Just wrapped up an unforgettable two-week robotics marathon across China, and I still can’t quite believe it’s over. Every single day of the trip and even now looking back, I was overwhelmed with gratitude for this incredible privilege. I love this robotics/AI world, and this journey reminded me exactly why. Gaining access to 18 of China’s top humanoid robotics companies in such a short time felt almost surreal. It was a rare and special experience. None of it would have happened without the amazing people who made it possible. Huge thanks to @TheHumanoidHub, @XRoboHub, our interpreter Winnie and everyone else who opened doors, shared insights, and supported us along the way. You made this trip truly beautiful. I’ve long been fascinated by China’s robotics ecosystem and always wanted to visit someday. Everyone I spoke with in the US and elsewhere kept saying “If you’re serious about this field you have to see China.” Spending two intense weeks on the ground, visiting different cities and meeting brilliant founders and engineers was deeply inspiring. I loved every moment of it. It all started earlier this year in SF. @TheHumanoidHub and I kept crossing paths, diving deep into robotics conversations and I knew he was planning this journey. When he turned to me and asked “Want to come?” I said yes immediately and booked my ticket the same day. Future me is still thanking past me for that decision. The trip was wonderfully intense and ridiculously fun. I’ll never forget sharing takeout pizza in taxis while racing to the next company, passionately talking about the future of robotics the whole way. Those moments were pure joy. There’s so much I saw, felt, and learned. Before the memories fade I’ll be sharing more of it here. And yes — we have more content coming so stay tuned😎🤖

Hyundai bought control of Boston Dynamics at a $1.1B valuation in 2021. Today Unitree opened near $66B - 35x it's last VC round of $1.9B and ~7x its IPO price. Unitree has real revenues and major brand presence, but the company is not very well understood by most investors. In 2025, it generated $131m in Humanoid sales. 70-80% of its Humanoid sales were for research use cases, 20-30% for education and entertainment, and very little for genuine robotic use cases. Unitree's main product was a robot body that lacked a brain. An affordable, programmable, at least semi-reliable brainless humanoid body however, is exactly what the research market demanded. Unitree G1s are ubiquitous across robotics research groups around the world. Just as AI research is dependent on physical computing hardware, robot hardware is indispensable for research in robot learning, control systems, simulation, etc. While there are few Unitree humanoids currently deployed for real robotic work, the rise of the company has greatly accelerated global research progress. It optimized for an axis (cheap dynamic locomotion) that allowed it to capture the research market but is different from the requirements of of the deployment market (intelligence, durability, payload, safety certification). However, a lot of the engineering capability they've built as a company can and is starting to be used to develop more deployment optimized hardware models. This is a fundamentally different approach from most American humanoid companies which are building towards operational products for consumers and businesses in a straight shot. Companies like Figure and Apptronik invest more resources in R&D and don't yet offer it to retail because they want to go direct to the larger deployment markets. They are building towards a highly functional polished product that can eventually become a development platform like Apple (as opposed to starting as a development platform). For AI, Anthropic took the mass market product capital intensive approach and it took a lot of dollars and time before lifting off on revenue. Cohere and AI21 Labs have existed for a similar amount of time, and took a more capital-light path. AI21 had to pivot, while Cohere has continued to grow, although significantly more slowly than Anthropic. Neither approach is right or wrong and history is filled with examples of successful parallels for both. You cannot compare companies taking different approaches solely on a revenue multiple basis. The company's commercial approach is a byproduct of the Chinese private capital markets. A market where there are not as many venture dollars as the US that are willing to fund hundreds of millions to billions for R&D before any revenue is generated. Revenue growth is required to fund the next rung of capital even for potentially massive TAMs. Actuator scaling parlayed into quadrupeds, quadrupeds into the dominant robot hardware research platform. This IPO funds their transition to the most ambitious phase yet - a company building vertically integrated intelligent robots across a wide variety of form factors. The current market valuation is suggesting that they will accomplish this transition, although it is not final yet. The outcome for Unitree differs dramatically based on if they can successfully move up market. Companies like DJI and Toyota have previously done so, while a failure to do so could have the company looking like Raspberry Pi. A company that cemented themselves within experimentalists and niche industrial markets. However, it may not be necessary for Unitree to build SOTA research capabilities in order to scale robot sales into real deployments. If physical intelligence commoditizes, which we believe it does, then they could have plenty of externally produced models for their customers to choose from. The companies that can produce high quality hardware at scale stand to be large benefactors from the development of physical AGI. The focus on hardware has enabled Unitree to raise a huge war chest and have access to thousands of robots that they can use for robot learning data collection and research. While various data types can be used in pretraining for robot foundation models, robot data is required for the models to become performant. To collect a large set of robot data, you will need a lot of robots. They may have actually created a stronger path for themselves to produce performant physical AI models than companies that have focused purely on robot model development years ago. Wang Xingxing is famous for his technical chops, but he has also been an excellent business strategist.

One brain. For any robot. 🤖 We’re launching Gemini Robotics 2: our next-generation physical AI bringing full body intelligence to humanoids, advanced dexterity, multi-robot teamwork and more.

Xynova tour in Hangzhou. The startup is building dexterous hands for humanoids. Their tech lead chatted with us about the 23-DoF hand design, a hybrid-drive approach, tendon durability, and more. Shoutout to @dolylupec for co-hosting the interview. @Xynovaofficial_ @XRoboHub

Introducing ACT-2 Preview The first robotics model to unify broad generalization with high reliability. A single fine-tuning example can teach Memo a new behavior that generalizes. Zero shot, real unseen homes, 99% success rate.

Remember Figure's package-sorting demo? We saw RobotEra M7 humanoid perform a similar use case autonomously for 6 minutes without a mistake. We toured RobotEra's headquarters in Beijing and chatted with Lucas Wang (@lucas_wang_X), VP of Product, on their direct-drive hand design and where dexterous hands go next. By the way, this package-sorting use case is already deployed at real sorting stations in multiple locations in China. Thanks @dolylupec for tag-teaming the interview. Thanks @roboterax for the tour.

In January, I started "building something new" with an incredible team. Today I finally get to share some first details about what we've been building. We've called it Walden Robotics (waldenrobotics.com). I thought long and hard about my own reasons for starting this company. It's not only about the robots. It's also about people. I've tried to capture those thoughts in my first Walden blog post: waldenrobotics.com/news/why-… It's been an incredible ride so far. Within just a few months of forming the company, we were already operating a general-purpose robot with an end-to-end policy in production in one of the most important factories in North America. It's amazing at how much I've already learned from that experience. There is a lot of work to do, but the mission has never been so clear. Please help me welcome Walden Robotics into the world. And stay tuned for more updates! piped.video/watch?v=fewvZrck…

my favorite poster at ICML

Introducing NEO’s 25 Degrees of Freedom, tendon-driven hands — nearing or surpassing human-level dexterity, strength, speed, and reliability. For seventy years, robotics worked around the hand problem. The humanoid bet is the reverse: it lives or dies at the fingertips.

The best event at ICML by far was morning pilates & brunch with @mmmbchang 🔥

A year ago, when I left Tesla Optimus to co-found Mondo, I made a promise to my son: I would build him a little robot buddy — something like the robot friends from our bedtime stories, but real. At the time, it was not really a product statement. It was a father’s promise to a little boy I love more than anything — a boy who loved stories about a child and his robot companion. Over the past year, that promise slowly became real — Beni the robot. Beni follows you, films you in 4K, moves across different terrain, jumps over obstacles, captures moments from new angles, and helps turn your clips into highlights. We designed it for families, pets, creators, athletes, and anyone who wants a robot that feels less like a gadget and more like a little sidekick. I’m deeply grateful to the Mondo team for turning countless prototypes, failures, late-night tests, and difficult tradeoffs into something people can finally meet. Although we have only just launched on Kickstarter, this world-class team is already deep in the “production hell”, and we expect to ship the product in the next few months. I’m also especially thankful to my family. Building hardware takes time, patience, and sacrifice from the people closest to you. Beni would not exist without their love and support. My son often visits our office and plays with Beni. Watching them together has reminded me why I started this journey in the first place. I do not want him to grow up thinking of robots only as distant machines in labs, factories, or videos. I want him to grow up in a world where robots can feel close, friendly, safe, and present — something he can play with, trust, and remember as part of his childhood. For the robotics community: Beni may look cute, but it is a serious robotics product. We used reinforcement learning to train Beni’s motion policies, with NVIDIA Isaac Lab and Google MuJoCo as part of our training and validation pipeline. Under the playful exterior are real robotics problems in locomotion, perception, control, and sim-to-real transfer. The lessons we are learning from Beni also directly inform our future humanoid product. Different form factor, same belief: advanced robotics should not feel distant or abstract. It should become close, present, and part of everyday life. For my son, Beni began as a promise. For Mondo Robotics, it is just the first step. Meet Beni: mondorobotics.com/

We visited One Robotics' lab in Shenzhen. The company has a lineup of home robotics products but is now heavily focused on its wheeled humanoid robot, Onero. We chatted with their Chief AI Architect, Hu Yusong, about the AI foundation model work his team is doing. @dolylupec @XRoboHub Thank you @OneRoboticsTech for the tour!

We went to Wuji's Shenzhen office to get a grip on their dexterous hand design. Each finger joint is driven directly by a tiny motor paired with a worm gear that redirects the axis of rotation. We spoke with a senior team member there to get more insights. @dolylupec @XRoboHub Thanks @wuji_global for the tour!

Really glad I finally got to visit @XSquareRobot. We got an early look at XRZero-G0, their open-source full-body data collection and training system that works without requiring a physical robot body. At the time it hadn’t been officially released yet, so we couldn’t share anything haha. Now they’ve officially launched the QUANXTA Zero series — a robot-free embodied data production platform. This is not just a hardware setup. It’s a full-stack system that connects data collection, high-fidelity synchronization, automatic cleaning, intelligent annotation, embodied model training, robot inference, and evaluation loops — all in one pipeline. The goal is very clear: close the “last mile” gap between data and models, and build scalable infrastructure for embodied AI. QUANXTA Zero comes in three setups for different capture scenarios: → G1 (UMI-VIO): dual grippers with a head-mounted rig, balancing quality, usability, and endurance → G0 (UMI-VR): full-body mobile data capture → E0 (Ego): lightweight head-mounted setup for first-person data collection In short, it’s building the data foundation for embodied intelligence. Data is what drives intelligence emergence in large models — both scale and quality matter. If the data flywheel is already spinning, real home robots might be closer than we think. @TheHumanoidHub @dolylupec

If you’re coming to Seoul for @icmlconf 2026, welcome to my beautiful hometown!🇰🇷 This year’s ICML had a record ~24,000 paper submissions. What stood out most this year was a shift away from simply making models smarter, toward building AI that can actually perform tasks reliably and safely. A few things stood out to me in the papers: - Agentic AI made up a big portion of the submissions. At the same time there was a clear focus on safety, reliability, and real-world applicability rather than just performance. - On the Embodied AI side, Vision-Language-Action (VLA) models showed up a lot. These models don’t just understand vision and language - they generate actions, which is getting a lot of attention in robotics. Research on memory and long-horizon behavior in robotic systems also stood out. - AI for Science was also quite visible, especially in biology-related areas such as genomic foundation models and protein modeling. If you’re working in any of these areas (especially Embodied AI/Robotics and AI for Science) I’d love to hear what you’re focused on. Happy to grab a tea while you’re in Seoul, just reply or send me a DM :)

Today, we’re launching our home robot Isaac 1. Isaac 1 deliveries will begin this fall. Order yours below.

We visited Paxini in Shenzhen, a company focused on the sense of touch for robots. We got hands-on with their dual-modal Dexter Hands (cameras + tactile sensors) and their Tora Double One humanoid. Humanoid robots will need this touch layer to get truly dexterous. @dolylupec @XRoboHub

Tour video of our visit to LimX Dynamics (@LimX_Dynamics) in Shenzhen. We saw some impressive demos of Oli and Tron and had a conversation with Paul Li, Head of R&D. We discuss LimX's tech stack and the fundamental engineering challenges with humanoids. Paul has refreshingly honest takes on the state of humanoids. Thanks @dolylupec for filming and jumping in with questions.

South Korean companies are becoming leading suppliers of humanoid robot parts and technology, and they are forecast to command a 30% share of global production by 2035. Read more from Goldman Sachs Research: click.gs.com/6oym

. @AGIBOTofficial reached 15,000 robots in production It took 2 years for the first 1,000 units. Now their production rate has increased nearly 40×; they’re making 5,000 units in just 3 months Solid scaling progress