Greetings from my old South Bay stomping grounds, where the mercury now hits 95 degrees in mid-March, apparently. I’ve been entrenched deep within the confines of the San Jose Convention Center and a two-star motel (when rooms are still available and $150 a night a few weeks ahead of a big event, there’s probably a reason for it), covering NVIDIA’s big developer event.
As such, you find a lot of GTC stuffed into these pages. It’s a huge show and a veritable who’s who of physical AI. Downtown is awash with green lanyards (kudos to NVIDIA for getting 25,000 people to wear green on St. Patrick's Day) and Cybertrucks bound in promotional wrapping for various Kubernetes companies. I spoke with a lot of humans, and shook many hands, of the human, robot, and prosthetic variety.
It’s dawning on me now that I’m going to have to save some interviews for next week’s newsletter, so do me a favor and no one make any major physical AI or robotics announcements in the next seven days. Meantime, we’ve got a lot for you below (apologies in advance to my editor, Maureen).
I caught up with Dexterity CEO, Samir Menon, ahead of the event, to discuss the startup’s physical AI headstart. At GTC, I chilled with Olaf the robot snowman and his creator, Moritz Baecher, the director of Disney Research Robotics Lab in Zurich. Amit Goel, NVIDIA’s head of robotics and edge computing ecosystem, joined me to discuss sim-to-real, real-to-sim, and the humanoid form factor as a developer platform. Frog’s Head of Industrial Design talked about the firm’s outside-the-box approach to home robots.
This week on the pod, we’ve got iRobot CEO, Gary Cohen. Grab a snack and settle in, because we’ve got a lot to talk about.
Halfway through our call, Samir Menon asks me to do a “little experiment.” I hesitate slightly. These sorts of interviews rarely call for interaction, besides the bit where we talk back and forth for 30 or so minutes. Dexterity’s CEO instructs me to close my eyes, put my hands down by my side, and then reach for an object on my desk. I lift my AirPods case.
“The first thing that happens is when you close your eyes, you have this model of your body,” Menon explains. “We call it a ‘body model.’ You know what the capabilities of your body are. You've got a model of your AirPods, and then you have a model of the physics of the world and how to reason about the spatial trajectories. All of this for us gets combined inside what we would call a ‘world model.’ The world model doesn't actually do the task, but it helps you reason about the world, and it's foundational.”
Earlier this month, Dexterity introduced Foresight — or, rather, it “publicly nam[ed] something [it has] been building for years.” It’s what the Bay Area-based firm has taken to calling the world model it’s been constructing in labs and warehouse floors for the last seven or so years. While many of its newer and more theoretical brethren have been raising billions to take those first steps, Dexterity has been helping robots get very good at packing and unpacking boxes from pallets.
That Foresight’s introduction didn’t light the startup press on fire probably has much to do with the fact that the company’s announcement wasn’t focused on humanoids, and never actually got around to uttering the words “general purpose.” For better or worse, nowhere in the demo does a robot execute a flawless backflip. If, however, you’re in the market for a browser-based game that allows you to stack boxes in the back of a simulated truck cab, friend, you’ve come to the right place.
The past year has seen a massive spike in funding for physical AI companies, many of whom are working to build massive, generalized world models for robot deployment. Meno —- a Stanford PhD, who founded Dexterity in 2017 — believes the attention being paid to startups like Physical Intelligence, Field AI, and the like, will be a net benefit for the industry at large.
One company has driven humanoid robotics’ growth more than any other. It’s not Tesla or Boston Dynamics, Figure or Agility. In fact, it’s a company that doesn’t make humanoids at all, even if its leather jacket-clad CEO loves nothing more than a great photo-op with their kind.
I can’t conclusively say that I saw every humanoid present at GTC this week, but it sure feels like I packed a lot into two days. Agibot, Humanoid, Noble Machines, [tk], and a seemingly endless stream from Unitree were present on the show floor. There was also that Gaddian anthropomorphic snowman, if we want to expand our definition to include human-adjacent, ice-based lifeforms.
And then there were all of the humanoid companies I chatted with at the event that didn’t have a robot with them (turns out bringing one to a tradeshow is a bit more involved than buying a couple of plane tickets). I won’t attempt to name all of them, because I’ve slept poorly this week and will definitely leave a bunch out.
Not too long ago, if you had asked me why NVIDIA is being so bullish about humanoids, I would have given you some straightforward answer about market size and capitalism. Those are all still important factors for a trillion-dollar public company, of course, but my recent chat with the chipmaker’s head of robotics and edge computing ecosystem, Amit Goel, surfaced another compelling point.
NVIDIA is a facilitator. The company excels at components and platforms. For this reason, it’s become a fundamental connective tissue across all sorts of tech verticals. Robotics/physical AI has become a massive one of late, thanks to all the work it’s been doing on the Jetson front over the years. There’s a reason so many humanoid companies bring their robots to GTC, appear on stage for a Jensen photo-op, and/or take part in some animated campfire.
Now what if we started thinking about humanoids as a developer platform? Some form of general intelligence is the stated goal of so many of these physical AI startups. What better way to experiment with generalized skillsets than a general-purpose robot? I won’t say humanoids are the end-all, be-all of this particular category (us humans are fine, I guess), but the proliferation of the form factor has made it a great place to start.
As real world, free-roaming robot snowman testing ground, one could do far worse than the San Jose Convention Center on GTC day two. It’s no Magic Kingdom, sure, but you’ve got plenty of awe-struck, extremely distracted throngs of humanity to bob and move around. Fresh off his second stage appearance in as many days, Olaf – and his Disney Research Designer, Moritz Baecher – decided it was as good a time as any to walk to show floor.
Both wore badges. The robot’s read,
Olaf
Walt Disney Imagineering
Event Staff
It wasn’t much of a cover, however. A crowd soon gathered around the movie star, eager for a better look a day after he helped close out the day one keynote. Even in at a show populated by a dozen or show functioning humanoid robots, a waist-high, anthropomorphic snowman is going to draw a crowd. But hey, it’s a good dry run a few weeks ahead of the system’s official Disneyland Paris debut.
Olaf represents the latest addition to Walt Disney Imagineering’s growing army of advanced robotic characters. Animatronics have come a long way since The Hall of Presidents’ early 70s debut. The goal here is to blur the line between audience and attraction by building characters that can walk amongst show goers.
This year’s GTC was a return appearance for Baecher, who was joined on stage last year by a BDX Droid. Those Star Wars inspired systems have since taken up residence at Disney resorts and aboard the company’s cruises.
At the risk of getting all 80s standup comedian on you, people can’t help but tidy up before having a professional over to clean their house. Maybe it’s vanity. Perhaps it’s a social contract. It could just be that people are okay with a stranger knowing their house is messy, but just not that messy.
Turns out the same is true for robotic vacuums, though this one doesn’t take a degree in human behavior to crack. It’s simply that the little puckish appliances have historically had trouble with big messes. Socks, paper, and computer cables can all wreak havoc on a robot vacuum. And then there’s the pet poop problem that infamously vexed the industry for so long. It’s one thing to get a cord stuck in your rollers. It’s another to paint the carpet with dog droppings (sorry, but it’s a thing).
“One thing that's interesting about the robot vacuum space is like for so long that space has been one where you actually have to clean your house before you can have it work,” Frog’s Head of Industrial Design, Inna Lobel, told me this week at GTC. “It was really interesting to see Roborock start to experiment with arms and things like that. I think this is a place where people are going, but they're going at it in a very different way.”
Roborock did what it does best – it built added stuff to a robot vacuum. In this instance, it’s a mobile manipulator that reaches down to pick up socks and other human detritus. Frog, best known for its design work companies like Apple, Sony, and Disney, took the unstructured path.
Lobel uses the word “unobtrusive” when discussing Nome. Frog’s site describes it as “an internal R&D platform to consider how physical AI should show up in human spaces, shaped by context, culture and everyday behavior.” I’m pretty sure I called it something like a “robot coatrack.” We contain multitudes.
Automated is getting ready to take this show on the road for the first time ever. To mark our Beantown blowout, I’ll be talking with MassRobotics cofounders, Tom Ryden and Joyce Sidopoulos. We’ll discuss the Boston ecosystem, the organization’s history, and talk about some of the more fascinating startups that have gone through their system. The event will stream live at 11 a.m. ET/8 a.m. PT Wednesday, March 25 on both LinkedIn and YouTube. Earlier this week, MassRobotics announced the members of its second Physical AI Fellowship cohort, featuring some familiar names, and several new ones: Burro, Config, Deltia, Haply Robotics,Luminous Robotics, Roboto AI, Telexistence, Terra Robotics, and WIRobotics. The eight-week virtual program launches in spring.
If you came to GTC for understated expressions of technology, good news, the Rosicrucian Egyptian Museum is walkable from here, and it’s a pretty good way to kill an afternoon. Those looking for some neon green bombast, on the other hand, have come to the right place, as NVIDIA proudly announced, “the big bang of physical AI has started.” Surely this means Gil Pratt’s “Cambrian explosion of robotics” can’t be too far behind (whatever 13 or so billion years amounts to in tech terms). The chip giant just unveiled updates to some of its key properties in the space, including new versions of Cosmos world models, NVIDIA Isaac simulation frameworks, and NVIDIA Isaac GR00T N models.
The news also sees a veritable who’s who of robotics firms partnering with the company. What’s notable here isn’t the number of names, so much as the breadth of categories, ranging from industrial stalwarts to humanoid and physical AI startups. The list includes (in alphabetical order, mind) ABB Robotics, AGIBOT, Agility, FANUC, Franka Robotics, Figure, Hexagon Robotics, KUKA, Skild AI, Universal Robots, World Labs, and YASKAWA. While much of the coverage around the company's physical AI plays has focused on “general purpose” form factors, this list includes tentpoles of the industrial arm world, including ABB, Fanuc, KUKA, UR, and Yaskawa. Those partnerships are focused on Omniverse libraries, Isaac sim, and Jetson modules for controllers.
Accel and Andreessen Horowitz want you to know they’re serious about physical AI. Both investment firms participated in a massive $500 million round for the relatively unknown Mind Robotics. It certainly doesn’t hurt that the Silicon Valley firm is founded and led by RJ Scaringe, CEO of Rivian, makers of electric trucks and SUVs with headlights shaped like Tintin’s eyes (good luck not seeing that now). In a release tied to the news, a16z general partner, Sarah Wang, notes, “RJ is one of the very few founders who have built and scaled a vertically integrated hardware company. “At Rivian, he architected the full stack — vehicle architecture, electronics, battery systems, embedded software, manufacturing processes, and supply chains — integrating each layer into a cohesive system. As other EV makers are likely figuring out, robotics and physical AI are their own beast, but building Rivian was no small feat and no doubt gives the firm an important leg up on other startups in the space.
For its part, Mind is launching a full-stack operation — a major motivator in raising such a massive amount so shortly after announcing a $115 million seed round. That includes the foundation models, the robots themselves, and the eventual deployment in industrial settings. “As AI enters the physical world, we believe the largest, at-scale application for advanced robotics will be across the industrial sector,” Scaringe says in a release. “Advanced robotics are going to be critical for global competitiveness, as well as addressing the substantial industrial labor shortages that exist today. We’re building robots that will perform real tasks, in real plants, at real scale. I am grateful to have partners that believe in what we are building at Mind Robotics — looking forward to having Sameer join our Board.”
Self-driving cars and humanoids tend to hog the headlines at these events, but on Monday morning at GTC, NVIDIA announced new offerings designed to accelerate industrial manufacturing. The news features big-name partners, including some top industrial robotics firms, chip fabs, and automakers: FANUC, HD Hyundai, Honda, JLR, KION, Mercedes-Benz, Mediatek, PepsiCo, Samsung, SK Hynix, and TSMC. All of the above will be deploying NVIDIA’s CUDA-X and GPU-accelerated software to improve manufacturing, engineering, and design. Each will also be deploying NVIDIA’s agentic AI for customer interfacing.
“The dawn of a new industrial revolution has arrived, where physical AI and autonomous AI agents are fundamentally reinventing how the world designs, engineers, and manufactures,” per NVIDIA CEO, Jensen Huang. “Uniting our global ecosystem of software giants, cloud providers, and OEMs, NVIDIA is delivering a full-stack accelerated computing platform that empowers every industry to turn this vision into reality at a scale and speed never before possible.”
Almost immediately after spinning out from Intel over the summer, RealSense announced that it was strengthening ties with its former parent’s rival, NVIDIA. Fueled by $50 million in fresh funding, the 3D camera maker also announced that it was making humanoids a primary focus, along with existing categories, autonomous driving and AMRs. Monday at NVIDIA’s GTC conference in San Jose, the company showcased LimX Dynamics, a navigation system designed to help legged humanoids map, navigate, and localize autonomously. The system was trained in NVIDIA Isaac simulation, and utilizes a combination of RealSense cameras and CuVSLAM, NVIDIA’s SLAM/visual odometry system.
“Humanoids operate in three dimensions, alongside people, in environments that are constantly changing,” RealSense CEO Nadav Orbach notes in a statement. “If robots are going to work safely beside humans, perception carries responsibility beyond raw sensors. It must function as the robot’s visual cortex, enabling accurate localization, collision avoidance, terrain understanding, and stable, predictable motion in unstructured environments.”
All those Pokémon you caught a decade back helped create the geospatial data that formed the foundation of Niantic spinoff, Niantic Spatial. The startup got real-world data points, and you got to evolve pocket monsters — let’s call it a draw, I guess? As you’d imagine, such mapping is a highly valuable asset if you run a, let’s say, robotic food delivery startup. It’s the kind of information that lays the foundation for the deployments that’ll really get your flywheel spinning. As such, Coco Robotics has signed on as a partner. The Southern California company will be utilizing Niantic Spatial’s AI and its Visual Positioning System (VPS) to help its delivery cart get food to your door. Here’s Coco CEO and recent Automated interviewee, Zach Rash, “We’re excited to bring the Niantic Spatial and Coco Robotics engineering teams together in this unique design partnership. It gives us reliable access to localization services that further improve robot navigation. Looking ahead, we’ll jointly explore new ways to enable Coco’s robots to operate with increasing safety and autonomy in any city.”
“May the road rise up to meet you. May the wind be always at your back.” Listen, it was St. Patrick’s Day on Tuesday and I’m not aware of any Asimovian rule that says traditional Irish blessings can’t apply to robots. And honestly, it’s hard not to root WANDER-bot on as it traverses hostile terrains, running on pure gumption and wind power. The 3D system was designed by researchers at Cranfield University in Bedfordshire, U.K. to navigate hostile environments on Earth and — they hope — other planets.
I had a nice chat with Physicl’s CEO, Alex de Vigan, on Tuesday, roughly 24 hours after the French podcast came out of stealth at GTC. This is one of those interviews you’re going to have to wait until next week’s newsletter to read in full — it’s been a busy week, and contrary to popular belief, I am not, in fact, a machine. In the meantime, the Physical AI company’s story is an interesting one. It’s kind of a spinout/sub-company of Nfinite, which is best known for working with Getty Images to turn the wire service’s photos into 3D images. The new startup’s goal is, in its own words, “scaling world-ready data for embodied intelligence.” It’s still very early stages (again, it came out of stealth on Monday), and de Vigan tells me the company hasn’t partnered with any robotics companies yet, though he’s hoping a high-profile launch at GTC will change that.
I also had a nice chat with Leo Ma, CEO of RoboForce, who also dropped some big news this week at GTC. That, too, will have to wait until next week to get the full Automated feature treatment. Meantime, the Sunnyvale startup just announced an oversubscribed $52 million round, pushing its total funding to $67 million. Confession time: I was like 95% sure this was a battlefield robot when it was initially pitched to me ahead of the fundraising news. 1. The company is called RoboForce, which sounds like an 80s anime where a team of people with different colored helmets unites to form a giant robot. 2. I mean, just look at the thing. Click through the link to see more images of the tank-like tread locomotion system.
Ma, however, assures me that the startup is a “force for humanity,” by which he means it’s built to perform the three classic Ds (dull, dirty, and/or dangerous — that describe the work, mind, not the robot). A big part of the reason the thing is so robust is that it’s meant to do much of its work outside, be it solar panel installation, mining, or shipping, along with the more standard manufacturing and logistics gigs. Per our chat, the CEO says the company is really focused on scaling, deployment, and that sweet, sweet ROI at the moment.
iRobot CEO Gary Cohen is the guest on this week’s pod. It’s a great conversation reflecting on the state of the industry and the organization when he took over from cofounder Colin Angle, midway through 2024. We also discuss the company’s recent acquisition and journey through the Chapter 11 process. During the interview, Cohen alludes to an upcoming product for the Japanese market — mind you, the conversation was recorded just over a month ago. It seems clear now that he was referring to this lil guy. The Mini is the “smallest Roomba yet,” and something I no doubt would have had my eye on during my decade or two of urban apartment dwelling.
One of the more immediate bits of Reddit criticism we saw around last week’s pod with Matic cofounder, Mehul Nariyawala, centered around that vacuum’s height and its inability to get under certain pieces of furniture. I suspect we won’t be hearing similar feedback about the Roomba Mini. Battery life and bag capacity could be issues, however, so I suspect it will be making regular appearances at its similarly small AutoEmpty Dock. After debuting in Japan, it’s since arrived in the U.K. and the greater E.U., priced at £379 and €399 ($506/$460), respectively.
Buried amid the deluge of AI and robotics news at NVIDIA’s GTC kickoff Monday is news of IGX Thor’s general availability. This is the industry-focused version of the company’s physical AI processor, focused on a considerably broad range of industries, including manufacturing, logistics, construction, sciences, healthcare, and space. It wouldn’t be an NVIDIA developer event without a long list of partner companies, and this morning’s news didn’t disappoint. Agility and Hexagon Robotics arrive via the humanoid side — both companies have recently announced pilots with prominent automakers (Toyota and BMW, respectively). Supply chain firm, KION Group, is using the platform to improve safety perception, while a variety of companies are adopting the tech for autonomous surgery. Those are Johnson & Johnson, Medtronic, and Horizon Surgical Systems. As for space, Planet Labs is utilizing the tech to convert satellite data into “actionable intelligence in orbit,” while CERN has picked it up for AI model creation.
Get a load of this goober. Better yet, watch a video of the thing in action. When stationary, they look like the sort of modern minimalist sculpture you’ll find on most liberal arts university campuses. When in motion, on the other hand, it’s a whole different ballgame. The “legged metamachines” are comprised of modular building blocks that snap together into a wide variety of form factors. As its creators at Northwestern University note, “Each module by itself is a complete robot with its own motor, battery, and computer. Alone, a module can roll, turn, and jump. But the real agility and indestructibility emerges when the modules combine.” Agile? Sure. Graceful? Not so much. Still, the project is impressive nonetheless.
“These are the first robots to set foot outdoors after evolving inside of a computer,” says the paper’s lead, Sam Kriegman. “They are rapidly assembled and then quite literally hit the ground running. They can move freely in the wild and easily recover from major injuries that would be fatal to every other wild robot. If flipped upside down, they instinctively bring themselves upright and continue their journey. They can survive being chopped in half or cut up into many pieces. When separated, every module within the metamachine can become an individual agent.”
Plenty of folks I met within San Jose this week were planning to take a quick jaunt over to Stuttgart for LogiMAT. I’m sitting this one out, unfortunately — you know how I love a good Messe. Austin’s BrightPick will be there, however. The company Gridpicker, a new storage retrieval system with a built-in mobile manipulator. CEO Jan Zizka says the system is an evolution of the company’s AutoPick platform. “[W]e created Gridpicker by essentially taking our Autopicker robots and placing them on a high-density grid,” he notes. “It combines shuttle-level throughput with the simplicity of AMRs, delivering up to 40% lower cost and requiring 5x less labor than shuttle systems.”
One of the biggest hurdles standing between physical AI and its “ChatGPT moment” is a lack of quality data. A big part of the reason LLMs have been such a massive — and often surprising — success is the fact that humans have essentially been creating training data for 100,000 years or so. The same can’t be said for the input required to train robots. NVIDIA is among the companies working to address the gap, and this morning at GTC the company announced Physical AI Data Factory Blueprint, an open reference architecture designed to improve how both real-world and simulated data is gathered, shaped, and assessed. The company has already recruited some big names from across autonomous driving and robotics, including FieldAI, Hexagon Robotics, Linker Vision, Milestone Systems, Skild AI, Uber, and Teradyne Robotics.
The platform is host to a number of processes designed to do right by the real and synthetic robot data. There’s Cosmos Curator, which processes and annotates datasets, Cosmos Transfer, which is designed to address edge cases and long tail scenarios, and Cosmos Evaluator, which, you know, evaluates data. “Physical AI is the next frontier of the AI revolution, where success depends on the ability to generate massive amounts of data,” says Omniverse VP, Rev Lebaredian. “Together with cloud leaders, we’re providing a new kind of agentic engine that transforms compute into the high-quality data required to bring the next generation of autonomous systems and robots to life. In this new era, compute is data.”
Gecko Robotics just landed a $71 million contract with the U.S. Navy to deploy its wall-climbing robots, drones, and sensors for monitoring and collecting ship and submarine health data. The systems will survey the state of welds, decks, hulls, and components, examining the info with AI to create models designed to showcase existing issues and help predict future problems. The partnership is part of a push to hit the Chief of Naval Operations’ stated goal of 80% fleet readiness.
The Association for Advancing Automation (A3) is North America’s largest automation trade association representing more than 1,400 organizations involved in robotics, artificial intelligence, machine vision & imaging, motion control & motors, and related automation technologies.
.jpg?upscale=true&width=1120&upscale=true&name=b3e013a5-910c-4edd-9c0a-5bea29a9ee88%20(1).jpg)
.png?upscale=true&width=1120&upscale=true&name=industrial-software-giants%20(1).png)
