I'm kind of picturing home offices now with like multiple sparks stacked up because some people said they have multiple Mac minis because they have so many agents. But of course, that brings you back to the question of are you going to run out of tokens? Yeah, so the way ⁓ so Anthropic right they had their announcement on Saturday that like noon time, right? We have this cutoff and it's no longer an all-you-can-eat buffet with with ⁓ tokens if you're running it through through Claude and thropic ⁓ but it's ⁓ the way their pricing system was set up was for humans because there's no human that's on ⁓ Claude or you know, code 24 hours a day. It's set up because there's pauses because we sleep and we eat and we go out with our friends and hopefully do all these things. But an agent doesn't do those things. So, and that's why the tokens were going all over the place and they had to change the whole pricing structure. yeah. Part of what I'm doing is I'm using open source ⁓ models, open models. I have no cost per token. But I know like once I master this and I want to hook in You don't start paying for tokens because you can still use anthropic you set to pay the token rate right not the yeah, not all you can eat rate so I can start doing that if I need that or I feel like I want that or maybe just some of my subagents that need access to the full Opus 4.6 or whatever it is at the time models can do that and hopefully they're not running 24 7 only running when they need them to But this is a good way for me not to run up a huge token bill because I'm hearing people are running up like hundreds of dollars Easily and before you know it this Spark is not cheap, but you could easily clip $5,000 spark. Token fees. Yes. If you're just letting it go all day long all the time and not even thinking about it. So it's a good sandbox for me. And I can have all of my data that's proprietary from our company not leave. So that's a big yes. Right. I am not giving it for my code. I am not giving OpenAI my code. That's yeah. That's the beauty of it. And runs CUDA. So I could, people also, why didn't you get a Mac mini? Or Mac, what are they called? The Mac Studios. I made their backlog now. it's Mac mini, yeah. But it doesn't run CUDA. So, that's a big deal for me as well, because I didn't want to run Omniverse stack, and that's all CUDA's that are rated needs the CUDA framework. all right, well, let's talk about robots at GTC's keynote. There wasn't much, but there was some. There was some. Yeah, so it came toward the end. But there was Olaf. Right. Last year we had the little walk the guy. Right. No, but even before that. So it started with, yeah, so the physical AI conversations at the keynote started with the talk about autonomous driving, which of course is a very important part of it with the discussions about Alpha-Mio and where that's going. Right. So that's part of it. ⁓ And then it did get into showing the robots. ⁓ then at the, yeah, and who their robotic partners are. ⁓ You know some of the interesting ones the construction ones bedrock robotics and and Caterpillar which is really interesting to think about, know, heavy machinery driving around on its own. That's Fascinating. Yeah, talk about Caterpillar. Yes, Tony Ward right before I joined light wheel I went to a technology conference for for mine and ⁓ Those two things don't mix. It's like military intelligence, right? ⁓ You know, people go into mining to make big rock into small rock and dig things up and explode things, not to work on tech necessarily. So a lot of them aren't tech forward. There are CTOs there. There are people who are very tech focused in these companies, but only on the large companies. And Caterpillar was co-presenting with the company and they had done a trial with these autonomous piloted machinery in one of their quarries. And... It was an amazing presentation showing how it was working and everything was kind of orchestrated together. And the first question is, are you worried about safety of one of these running over someone since there's no actual driver? And then that's when ⁓ there was almost a smirk on the guy from Caterpillar who was like, when everything's autonomous at the quarry, there is no humans on the quarry floor. So no one can get ran over when there's no humans on the floor. And I see that already now in robotics with warehouse automation as well. Yeah, the warehouse could potentially get damaged, but there's no one in that section of the warehouse where the robots are running. We have 100 % safety rate. So that's really exciting. And I can tell you the mining industry is like one of the slowest adopters. And when they heard that, I could see so many wheels turning and people saying, I could probably get a budget for this on safety alone. Just the fact that you no longer have humans. I mean, no one wants a human to get even hurt, let alone like a... Major catastrophe or someone really, you know have it like it ran over a die and that's so that to me I keep seeing this in the robotics or people keep talking about efficiency of how quickly we can turn over things how long it can do things no one's talking about the the fact that the manufacturing and mining have been putting in automation just because people's backs have been getting blown out and now you have these comp worker comps and they don't want that they want their safe the safety of their workers they want people to be healthy kind of like a family when you're working in mining and so I don't hear that safety talk as much in these robots in the factories as much. I'm looking forward to that. And I'm always hoping having Caterpillar at like GTC is kind of like a way to talk about safety and getting that. So I always ask that question when I talk to people who are doing industrial robotics. I always ask them about their human safe aspect or when is this going to be human safe because those ISO standards are still being approved. And all the CEOs and the founders, all say that that is something their customers always ask about. It's very top of mind. All of them have different methods of keeping the robot separate from humans right now, ⁓ where like agility uses light, which is very interesting. lot of them use physical barriers, but the robots have to be kept separate from the people because they are not yet human safe. And they're not rated to be human safe. Yeah, I don't think people see these robots on social media and realize that they are sometimes six feet tall and 180 pounds and have the strength to throw a football across the whole entire warehouse. These things are not soft toys that when they fall over, ⁓ oops, we'll just pick it back up. No, they're heavy. They have high torque motors in some of these. So I posted a video last night just on my YouTube because I figured it was just good fodder for YouTube shorts. But what it was was the H2 robot from Unitree. So Unitree has the G1, which is, you know, it's really cute. It's, you know, it's the one that the UFB uses for their fighting bot, right? It's size of a child, right? Everybody loves that one. But they also have their H2 robot, which really has the stylization of something from 50 sci-fi, but it's, it's tall. It's six feet tall, just like what you said. And the reason why it's so tall is because it's designed for factory work. But at ETC, they had it out in the middle of the corridor, just standing there, right? It wasn't doing much, right? It was just standing there. It wasn't really even moving. But I got footage of people just walking by it and people were walking by like this. Like they were so unsure about it because the average height of a person is between five, five and five, seven, if you average everybody out. And so this thing is half a foot taller than the average person. And to walk by a robot that is that much taller than you is very intimidating. So it's a little unnerving. I'm trying to find a photo of a former guest, Chris Paxton, next to one. And then I posted a video of the full thing standing. ⁓ He's tall. Chris Baxson, he's not in the five-five to five-seven range. I think he's at least six feet tall. Standing in the Nexus thing, I think it was actually taller than him. And I'm like, wow, this thing is... And it's really skinny, which almost makes it feel taller. I don't know if that makes any sense. Yeah, and it's because I looked up the specs on it, and it's got very strong actuators, you know, as you'd want for something. And it's got good battery life and everything you'd want in a factory robot. But it's just... when it's around people and wisely they didn't have it moving a lot, it's just, it's unnerving. It's not your ambassador robot, let's put it that way. I didn't see it moving. When I saw it, it was just standing stationary, like it had been powered off but standing up. Maybe that's what it was doing. Yeah, but I do know some people saw it move. Okay. And then... Another person I he's got a great YouTube channel, and I'm forgetting the name, I will have to link it. He did a great video where he had, it was laying on the ground, and they literally went through the actuators. He's a mechanical engineer, so was going through all those sort of aspects of it, and I thought it was very interesting, because he was breaking down choices and how they built this thing, because it has to be, it's tall, can't just build it the same way their G1's built because of that. The torque you get on a lever, from a long armed industrial robot is different than a short armed industrial robot. Right, something with that kind of strength, Yeah, I can't find the video, there was a great video we had from GTC. I think the problem is, though it was only a couple weeks ago, it feels like a year ago we were there. That's right. Well, you will link to it in the comments then. Yeah, yeah, it's great because he posted a picture with it and then I post, ⁓ here it is. See if I can get this up on the screen. look forward to seeing it. Let's see here. Chair so people who are listening in on this I do suggest that maybe you watch the YouTube version ⁓ So there it is there's Chris and I think Chris has got to be Six foot about so this robot is taller than him and it's really skinny looking guy. Doesn't it doesn't look strong Yeah, so in a little video that I posted I made my my assistant who's maybe five seven stand next to it and he was very Uncertain having to stand next to it because of the height difference yeah, I ⁓ I I Took a video of it vertically panning as well up and it made it feel like it was tearing over me so yeah, yeah, so now let's talk the opposite end of that thing was Was Olaf? Which ⁓ I think they had they had on stage, but then shortly after I saw a video of it falling over at Disneyland Paris so Which actually, in defense of, of course I wasn't at Disneyland Paris, but when that fell over, they had such like a human safe circle around it, which, I mean, Disney does that very well. They always have handlers around their characters, but it was almost like a trust fall, right? Like when it fell over, had plenty of space to fall. It had all its handlers around it. So it wasn't like some of these other things that you've seen with, again, with your Unitree robots, but... where they've been too close to people and they've kind of whacked people in the face. with either their hands or their feet. I'd say Disney, if you dive into the imagineering going on around the robots they're building, you will find quickly that their papers really are about safety. They did an interesting one about falling whimsically. So if they know the thing's gonna spontaneously fail, it like dives into a way that it... It fails and looks like it's trying to fall like, like it I don't know to do it like falls like it's like laying down or in a funny way But it's doing to protect his battery But also if it knows that there's humans around it can fall in a direction that there is no humans So it has to constantly be aiming around it for an open safe route So if it feels like it's getting outer distribution and boy it's gonna fail It's gonna fail where no one is which I think is a beautiful thing Maybe that's why it looked like a trust fall. Yeah, I remember at the Humanoid Summit, I watched the Disney robotics where they were talking about the engineering behind Olaf and the engineering behind it is beautiful. It's amazing how much thought they put into it. And I think when Olaf came out onto the stage with Jensen, I mean, he had a little bit of a dialogue with it. ⁓ I don't think it gave justice to how much engineering is actually in Olaf and all the things. that it can actually do because it really is an extraordinary robot. They had sort of just like, a little chipper, a little back and forth. yeah, but and then it looked to me like they had to cut Olaf's mic because he was going, Olaf was willing to just keep chatting and chatting and then they had to drop him back through the stage and he was just still going. But ⁓ he is an extraordinary piece of robotics. Yeah. ⁓ They pointed out that it was possible because they trained it in simulation and they keep, NVIDIA likes to talk about Newton, which is their ⁓ humanoid forward physics solver for their simulation engine. ⁓ I'd be remiss to mention that Light Wheel has joined the technical steering committee for Newton. ⁓ Because we're really trying to push that forward. Because the better high quality physics that we get in these engines, the more we can do in simulation. And I think what's important about that, because back to safety, I can set up a situation where there's a bunch of children around Olaf and it has to fail and what is it going to do? And if it's in simulation, no child gets hurt. ⁓ It's a very, very safe environment. So yeah, it's pretty cool that they're able to train a lot of that in simulation and then just let it go. It's just, it's out there in the world. Yeah, because it bothers me when I do see these kind of... Unitry videos and they may be the exception and not the rule because of course anyone can go out and buy one of these and tell off and bring it out into a crowd but where you do see it going through its tricks and and and hitting somebody and hurting somebody because then you have to think that that child then is going to be afraid of robots right because it was it was just like ⁓ how cute and it's enjoying it and then it's just out of the blue it got it got smacked and that's that's traumatizing for a child It goes back to when I used to work with the DOTs with drones back in the day and they would not fly drone anywhere near a highway corridor because even if they did everything they could to make sure they were safe as possible one drone that decides to all sudden malfunction and veer towards a highway and hit a car going 65 miles an hour even if no one got hurt it would be on the news and it would change this perception of safety for these and it may mean the end of drones for the workers at A D.o.t. That could be using it to improve safety for their own selves so it's there's a lot of perception right so even if one of these things fell and hurt a kid That could be game over for robotics Deployed in in parts for a long time because you just to rebuild that trust and no one wants to kid get hurt anyway, Yeah, it's it'll be a tricky game that they have to play of how can we get this out there safely? Effectively because I I do think a lot of people don't realize how these things are heavy and spontaneously can move. Yeah. Is there anything else from the keynote you want to talk about? Or before we can show some other notable what we saw on the show floor? I think we've covered the keynote ⁓ because that was it was it was a great keynote as all. you know what we didn't cover was that one trillion dollar number. ⁓ yeah. ⁓ yes. Get over that. I can sell a trillion dollar worth of Blackwell and Rivera Rubin. Yeah, but it was also the way Jensen delivered that number. He was he was quite the showman when he delivered it. It was kind of he he drew it out with great pauses and it took him like two minutes just to deliver that number before he came out with that, you know, one trillion dollars. And my brain was still trying to wrap itself around like, wait, wait, what? does that number include? What does that number mean? Like I couldn't even wrap myself around the math of what it is. But as you correctly stated, that number is just Vera Rubin and Blackwell. That's that hardware number. Yeah, which is doubled over there a lot previous year. I believe yeah 500 billion Yeah, cuz he then put up a chart and the slide showing what it was What it's projected to be and then he said that that number may even be short may even be a conservative estimate Well, so anyway, it was it was a mind-blowing number. I'd say he's good at putting sky-high numbers up there and everyone doubting him and then beating that number so if there's any company that'll do it it's probably this one just I mean they have such an advantage as well like a AI world that keeps celebrating is just really being powered by their chips and no one else is competing yet so it's not theirs to win. products right that's that's that simple. so again I don't want to be biased being an Nvidia in full invite an influencer on their team, but it's true. It's like everything I've bought around here for computing has been in Nvidia cards. It says well before I worked with them, just because it worked. Jensen had an interesting quote, and this was during the press Q &A's. I think it was during the press Q &A's, but he had just said that a lot of you were Nvidia customers before, you know, just growing up. because you'd asked your parents for Nvidia cards or your kids have been asking you for Nvidia cards. And I was just thinking like, you know, there's so much truth to that. I have a son who's a gamer who, every new generation of the Nvidia card was something that ended up. in our house, but it profoundly influenced him. He ended up going into engineering. Of course, he still loves NVIDIA to this day. My first NVIDIA GPU was in college and they had just started coming out. I'll tell you how long that... He wasn't wrong about that. Right. Then my first one at work, my first computer... Well, I always had NVIDIA GPUs everywhere I went. It's weird at work too. When I worked in computer vision, I got a workstation laptop with a quadro card, and boy, that thing was amazing. It only lasted like 90 minutes, but you kept it plugged in, you had a high speed. Oh, good. Back to efficiency. I still have one those computers that last 90 minutes. This is 1590 laptop, and it goes fast. One thing I talk about with robots as well, again, if you're watching this, you'll see I get a Ricci Mini while I was also at DTC. Oh my gosh. I didn't you had one. by the keynote. let's just say I took it from our office. Wow. I had no idea you owned one. I commandeered it from the light wheel offices. It was sitting in a box. No one had a plan for it, except for me. So I want to be my voice agent for my open client. ⁓ my gosh. you have to be It's exciting and they like it so far. So, yeah, I'm very excited about that. Okay, let's swap to some of the interesting things we saw on the show floor. barely got a walk around it I think I had like 30 minutes of a whole time. I walked around it, which is fine So you're my eyes and ears. I can start with worker who I also had a chance to talk to Ken Mackie the CEO we actually presented at the same talk which was kind of fun that we got to just talk but Tell me about them like I I again didn't actually get to see their see their booth in action Okay So Worker is very ⁓ interesting because they are fulfilling a need and they're actually out, their products are actually out there working right now. So their booth was showing moving tile for a company called Fire Clay Tile, which exists in, I want to say San Benito County, right? They're really local around here. The owner of that company couldn't get enough workers. at like 40 or $50 an hour to lay tile for him. So he needed tile workers and he just could not hire them. And this was in an area where, you know, he was willing to pay, but he just could not get the people. So he brought in worker and worker created a solution for him at the same rate, right? That he was willing to pay. to lay the tile for him and that's exactly what they were demonstrating. So what I found interesting about that, this is not that a robotic solution was taking work away from tile workers, this was fulfilling a labor gap, fulfilling a need is what he was showing. So yeah, it was a very compelling solution and I think we're gonna be seeing a lot more of that in all these places where there's labor gaps, where robots are coming in and they're filling solutions. I saw that in the mining industry. Everyone was talking about autonomous trucking. Like, it's going to take truck-carers jobs. It's like, no, we cannot find enough people who want to go into trucking. Some of these skilled labor jobs, there's not enough people wanting to do it. So you only have one other solution is to make Robox do it. I like how that they're solving a need and not just, hey, we built this. Love home robots, but I use as an example like everyone at home can fold their own laundry There's not like a yes like a labor gap at my house It's just I don't want to do it and I don't want to prioritize my time for it But it gets done every day or every week we get all the laundry done But versus this is like I could do more business at my company if I had more workers I can't get enough staff to to fulfill all the the sales I want to do or all the jobs I want so that's I like that he comes in with that that That mindset at Worker. So how are they doing that? How are they laying tile? I saw they had robotic arms. Yeah, that's what it was. So it was a robotic arm that was a pick and place solution that was laying down the arms. yeah, it was a... So then is that like a... You can't get workers, but then you have to have someone who's managing the robots. So would you hire workers as a service company? Is that how they work? Do we understand that? So Worker ⁓ is robotics as a service model, which is exactly what you guessed. they, ⁓ have here they're paying $25 an hour. I'm not quite sure if that's exactly right for that one, but it doesn't even matter. So it's, ⁓ you pay only for the productive hours that the robot is working. ⁓ So it's learning ⁓ with no coding required. but it's training in NVIDIA Isaac Sim using synthetic data. So there's your NVIDIA tie-in. the demo we saw at GTC, it was moving ceramic tiles for fire clay tile. So a California manufacturer. fire clay is, I did remember that correctly, they're in San Benito County. So they're in Aromas and there the unemployment runs at 2.6 times the national rate. So it's not that people didn't need work nearby, but the specific skills that were needed, which was hand glazing, ceramic finishing, and precision tile work, represents a dying craft with almost no entrance. So to Ken Mackie, there's a real shortage of skilled labor, so that could fulfill that need. Seeing the robot as a service is gonna be the model that I think the industry adopts early on. Right? Because I would hate to spend a ton of money if not building my own robots. To spend a lot of money on a robot that in a couple of years might be outdated. It just makes more sense if you can have someone bring in the robot as a service, pay per hour, pay something like that. then... Like if your business slows down, right? Or your business speeds up. So, yeah. I'm thinking like my modem at home, right? It's like, I rent the modem. Everyone says, you should own your own. But then I ask people, when's last time you updated your modem? They're like, I need to update my modem? I'm like, yeah, they get the newest speeds. You have to have the newest modems because telecommunications keeps improving. And you're on Wi-Fi 4, not Wi-Fi 6 or 7. And they're like, ⁓ do you know those different Wi-Fis? So it's kind of like, well, yeah, you rent one. Yeah, it costs more. But you always have the best as well. So can imagine a worker will have version 1, 2, 3 of the robot. And over time, they can just swap out the new robots as well. ⁓ that's very profound. Yeah. And ⁓ I was thinking about that tile company and without a solution like the one they were able to get, and of course they were willing to try robotics, their business could have closed, right? They had such a critical need for workers. So this allowed them to keep their business going because obviously they had customers, they just didn't have workers, which is the exact name of the company, right? The name of the company is Worker. So they are providing workers that are just badly needed at these different places. So yeah, so I thought they were very interesting. They were right in the NVIDIA booth. And so I really like companies that are out there actually working, providing solutions. And they were one of them. So they caught my eye. liked that. was like agility last year was in the NVIDIA booth. And they were deployed. They're not just a demo unit. They actually have Digit out there. And it's exciting to see that. Yep, we're getting factories. robots that are out there, not just, ⁓ this one's just a big partner of ours, but we want to make sure we're showcasing deployed. At least if they're only doing a trial, that's fine, but at least they're being trialed and they're not just a demo somewhere. Right. And I do want to say you had one note. You said last, at Humanoid Summit, which we last saw each other before GTZ was at Humanoid Summit in Mountain View, everyone was enamored by Psionic. have a... have a really cool hand. I suggest people look into that company. They are a fun company. They have really good ⁓ technology and marketing. But this time it was worker. And it's funny how I feel like at every single conference, there's like one standout that's doing something that everyone has to go to see. It's like, ⁓ did you see that booth? And you're like, no, should I? And like, yeah, if you don't see that one, you missed the whole conference. ⁓ Exactly. Let's that one. Yeah, and psionic was at GTC, but I think they were only able to make it for for one day Yeah, yeah a deal the the CEO was walking around with the laptop and one of their rope Yes, and and you could like touch the finger and you would see on the laptop picking up the sensors It was quite yes, and then and then you could move the bionic hand. It was a really cool Demo now he stole the show at the human rights on it because he was wearing a doc OX suit with all his bionic hands on it Which was incredible But this was also a great demo because it just gave you an idea for what their robotic solution is, which is a solution for the hand problem, literally. But the same hand that goes on people as a prosthetic is the same hand that goes on robots, which is what I love about that company. So yeah, it's also a great company. All right. Well, let's not keep people, our audience, hostage for... ⁓ Exactly, sorry, I could talk about robots all day. so Humanoid, this is a UK based company. know a real Humanoid. I've met ⁓ at least maybe two or three engineers. ⁓ Very smart people there. Any of those robotics companies, they're full of smart people. That's the really cool thing about this industry. You can't work in robotics and not be brilliant at something. So, because you have to take a lot of really hard disciplines to make them work. But ⁓ what did you see about Humanoid? Again, I saw Humanoid. when they opened up the floor for us content creators and they weren't moving at the time. They were like powered down, the two robots. But I do have pictures of them I can bring up. ⁓ what do you see there? My assistant and I lined up later so that we could try the demo and there was like a really good line for it. So there was two humanoid robots that were working together in tandem and you could order from, it was a tablet menu, so they weren't actually that it wasn't actually voice ordering at that time, right? But they would acknowledge ⁓ by voice so that, you've ordered the dried mango and this snack bar. And then one of the robots would hand you with their grippers, one of the parts of your order, the other would hand you the other part of your order as you go. And they were ⁓ dropping into your hand the parts of your order. And they had to go to a crate because you had a lot of different options. They had to go to the crate and they had to ⁓ fetch and pick the correct option to then hand back to you. So, yeah, so that's what they were doing. was sort of a fancy tandem to robot pick and place, I suppose, solution, you could call it. There we go. There you go. Yep. Handing out the snacks. Yeah, not quite exposed correctly, the picture, but yeah, you can. And these are both wheeled. Yes, you can see it. Yeah, you have to look at the bottom to understand, right, actually they're really quite large. Now, my understanding is these robots come in two different versions. They actually do have a bipedal-legged version and they have a wheeled version and they have different use cases, right? Like you might need to have legs in an environment. They have a really great video where they orchestrate humans interacting with these robots as they go through a storefront where they have legs to the back of the store where they're on these rolling legs and ⁓ how they orchestrate the different robots work together. And I think that's what they're showing here with the two robots. You're not going to get the demo. I got enough from just seeing that they were working together. And that's a really interesting thing is because in the future, we're not going to be having solo robots. I don't go, people don't work in factories by themselves usually. It's a ⁓ collaborative effort. to get things done and so I've seen that that's what they're showcasing. Yeah, because the two robots were pulling from the same bins to pull the snacks, right? So they both had to go back and grab things, you know, out of the bins. yeah. All right. Interesting. ⁓ I suggest looking into those guys. I've posted content about Human Raid. think they're doing, what's interesting, I think the first robot they deployed, which we saw there, They built and trained and deployed in a seven month span. That gives you the time frame in which we're building robots and deploying. ⁓ Now, that's just like V1 that's being deployed. So they're gonna get better and better, but it's speed. Speed's the name of the game with these guys. ⁓ But yeah, I think there's some confusion with their names as well being humanoid. Yes, right, because it's, was teasing you. was saying like, it's like... coming up with a new car and just like, I'm just going to call it car. Yeah. I was thinking about the humanoid. What should we call this? This is called humanoid. is... Yeah. Oh, tried to Google them. Kinetic AI, kinetic with like, KinetiQ AI framework, coordinates multiple robots. There we go. With voice instructions. So yeah, these guys are interesting, you know, because I've seen a lot of people build these G1, Unitry G1s, and they're working on their own. You know these autonomous robots just working as solo solo workers, but it's really gonna be a collaborative robot Future that we're gonna see I feel like that with cars too like we're building these cars that aren't really talking to each other But the real future to self-driving cars is when all the cars are talking to each other and we pull the humans out of the driver's seat And then they just won't run into each other because they will know what each other's doing I love that That is the vision right there. Like, hey, I'm here. Don't hit me. It might be foggy, right? It be really hard to see. You might be coming around a blind corner or in a city. You know, you're coming up out of one of the parking garages. One of the parking garages that's really steep like this and having to go around a corner. But because the cars could talk to each other, you need to rely upon human vision to be able to see because the cars are communicating with each other and know that's there. I love that idea. Yeah, the worst part about self-driving cars is that we have humans driving. Yes. We're most unpredictable drivers. are terrible drivers. People say I'm better than a self-driving car. Well, maybe, but I have a hard time thinking that we're good at it. Or else we wouldn't have millions of accidents a year. Whenever I'm caught in a curiosity slowdown, I always think like if all of us were on self-driving right now, we could all looky-loo and traffic would still be moving. Yeah. Yeah. Alright, let's move to noble machines this one. I actually did talk to I talked to waiting I think we both did yes, maybe did you yep, and he's a co-founder ⁓ They had a very interesting robot that was actually picking things up and moving it ⁓ I I will try to find a picture of one while you tell us who they're what they're about a lot of my while I go try to find a photo for reference so noble machines just came out of stealth for GTC This is Mobi3, which I believe makes it their third generation robot. They are already have a customer where they're working and out of like a Fortune 500 customer. There's some really smart fellows over there. They're from Caltech and NASA and Apple and other very notable companies. But what's interesting is when I spoke to the co-founder, ⁓ Ding, is he was talking about, like he said, what do you notice about the grippers, right? And the grippers are just blue and rubberized and he's like, they're dog toys, they're a dog shoe toy that you could get on Amazon for less than a dollar. And he goes, hey, if you have prime, you can get it by tomorrow. You he was joking. But also, if you look at the footwear on the robot, it's just a pair of sneakers. Like Spencer Fong had come by and he'd even, pointed out, he's like, ⁓ that's a hundred dollar solution right there on the robot. And I was reading up more about Noble Machines, and I guess its first pair of sneakers was from Ross Dress for Less for like $20, but it wore them out too quickly. And so then they put it in a pair of ⁓ caterpillar work boots to give it tougher footwear for when it needed to walk up hills. And so it's just such a neat solution. so... They're out there to not reinvent the wheel, but just find solutions that just work on their robot. So yeah, let's see. Yeah. Here we go. I had a of problem grippers there. There's these grippers, the blue chew toy. Yeah, we've got to find a shot, though, where we can see those really nice sneakers. Oh, I don't have one. If you go to noblemachines.ai, right on their home page as a hero shot, and you can see it right there. You can see those grippers there, though. And they have this nice. They have a structure light sensor on it, a LiDAR unit, really sturdy. They were saying, you know, I thought it was interesting, I just about picking up these heavy objects. We pick up 50 kilograms, I think they were saying, or it could have been pounds, but I think it was talking about kilograms. just think about the distribution across this robot when it picks up something like that, and how instantly it has to be able to adjust to be able to keep balance. We do this as a human, no problem. I can pick up something. I might have trouble picking up something 50 kilograms, but I can pick up heavy objects all day long and just know how to kind of like brace and put my center gravity backwards so it evens out. Robot doesn't know how to do that necessarily at first, so you really have to be good at that. Fast. Yeah, it was interesting to watch it like set it down again and then pick it up again. And so, yeah, it was an interesting demo. Yeah. What was it? .ai? It's noblemachines.ai is their website. Okay, I have two more robots I want to talk about. Generalist was another booth that wasn't in the Nvidia booth, but boy, every time I walked by, which was only a few times, but every time, I couldn't get to it. It was so packed. ⁓ And they had basically a pair of arms with a bunch of objects when I walked up, and then people could tele-ops and see how that worked. And what I think is cool, like we do tele ops and SIM in the light wheel booth, but what they were doing is tele ops on a real robot. And you could see that and it helped people connect like an industrial robot being trained by human. ⁓ And then they shortly after had announced Gen 1, their new model that has 99 % success rate on easy tasks that they have trained it off of almost like very little amount of training data, ⁓ like actual robot training data at 3x faster than the state of the art. So those are some big claims. I like to say big claims, own tests. We'll see how it works on someone else's testing. But also, then a day after I heard about Gen 1, one of the founders had wrote a essay on Gen 1 and telling people that there was VLAs came first. was a vision language action models were a big hit in 2020, like 20, was it 2022, 2024? 2025 world models came and I was like everyone's talking about world models and he's saying like don't think about Vlas and world models think about what you're trying to do and build your own model to solve that and that's how they did a better model now this person who wrote the essay was also one of the very first people to ever create he created the first VLA him in another researcher so the person who created Vlas is saying forget about Vlas forget about that framework build your own models piece together what you need, don't use someone else's off-the-shelf model and fine tune it. You're better to train it on your own data to solve your own problems. And I think that was a really interesting thing, because I think that might be something these really big companies are going to have to pull off to get the best robots they can. But for someone who's not a giant company, still going to have to, they can have the resources to do that themselves. But it just blew people's minds, I think, to see this robot and then... get home, hear the news about how that same robot they were playing with is now perhaps one of the smartest robots on the planet for easy tasks. I say easy tasks because again, there's some really hard things it couldn't do at 99%. And then how much faster and they were just way out ahead of some other companies that felt like when I was there, like the way they were talking to people. Not to say one company is better than another, but they really were talking intelligently at a level that think everyone is connecting with. So I was excited to see that. They had like, their arms were like, like picking up an iPhone and putting it in like the iPhone's box, which is not like a little, it's an easy task for a human, it was a paradox. That's really hard for a human, or really hard for robot. I could also find a, I do have pictures of that as well, that they were doing some. It's just incredible. So that's one I wanted to point out. I tell people if they're listening to my podcast, go look up Generalist Gen 1. ⁓ That's probably state of the art now. ⁓ And that's the crazy thing. We go to this conference and the state of art model comes out the week after and there'll probably a new state of art model coming out the week after that. ⁓ you can't keep up. challenge is to untangle headphone cables. ⁓ here. I got some pictures. But yeah, no, that was interesting. I don't know if you've looked into them at all. Quite an interesting company and how they're moving. So this was their booth. There's their arms and it's hard to see, but I have someone who's tele-opping those little arms. And I think I have a better view. Yeah, so you could tele-op and then there's like an iPhone, some Jenga blocks, and they were basically training this thing. while they were doing it ⁓ and doing these really fine motor things. They also talk about how it could improvise. So like you could be having it do something and then they were using a hockey stick to like mess up the scene and it was having to be improvised on how to complete the task midway through executing a task. So it was constantly updating its model trying to figure out ⁓ how to complete whatever you're trying to do. So ⁓ I thought that was really interesting. So this is some pictures. that we captured from there. So I got to see them. I thought they were good. And then the last one, me and you both saw was Open Mind, but we didn't see them inside GTC. We saw them on the walk from the keynote to... definitely was great placement for them because everybody walked by there on their walk back from the keynote and they had a lot of robots out from a lot of different companies. So their ⁓ Open Mind is open source robotics. ⁓ Correct? Yeah, that's what they are. That's why they're called Open Mind. Yeah, and they were just showing their solution running on all of these different robotic platforms. And the one that stood out to me was the one that just kept going up to everybody and saying, hello, I'm Pam. Hello, I'm Pam. It's just wanting to interact with the people. Yeah. And some notes I took on that was ⁓ I already looked into them. I think they were new to you, but... It's like a strap of Jetson Thor on the back of this and you give it a supercharged brand. I say back of this. you could put a Jetson Thor is what you run open minds software stack on and you put it on like an industry G1 you could put on one of these robotic dogs and you also a booster K1 the little booster robot and all sudden is smart or smart much smarter than it was and They have their own operating system where it can basically figure out the robot and start working with it. ⁓ They have a shared intelligence network, Fabric. Again, that's what we were seeing with Humanoid. It's a way for if you have multiple open mind power robots, they can now talk to each other, coordinate their activities, kick a soccer ball from one to the other and kick it back, things like that, ⁓ which is like almost a hive brain they start to create. ⁓ They also have an app. Ecosystem where you can install apps on your robot. So it's this fun open source ecosystem that you can just literally if you have the budget you can strap this GPU on the back of your Unitree or your booster or your robotic dog and all sudden it's a smart version of it and ⁓ I think that's really cool because there's a lot of things I think people like I think there's a lot of people who are trying to Early discovery and budget discovery. You're like, how can I get a robot to help me? Well, instead of building this all yourself, there's now these open source alternatives that are coming out. can just buy an off-the-shelf robot, throw it off, a decent computer on there, and get an open source brain on it. And you're off to the races trying to make it do some interesting things. Kind of like the robot I threw you on my death. It's like, this thing's open source. And there's a community of apps people are developing. You can just grab it. Now, this can't do anything, all that. compared to a true G1. But the fact is that it's a good way for people to learn and get it to do interesting things without having to build your own software from scratch. But it kind of makes sense what you're describing, right? If one company is good at building the bodies, right? Like Unitree, but you get a G1 and a lot of parts of it are, let's say still, tele-op. but you want to give it this big boost and you have Jetson Thor, which is a really great way to add a brain. Now you're combining the body and the brain and then you can just turn it into whatever it is that you want it to be. And so you're just this best of both worlds approach. Yeah. There's another company doing this called Flexion. They're out of Switzerland. I don't know if you've looked into them as well. They're building the auton... I'm reading this off their website. not that well versed, but they're building an autonomy stack for humanoid robots on command and control from manipulation to locomotion. Kind of the same deal. It's like, you buy this off-the-shelf robot, you put the Flexion brain on there, and now you got a smart off-the-shelf robot to do a bunch of interesting things. They showed like a Unitree G1 walking through this nice Swiss forest picking up trash and navigating around roots and rocks and moss, and it was really cool. It actually makes sense to me, right? Because everyone might want their robot to look a little different, but then again, too, you might want the brain to be a little different. So you're picking these different pieces as a solution to create what you want in your robot. that makes a great deal of sense to me. What I'm hearing here is there's so many robotic companies out there. There's no way even someone who has the droids newsletter can actually follow all the news that comes out. I cannot. I'm actually learning things from you. I'm like, oh, I didn't catch that demo. did not did not know that. jumping out of GTC, unless you want to cover anything else, I just want to talk about a couple new things coming out. going. love to watch the latest news. I'd say if you've never been to GTC and you can afford it, you want to meet some of the brightest people in GPU computing, worth going to next year. Definitely absolutely go if you can. It is not a cheap trip because it is expensive to stay in the value valley when it's not GTC, but when it is GTC it gets even more expensive. So, ⁓ plan early for it, book early. But you can also see all these sessions virtually for free. And Nvidia is very generous with the fact that they just literally ⁓ release almost everything that you could have caught in these sessions for free for anyone after the fact. ⁓ I'm actually just catching up on the sessions now because I had to spend so much of my time on the show floor There were so many people I wanted to interview I think I only caught three sessions including yours But so now I had a whole calendar of things I wanted to catch and I'm just going through them one or two a day because there's such a wealth of good knowledge and I'm like Well, I can catch them now rather than trying to catch them all then so yeah, I mean that's what I was told at the last ⁓ Last year as well was like don't worry about going to all the sessions you watch them later ⁓ Just go meet as many people as you can interview people talk to people that's where that's where you get the inside scoop, but yeah, so now I'm gonna talk about a lamp a robotic lamp now that sounds crazy yes, and ⁓ This just came out a few days ago. Have you not seen it yet? Okay? Okay? No, so loom came out last year, but that's that's what I want to talk to you about There was a little bit of a scandal with it last year. Yeah. I don't know that. But all I know is they released their video of their lamp and pricing. the video is very beautiful. It's very elegant. I don't know. It looks like a pair of praying mantis arms. Yes. But when you watch the video itself, right? So you have this gorgeously beautiful Hollywood couple. And she's she's she's I don't know what she's doing to that wall. Right. And it's this enormous mansion. And you know, they're dancing and there's this waterfall sounds in the background. It's folding, you know, a relatively small pile of laundry. We don't see it try anything like a fitted sheet or anything challenging like that. Okay, he's reading a book, you he's living a very nice life of leisure. Okay, it's folding a shirt, right? So, okay. And now we don't know what speed it took to fold the shirt, but, but it did. NX. did eventually. It's making the bed. There it goes. It just finished making the bed. OK, now, right? Right? I guess that's what now they're dancing. So what we do in my house between OK, exactly, exactly, exactly. Between right, right. OK, now there's the lamp. There's the beauty shot of the lamp against the sunset. Right. OK. So there's there's a whole story behind this company and I researched it and it's so intriguing. I'll see. Now you have the scoop. All I can tell you is I see this lamp and my son said cool, but not in my room. I would never want that thing. He thought it went choke him out while he was sleeping. My daughter are like, nope. And my wife's like, well, if it holds clothes, I'll do anything. Okay. So this and here I had created two pages of notes on this one, but the deal of it is so it came out last year. But the launch video of it, it had all these views. Let me see if I can get the exact number of views that I had. OK, it went viral. How many views did it That video came out last year and I didn't No, no, no, no, no, no. The 2025 video, the original video. This one just came out two or three days ago. The original 2025 video went viral with four million views. But it turned out what the press caught was that it was entirely CGI. OK, and it was not disclosed. The founder denied it until it was cornered by journalists and then he admitted that it was not actual footage. Now the company has yet to do a public demo of the actual product. They've done ⁓ a private demo to ⁓ Bezos and Dario Amodi. They got a private showing. William. ⁓ and William, yes, you caught that detail. OK. ⁓ So they won't see how long it takes to fold a load of laundry. So it's supposed to be shipping in summer 2026. It's meant to look very elegant. It's made out of anodized aluminum. So it has a feel like a luxury car. But here's another interesting fact. It's ⁓ meant to evoke like beauty and the beast, know, that whole. ⁓ That whole scenery and dancing. Didn't you catch that? know, you know, it's there. Don't you catch that in the mirror and in the spots and all the different things coming to life and cooking you. Right. The lamp coming to life is like beauty and the beast. So the concept's actually really cool. And hopefully this time the video is real. Now, I showed you this is a behind the scenes video now where you can see them filming it. I think their way to show that this is this one's real. Okay. mean, they could still shoot CGI that I say. no. I mean, I think what they're trying to do now is to regain the trust that they lost last year when, you know, the demo wasn't real, right? They're trying to they're trying to rebuild. They're trying to rebuild the trust, which is probably what that's about. So assuming this is a real product now and they're putting their dirty laundry, shall we say, behind them. This this could turn out to be a very interesting But if we're comparing it to something like Weave Robotics, which we both saw at the Humanoid Summit, that's a company that's also folding laundry, but they're actually already folding laundry. So they are deployed at two laundromats now in San Francisco. Yeah, there were more. Oh, is it? I thought, okay, put them in the window, right? Like not only yes. Yes. Yes. Yes. See it. Right. Yeah, they're in the end. Also, they are now as of February. Rolled out to their first customers so people can maybe that's how thinking there is Okay, yes. Yes, so So you now can have a weave robot in your home to fold your laundry. So that's a robot that's actually Doing actual work and it's a robot that we is You know is press and his creators have been able to see working So we've been able to see the speed of it We were able to see it fold a big pile of messy laundry like every which jumbled way and watch it sorted out. saw two of them working side by side. mean, yeah, it's not as fast as I am, but it was right. The speed was pretty good. Yeah, it was. And the uptime was good. I'm going to dump the load of laundry and I'll see you in four hours. Right. That's why I was wondering what the speed was on this one. Right. Like we're not given the speed of how fast it is. How long does it take to to fold the shirt and whatever else? And I don't think the expectations at this point. is that it's like, woo, it's faster than humans, right? Because it's such early days. at least some numbers would be good if you're trying to build trust and there's other solutions out there that are already out there and working, which is why I'm comparing it to Weave. So the narrative I like to think about this is I'm to take a Roomba and a car and I'm to merge those two into the narrative of robots. So. ⁓ I've been seeing robots, the home robots sold as Neo and ⁓ Sunday Robotics. He's great, can do it all. Home robots that are humanoid, the size of us almost. And the promise is they'll be like your, your butler or whatever, your assistant at home and do everything. And then that's like the middle point. I'd say we've robotics is towards the Roomba end of that scale where it's not a one Thomas robot. It's not walking around. It's not doing a bunch of things. But it's doing a few, like one or a couple things well for you. Because they did say they'll do more than fully in laundry in the future. That's not their Yeah, they might be able to tidy. Right? So, and it costs more. Costs less than, well, we've already seen subsidized pricing for these Neo and things like that. But I'm assuming that it costs more than $20,000 for that robot. We're seeing Roomba being like the very bottom of the, or robotic vacuums. and there's a big jump from a robotic vacuum in your house and Weave and neo and then a robotic car some reason we're okay with a robotic car Which is like the most expensive robot you can own and you can do a bunch of cool things but Kind of was saying like what I think like a car how we're get people to like really accept self-driving cars is people are already in cars that are doing more self-driving without people knowing it they're gonna be doing They already have ADAS in there, right? They're already detecting things and keeping you from going off the road. They're already slowing down your cruise control and so on slow in front of you. All these things, even in the lower end cars now, like my car's not a Tesla. And eventually we get to the point where it's like people who are saying, ⁓ I can never trust a car driver to be like, well, it's already doing half the driving. It's not a big jump to all of the driving. It's kind of baby steps. All robot. allow people to adjust. Yeah. Right? So like now most people have seen a vacuum. And bumps around your things and like that's the lowest form like that's probably like cruise control 1.0 you that's it. That's like all it can do. ⁓ And then the Weaver robotics is like way more forward than that. But we need more I think intermediate steps just like the car will have intermediate steps to self driving where we're going to need to have like these lamps that will start folding your laundry. Not scary not dangerous. They don't have high torque actuators. They're going to fall on you. I don't think they're going to crush you. I don't know the weight of them, but they have a heavier base it looks like. So, not as dangerous. And people eventually just get used to smaller robots doing small tasks. And that gap from that to a Neo will not be nearly as scary. It's like, already have like five robots orchestrated in my house. I just want it all in one robot now. And now you have a Neo. Yeah, 100%. I agree with you. And the first robots that I would like to see in my house would actually be more appliance. like robots, which, yeah, so I wouldn't mind a robot in my kitchen, but it just does kitchen stuff and it stays in my kitchen. ⁓ And then I'm fine with a laundry robot, but it's fine. You can stay in my laundry room and my laundry robot doesn't need to do kitchen stuff. My kitchen robot doesn't need to do laundry stuff. And then upstairs in the bedroom, ⁓ this loom robot or something like it. If it's specially designed, let's say for the bedroom, right? It would have to have really a lot of privacy considerations in mind, right? Because that's such a private space. But if all it's doing is designed to make my bed and fold the laundry and I had the money for a luxury like that, right? I'd be fine with that. And the aesthetic of it is very beautiful. It's actually, it's a much, no offense to weave it, it is a better looking robot. than Weave, is a much larger, let's say more industrial version, but a much larger form factor. So yeah, I like that idea of sort of these smaller form robots that perform specialized functions. This is how we're going to sell people on it in the future. People are just going to get used to robots being in all the different places of their house, eventually, especially as Neo gets better, right? We're not going to have tele-ops all the time. That gap will be a jump into a full on, you're gonna rent a robot, you're lease one, whatever it is. Own one if you have a lot of money. Or that'll be like, your second car will be the robot and it will be much smarter than it is now. It'll do a lot more things I don't think people are gonna give up though, having multiple robots once they get used to it. Because if you get used to robots doing things for you in the home, Neo can't be all things at all times. Can't be all at once. Right, and if you have multiple family members. So one of the interesting robots that was at GTC, was this little spider-like robot that was on the counter and it was just like mixing drinks, right? But it was small enough. Yeah, didn't either one of our fellow creators caught it on camera and then of course I had to look it up, So, but it was interesting because it could move on its own. But if you think about like... a coffee machine and a barista and a boba maker all mixed into one, but it was ambulatory and it could move and probably had a little bit of intelligence and who knows, maybe you could remember a little bit of your preferences, right? But it was just about, it was about, it was about yay high. Okay, so it's a glorified appliance at this point, but that is something people could probably get on board with and that might be their first kitchen robot, right? So it's doing small things for them. But you're not going to probably take Neo away from what it's doing in the other room maybe to make you your boba tea because it's busy doing something else. So you're going to have these multiple robots in your house doing different things. But your boba robot might only be a couple hundred dollars, whereas your Neo is a much bigger commitment. Right. I'm not going to rid of my smart thermostat just so Neo can walk over and change the buttons. Right. And you're not going to get rid of maybe your Roomba or your smart Vacuum just because neo is coming in because the room was doing its job and doing it very well It's very specialized and neo could do higher-end more humanoid functions. Yeah Multiple robots in the house. Yeah. Yeah. Yeah, we're headed that way. All right, we're headed Well, I only have three I want to keep this shorter will make the last three I think just quick notable things I saw I guess as we're talking about soft safe home robots and jump to the end of my list fa robotics which Both you and I got excited about is like a, I can bring it up. It is like a soft developer, family-friendly robot got acquired by Amazon. I was really excited about this robot. I will bring it up on my screen because it was safe. And my kids saw it and said, that's the robot I want in my house. This guy. a sort of Disney appeal to it. Right? It's dancing. It has a hand on the back. You can pick it up. It doesn't weigh very much. mean, it's not light, but it's light enough that an adult could grab it and pull it out. It's very child friendly. And you can tell they're appealing toward that market with the kid chair there and the primary colors and the dancing and the bright yellow and the soft corners on it. It feels very safe, the look of it. So Amazon bought this company, but they also bought River. I saw River at ⁓ as well at GTC and that is a home delivery like robotic dog. I'll bring that one up as well. ⁓ And I'm starting to see like Amazon is not new to robotics, but it looks like they're making a drive towards home robots. So ⁓ I think again, it'll be interesting to see like who's going to be the company who ends up getting these in the home. So here's the river. It's got wheels, but it can go up steps and stuff. So it's not just a normal one. This one makes sense for Amazon, right? Package delivery. This is probably their best. That makes sense. What's Fonda doing? ⁓ But I definitely see that. They have robots in the warehouse. They're going to put robots, self-driving vans probably eventually, robots delivering. And then you're have a Fonda robot. My family loved that one. So I definitely see that. Maybe the first robot in my house that's a general blue. just delivered the package out of its belly like an ad out of Star Wars. That was really cool. Just got to wrap it up with some twine and you got it taken out. But I think it's an interesting strategy, who they're acquiring. I think it might be someone like Amazon that will get it at home. The problem is, I think, is the privacy issue. Do you want Amazon to have FANA robot in your house? I think they already ran into that in the past with privacy issues. It's just something to keep an eye on. Nothing more notable, but then I see that they're building, they're buying the ecosystem and they don't want to just be a warehouse automation robots. They're trying to own the whole entire stack from delivery from the warehouse down to your house. you were saying Amazon's trying to get more into robotics. Do you know how many robots that Amazon has? I do not. I was going to say a million. A million, because most of the robots that they have are the ones that look like pizza boxes on wheels that move stacks for them. I've seen those. Yeah, those are the ones they have a lot of. But they have an enormous number. So they actually are a very large robotics company. But of course, their robots are specialized to the factory. What's that? Yeah, and so it's interesting to see them make acquisitions to move towards the customer, all the way to the house. And I kind of see that, okay, it's like Tesla as well. Like they have the car of Optimus, fill warehouse. They want Optimus being a butler as well in your house. They, not on my list was they have the Optimus 3. They've been kind of slowly showing ⁓ the new version, but it's very interesting. I will be paying attention to Amazon more and more because I've just seen the acquisitions they're making and they have a lot of in-house expertise on robotics. ⁓ The company that wins the home might not even be one we're thinking about right now. We're talking about so it's true You never know Last two I had a jobot. I wish I had my video queued up a jobot was at the light will party at GTC He was dancing. He was a great guy But a jobot announced that they in in three months had doubled the production shipped 5,000 human robots building from 5,000 to 10,000 in a three month period Um, and the, there's a lot of talk around how we'll get to the point where like robots will be everywhere collecting data and we won't need to do simulation. We won't need to be doing tele ops. Like robots will be collecting all the data we need to collect data for robots. Just like we have all these millions of cars out on the road collecting data for Tesla. Um, I think there's a huge gap between 10,000 robots and enough robots on the planet to collect all that data. But, um, it's interesting to see companies that are doubling production in three month period. now make 90%. China now makes 90 % of the humanoid robots that are being shipped. That's what I know is good. When you mentioned Adrobot to me earlier today, I had to look them up. And that was the number that caught my eye is it was two years for them to reach 1,000 units and then one year to reach 5,000. And then in just three months, as you said, they reached 10,000. Yeah. So. ridiculous. And then if you compare that to companies here, then of course we love. Figure, Agility, and Tesla are shipping, well, let's say a lot less than that number right now. Right. So yeah. That's not what they're shipping. Optimus says, the second half of 2026 is like a slow production start and then scale next year. So we'll see. But ⁓ yeah, it's really interesting to see these manufacturers in China, because again, these are more generalized robots. So we earlier talked about how you can strap the Jets and Thor on one of these robots and make it smart. Well, you can do that to an Agibot. You can do that to ⁓ a bunch of different robots now. And so the fact that they're shipping all these robots, they can't really do much necessarily on their own, but they're getting smarter. And we're having ways to make them cheaper and faster and better. ⁓ It just all has many points or signs up. We're going to have robots more often than you think, faster than we are expecting ⁓ in our everyday life. Yeah. I'm sticking with the take. There's going to be more than one in the house. Yeah. And then my last one was Asimov. So again, I don't know if you've been following their journey on X. They have been building in the open, with their open source. And it is the coolest robot. And I don't know how fast it's going to move on my screen here. I think it might be frozen, but they've been building this robot Asimov. See if I can get it to play. Free the robot. It's not playing well in our browser. I think I'm taxing my laptop here, but it's this open source robot. At first, I started posting pictures. It was like their feet and their legs and nothing else. I'm like, where's the rest of the robot? And slowly, you're able to buy a $15,000 open source robot similar to the G1. You can print it all out, buy all that. all the actuators, can get all the parts and build your own robot. Which is really cool because this company is literally just building it and giving it away. You can download their manual, starting the load. ⁓ And I can, again, have this on my screen moving faster. But it is a very interesting robot that I suggest people, well, here's might be more pieces for it. ⁓ Even have a GitHub repo for it. So this, after you mentioned it to me, I looked up the story of the founder of what inspired him, because he'd purchased a G2 and then something went wrong with the knee joint and he was told he'd have to wait months for a replacement part. And he's like, wait, what? And so he's just like, ⁓ I'm just going to develop my own. And that was the start of this journey for him. Go to their ex account and you can literally see them just building. The parts are, it looks beautiful. It's rose gold and black. And it's just the team. It's like they have Tony Stark working on this thing here as a picture. I love the concept behind it and just kind of the origin story and the inspiration behind it. And yeah, it's really cool. I'm glad you highlighted this particular company. Just one more. they just really dive into every aspect of the robot in like why they spent so much time just on the lakes. The design is beautiful. Right? And so I'd put this on your list of just interesting Ford thinking companies. And again, I don't know how they monetize this stuff or what. They even open source the gantry train crane for it. That's how open source they are. I introducing open crane or open source crane for humanoid robotics. They are just... Open-sourcing anything and everything on the list. Oh Now it's gonna play a song that I gotta turn off but anyways That's the last of my stories. I thought they were interesting so I Suggest that those are some companies. I keep an eye on especially maybe we read about them in the droid newsletter one day I Wrote that one down because you told me to put it on the list so you'll probably read about it sooner rather than later. Oh, man I've been following these from the first day, like probably day five or something, I started following these guys and I'm like, wait, are they building just legs? Then finally I realized they're building the whole robot, but they spent, again, months on just the legs. The feet, the feet flex, they're not just like solid flat feet, they actually have like a bend, just like a human foot. You know, our toes bend and our feet aren't just rigid, so they had a whole entire breakdown on why they decided to make the foot flexible like that, so it could better mimic how a human walks. So something look through. ⁓ So yeah, with that, that's like an hour and a half in. About GTC or the mid, well, do have any final remarks that you would put about GTC? ⁓ any final remarks. So ⁓ last year was amazing, but I thought this year was even better with the exhibitors that were there and the material that was presented and the energy that was there. it was just, ⁓ yeah, it was a really good show. I'd say my only regret was there were so many robotics companies there. Like Jensen said, there were 110 robots and all. I couldn't get to all of them. So afterwards I'd see some of the things my fellow creators would publish and I'm like, darn it, I couldn't get to that company. And it felt like I was just like nonstop, like interviewing people and out there, but just, I mean, it was just, was so big, which was great, but I just couldn't. couldn't get to it all. And you're a production assistant. I had a production assistant with me this year, right? And we were working in tandem and I was making him do interviews as well. So even with two of us, right? Which greatly increased my capacity this year, even with two of us this year. We couldn't get to all of it. So I think that's very healthy. It's a real reflection of how the industry has grown. I think next year we're going to be even more overwhelmed and I'm here for it. Yeah. I would say it was interesting for me. This was a first for me where I think I was interviewed more than interviewed people. ⁓ And I think this is going to happen. This happens to some of us creators I've talked to where it's like you become an expert by talking to enough experts that you actually now understand more of the ecosystem and your domain just because no one else is talking to everyone like you are. All of everyone wants talk to me because they know I've talked to everyone. Doing everything in their domain and they're like, you're the one who knows everything or one of the few. And I thought that was interesting how an influencer can become someone people actually put on camera and flip the script. And it was very weird for me because I'm so used to being able to control the narrative or ask questions I want. Now someone is asking me questions and I'm like, ⁓ I got to actually think about this because they put me on the spot here. ⁓ which is very interesting, but exciting. Yeah, well you think about the amount of research we always have to do before we talk to somebody, it's like you have to become an expert on every single company, and then you learn more talking to the company. so, yeah, so you do end up with this wide, you kind of become like ⁓ the Wikipedia. Yeah, I'm really excited to talk to more people doing hardware, just because again, I'm not a hardware guy, I've worked in software my entire life, so. learning about how hardware and software have to interact and all that. It's been really, really an unknown, uncomfortable territory for me, so ⁓ I don't shy away from it. I just want to talk to more people working on hardware and control theory. Now you get these robots to actually do actions with real hardware and real physics, not just telling it to move your arm there. Well, you have to make sure the actuator doesn't fall apart midway through and things like that. yeah, lots to come. On this podcast, I guarantee I already have a couple of guests lined up who are more hardware specific, so it'll be exciting to talk to them as well. Well, I look forward to it. I'm an avid listener of your podcast, so I look forward. I told you I've listened to some of them twice, so I look forward to hearing them. Yep. Not only in a few episodes in, but there is not that many robotics podcasts, to be honest, too. Yep. excited about that. Well, Diana, thank you for being on this episode. I'm glad for anyone who's made it this far. ⁓ So much going on. I suggest that you follow Diana. She has, well, I'll put the links, but you have the Joy newsletter. We also have Deep Learning with the Wolf, right? So people who are not necessarily robotics, but deep learning. You have that whole newsletter. that correct? Yes, that's how I started out. I was super interested in AI. And then when AI just became physical AI, that's what drew me over. into robotics, is how I ended up with two newsletters. yeah. You know, what I love is that they are a little different than each other, right? AI is like what people think about is usually not physical AI. So you can kind of have two different narratives and two different audiences that definitely overlap. So some people might want to hear about robots all day long like I do. Yeah, I tried to think if I could squish him into one newsletter and then I was like, there's slightly different audiences. So I put him into two. I also want this deep learning with the wolf and then deep learning came AI. Yeah, a of people would be talking about AI and what they mean is like neural networks or deep learning or all these different ways we can talk about it. they're all the same thing. Well, deep learning started out because because it was interesting to car. You know, I'd gotten my Tesla and I found out it doesn't really drive itself and I wanted to understand the neural network behind it. So that's what started it is I got interested in neural networks. But then after that chat GPT came out and then that sort of transformed everyone just wants to talk about AI, which it's not quite just AI, but anyway. ⁓ Yeah. So that that's kind of transformed into a lot of that. I mean, I learned a lot about AI when nurse came out, neural neural. ⁓ what do you call them, O-radience fields, which is how to neural network for encoding a scene into a neural network. And then Gaussian splining came out and everyone's like, oh, it's a different AI. And there's really not much AI in this, but sure. But you really have to understand AI to understand that there is no AI sometimes in things. Right, right. You have to be in the bubble to want to talk these little nuances. I think they steal concepts from classical neural networks. They use back propagation. use some of these techniques, but it's not. An AI researcher would say, that's not AI. That's like just using some shared techniques that we use in training AI. And I'm like, exactly, right? So yeah, exciting time. You talk about talking about this stuff with your spouse. I talk about it so much that my husband will now even sometimes mention back prop. And I was like, wow, I have done well with him. We have never talked about that in my house. It'll come. All right. Well, it's great having you on here. ⁓ I'm gonna jump to the outro. I'll see you guys in the next episode. All right, take care. Thank you. That'll do it for today's episode. Huge thanks to Diana and both tourists. Check out the Joy's newsletter if you want to stay up to speed on the robotics industry. And as always, thank you to Light Wheel for making this show possible. If you enjoyed this one, subscribe wherever you're listening, and I'll see you guys in the next episode of The Thinking Machine.