Manasi G.’s Post

China targets massive production of humanoid robots (Courtesy: Claude) The Key Manufacturers 1. Unitree Robotics (Hangzhou) Founded in 2016 by Wang Xingxing, Unitree employs roughly 1,000 people and ships robots to over 30 countries. Its flagship humanoid is the G1: priced from $13,500, weighing 35 kg, and standing 130 cm tall — though with a battery life of only two hours and a maximum arm load of just 2 kg for the standard model. They also offer a cheaper R1 model: starting at just $5,900 — the most affordable humanoid on the market. Unitree’s CEO has forecast deliveries of between 10,000 and 20,000 units in 2026. 2. AgiBot / AGIBOT (Shanghai) AgiBot shipped the most humanoid robots of any company globally in 2025 — over 5,100 units — according to Omdia. Its flagship A2 is built for service and interaction, carrying advanced natural language processing for real conversations during precision tasks, and is certified for the Chinese, US, and European markets. Entry price starts at $14,500. 3. UBTECH (Shenzhen) Shipped 1,000 Walker S and Alpha-series units in 2025, focusing on commercial and industrial applications. Other Chinese players: Galbot, EngineAI, Leju Robot, and Fourier each shipped between 150–500 units in 2025, backed by major investors including Alibaba, Bosch, CATL, Geely, JD.com, Tencent, and XPeng. What Jobs Can They Do? Logistics and manufacturing are the first key areas of deployment. Unlike conventional industrial robots bolted to a single task, humanoids can navigate unstructured environments and adapt on the fly. Specific use cases include warehouse work, factory floor tasks, unloading trailers, and even laundry folding in pilot home deployments. UBTECH’s models are also deployed in education, elder care, and services. Can They Work Independently? Partially, and improving rapidly. These next-generation machines are designed to handle a wider range of tasks at lower cost and with greater autonomy — meaning less need for human supervision on the factory floor, enabled by significant progress in embedded AI systems. However, a robotics professor cautions that many examples seen in the media are essentially demonstrations of what these systems can do. Full independent deployment at scale remains a 2030s milestone according to Morgan Stanley. The West’s Position Tesla and Figure AI each delivered only about 150 units in 2025 — a fraction of what Unitree and AgiBot shipped. Unitree is targeting a $7B IPO valuation and AgiBot a $6B valuation in 2026 , signaling the industry is moving fast toward commercialization. The gap is real and widening — this is essentially China running the same playbook it used for EVs and solar. The bottom line: China dominates on volume and commercialization speed; South Korea on density; Japan and Germany on precision and legacy manufacturing; the US on AI brains; and India is a future market rather than a current competitor. https://lnkd.in/ggzfWC3U

Last year, China's humanoid robots waved handkerchiefs at a tech demo. This year? They're doing backflips with nunchucks. This isn't CGI. This isn't fake. Not AI generated. This really happened. We're talking about a completely different species. 𝗦𝗼 𝘄𝗵𝗮𝘁 𝗵𝗮𝗽𝗽𝗲𝗻𝗲𝗱 𝗶𝗻 𝘁𝗵𝗼𝘀𝗲 𝘁𝘄𝗲𝗹𝘃𝗲 𝗺𝗼𝗻𝘁𝗵𝘀? The hardware was always there. Metal bodies, joints that bend, hands that grip. But the brain finally caught up. Think of it like this: you had a fancy car sitting in your garage for years, but no one knew how to drive. Now suddenly everyone's a race car driver. What was missing? The ability to process what they see, understand where they are in space, and make real-time decisions about balance, force, and coordination. The software. The intelligence. The part that connects sensing to movement without eating concrete. AI models that can see, understand language, and control physical movement finally matured. Large language models met vision systems met robotics simulation. And apparently "I should do a backflip while holding weapons" works too. 𝗡𝗼𝘄 𝘇𝗼𝗼𝗺 𝗼𝘂𝘁. → China has 140+ humanoid robot manufacturers → Morgan Stanley projects 28,000 units by 2026 → Unitree's G1 retails for $13,500 → Tesla targeting 1M humanoid robots at $20K each → Manufacturing costs down 40% in 12 months This isn't a science fair project anymore. This is mass production. With nunchucks. Five years ago, experts said useful humanoid robots were a decade away. That decade just collapsed into months. What this means for business leaders: 𝗜𝗳 𝘆𝗼𝘂'𝗿𝗲 𝘀𝘁𝗶𝗹𝗹 𝗯𝘂𝗶𝗹𝗱𝗶𝗻𝗴 𝟯-𝘆𝗲𝗮𝗿 𝗔𝗜 𝘁𝗿𝗮𝗻𝘀𝗳𝗼𝗿𝗺𝗮𝘁𝗶𝗼𝗻 𝗿𝗼𝗮𝗱𝗺𝗮𝗽𝘀, 𝘄𝗮𝘁𝗰𝗵 𝘁𝗵𝗶𝘀 𝘃𝗶𝗱𝗲𝗼 𝗮𝗴𝗮𝗶𝗻. The technology you're planning for in Year 3 will be unrecognizable by the time you get there. The question isn't "should we adopt AI?" It's "can our organization adapt fast enough to keep up with what AI is becoming?" Because the robots just did a backflip. And they're not slowing down. 💬 When you watch this video, what's your first reaction - excitement, curiosity, or "we need to move faster" or sheer fear? ---------------- Still on the fence? Still not sure where to start? Bring us your fears. Seriously. That's what we're here for. → Book a call here : https://lnkd.in/eMD2m-RY ------------- ♻️ Repost this. Your network and your leadership team need a wake-up call on the pace of change. ➡️ Follow Rujuta Singh for surprisingly honest truths on AI transformation & Innovation

One Model for Humans + Robots + Humanoids? This is UniT Architecture 🍓 Most robot systems are trained separately for each body type. Humans move differently. Robots act differently. Humanoids behave differently. UniT changes that. It creates a shared action language across embodiments so skills learned from humans can transfer to humanoid robots. --- I. Core Problem Today robotics suffers from: Separate training pipelines Huge data cost Poor zero-shot transfer Hard adaptation to new bodies UniT solves this by learning a unified token space for actions. --- II. Cross Embodiment Training The system learns from: Human demonstrations Humanoid robot actions Visual observations Motion trajectories Different bodies, same intent. Example: Human picks bowl Humanoid learns same goal with different joints. --- III. Shared Latent Action Space Encoders convert movement into compressed tokens. Then model maps: Human hand motion Robot arm motion Humanoid grasp motion into one shared representation. Meaning: Different hardware, same skill understanding. --- IV. VLA-UniT (Vision Language Action) Pretrained vision-language model + action expert. This enables: Understand scene Interpret task goal Predict correct movement Execute unseen tasks zero-shot Example: Human stacks bowls once Humanoid repeats task without retraining. --- V. WM-UniT (World Model UniT) Adds memory + video generation + prediction. System can: Imagine future frames Predict object motion Plan next step Correct trajectory before acting This makes robots proactive, not reactive. --- VI. Why Important Traditional robotics: Train one robot at a time. UniT: Train once, transfer across bodies. That means: Faster deployment Less data collection Better generalization Scalable humanoid learning --- VII. Real World Use Cases Warehouse humanoids Home assistant robots Factory automation Elder care robotics Cross-platform robot fleets Human imitation learning --- VIII. Big Insight Humans and robots may have different bodies, but actions can share the same language. That is what UniT builds. --- IX. Future of Robotics Next generation robots won't learn from scratch. They will learn from: Human videos Shared action tokens Simulations Other robots One intelligence, many bodies.

Humanoid Robots in Japanese Hospitals — A Strategic Signal, Not a Tech Story A recent article on Chinese AI startups deploying humanoid robots in Japanese hospitals and hospitality environments caught my attention. At first glance, it looks like a robotics experiment. It is not. 👉 It is a strategic signal about the next industrial battlefield. What is really happening? We are witnessing the extension of a model we already see in automotive and manufacturing: 👉 China is now exporting speed + cost + integration into AI-powered physical systems Humanoid robots are just the visible layer. Behind them lies a full stack: • AI training capabilities • Hardware manufacturing scale • Integrated supply chains • Rapid real-world deployment loops A paradox worth noting Japan is one of the most advanced countries in robotics. And yet… 👉 Japanese companies are testing Chinese humanoid robots 👉 Even with security and dependency concerns Why? Because: • Speed of deployment matters • Cost matters • Availability matters And increasingly: 👉 There are limited alternatives A familiar pattern If this feels similar to what we have seen in EVs — it is. Benchmarks from recent industry analyses (e.g., Roland Berger) show: • 20–40% faster development cycles • 20–30% cost advantages • Strong ability to industrialize at scale That same playbook is now being applied to robotics. The real shift This is not about robots replacing tasks. This is about: 👉 Who controls the next layer of industrial infrastructure Because humanoid robots will operate in: • Hospitals • Logistics platforms • Retail and hospitality • Industrial environments They will generate data, interact with customers, and execute operations. The strategic dilemma For advanced economies, the choice is becoming clearer: 1. Build internally → slower, more expensive 2. Adopt external solutions → faster, but creates dependency We are starting to see a pragmatic shift toward option 2. Why this matters for leaders This changes the nature of competition: • It is no longer just product vs product • It is system vs system And the key questions become: • Who controls the AI + hardware + data loop? • Who can deploy fastest at scale? • Who can balance performance with trust and security? Where opportunities lie This transformation also opens a new space: 👉 Combining: • Speed and pragmatism with • trust, regulation and system integration We are entering the next phase of industrial competition: From: • Digital transformation To: 👉 Embodied AI transformation And in this new phase: 👉 Speed is still critical 👉 Cost is still decisive 👉 But control of real-world AI systems becomes strategic This is not about robots. It is about who will shape the operating systems of tomorrow’s economy. #AI #Robotics #Industry #Transformation #Strategy #Leadership #Innovation

Humanoid robots are a $30,000 solution to a $300 problem. Most automation opportunities don’t need two legs. They need stability, efficiency, and uptime. Humanoid robots are not the answer for 90% of these use cases. I recently posted about the failed contenders in the Shanghai robot half-marathon, and a friend took issue: "One team had a robot that tripped and smashed, and that's what you focus on?" There were, in fact, many failed teams that had collectively spent millions of dollars in research, solving a problem that never needed a solution in the first place. As impressive as it was to see a robot beat a human in a speed run, there is no 'Honor' in that victory. CONSIDER THE CHEETAH If the researchers were serious about speed, they would have made a robot cheetah, or better still, a wheeled vehicle with a low center of gravity that wouldn't need pit stops to top up on dry ice to cool levered joints that should have been rotating axles in the first place. AMAZON HAVE THE RIGHT IDEA I recently toured an Amazon fulfillment center. It was full of robots. Not one of them was humanoid. It was a mix of arms, low-wheeled lifters, and scaffold-mounted articulators. This is what industrial automation looks like today. Humanoids frequently topple, risking human injury, frequent downtime, and expensive repairs. Strict regulations are on their way. Will the insurance be affordable? PR videos of humanoid robots doing something painfully slowly are designed to secure more VC funding for startups that are still years away from revenue. What's the point of a $30,000 humanoid robot holding a $200 vacuum cleaner, when a $300 Roomba can do the same task at a fraction of the energy cost, and at 10 times the speed, while also reaching the tricky spots underneath your furniture? A LESSON FROM NATURE • Primates: Upright, freeing hands for tool use. • Birds: Forelimbs for flight. • Kangaroos: Tendons to recycle kinetic energy. Nearly every other animal uses multiple limbs for stability. Why aren't our robots doing the same? There is nothing about humans that represents the zenith of evolutionary form; we look like this way because of a succession of random mutations and compromises that favored blood flow to the brain and flexible fingers. If we need to borrow a body plan from nature, consider the humble scorpion. Many legs, low center of gravity, stable base, forward-facing manipulators, and a tail that could swap out end-effectors from a spine-mounted tool rack. THE DANGER OF ANTHROPOMORPHISM Anthropomorphism puts humans at the center of the universe, just as our ancestors assumed the Earth was the hub of the solar system around which all other bodies orbited. But we know better now. I do not doubt that we'll have a future of publicly visible robots. But they will be as diverse as the natural world, not a facsimile of their creator. THE QUESTION Is your automation strategy focused on performance, or just 'human-centric' theater?

𝗜𝗻𝗱𝗶𝗮'𝘀 '𝗛𝗮𝗻𝗱 𝗙𝗮𝗿𝗺𝘀' 𝗣𝗼𝘄𝗲𝗿𝗶𝗻𝗴 𝗚𝗹𝗼𝗯𝗮𝗹 𝗛𝘂𝗺𝗮𝗻𝗼𝗶𝗱 𝗥𝗼𝗯𝗼𝘁 𝗗𝗲𝘅𝘁𝗲𝗿𝗶𝘁𝘆 𝗧𝗿𝗮𝗶𝗻𝗶𝗻𝗴 🛰️ [ROBOTICS] Indian 'hand farms' provide critical human motion data for training humanoid robots. Why it matters: This hidden industry is a crucial, low-cost backbone for the global humanoid robotics race, enabling machines to learn complex human dexterity. It highlights the significant human labor required to bridge the gap between AI ambition and practical robotic capabilities. 🤔 As humanoid robots become more capable, what are the long-term ethical and economic implications for the human labor currently training them? #Robotics #HumanoidAI #DataLabeling #AIethics #GlobalLabor 📡 Follow DailyAIWire for autonomous AI news 🔗 https://lnkd.in/dppGkrbH

Yes! The first phase of our technical robot benchmark is complete: 3 humanoid robots, 350+ technical data points, and conclusions for operators and investors that differ from the market narrative. As humanoid robots scale, industrial users, operators, and investors need to deeply understand their technology. However, public technical benchmarks for humanoid robots are still rare. So we started one (for our investments): we are building an ongoing benchmark series for industrial humanoid robots — strictly based on technical analysis, not demos, headlines, or AI claims. Our first desk research benchmarking phase compares the best-selling humanoid robots for industrial use: ➡️ Unitree H1-2 ➡️ AgiBot A2 Ultra ➡️ UBTECH Walker S2 We compare them across six categories: 1️⃣ Platform architecture 2️⃣ Actuation, joint performance & locomotion 3️⃣ Manipulation, hands & payload 4️⃣ Perception & sensor stack 5️⃣ Compute, software stack & AI layer 6️⃣ Power, charging & runtime High-level summary of the findings of this phase (350+ data points, comment BENCHMARK if you want a copy.): ✅ UBTECH Walker S2 currently presents the strongest overall package for industrial use: multi-shift architecture with autonomous battery swap in <3 minutes, the strongest heavy manipulation package with 15 kg payload, a 0–1.8 m workspace, 7.5 kg single-hand grasp, and 1 kg finger grasp, etc. etc. ✅ AgiBot A2 Ultra shows the most advanced SW / AI platform for integration into a broader agentic physical AI setup: the strongest developer-to-deployment stack, including AimRT, AimDK, HTTP JSON RPC, ROS2 Topic interfaces, ROS2 Humble with FastDDS, etc. etc. ✅ Unitree H1-2 is currently more a general robotics platform than a focused industrial deployment package. However, it comes with with the strongest hand-level tactile and precision capability via the dexterous hand stack and the strongest lower-body raw actuation reserve, e.g. knee torque around 360 Nm. Current bottom line: ➡️ Best suited for industrial deployment today: UBTECH Walker S2 ➡️ Most modern overall platform: AgiBot A2 Ultra We turned the first benchmark results into a technical fact sheet: • 15+ pages • 6 technical categories • 350+ technical data points • directly comparable specs We created the fact sheet for our own use, but we are happy to share it if anyone is interested (we need to be connected). Comment BENCHMARK if you want a copy. This is only the start: we will update and expand the benchmark, add more humanoid robots as they become industrially relevant, and over time move toward teardown-based and physical comparative analysis. // No investment advice and no guarantee as to the completeness or correctness of the provided data. //

Humanoid Robots: The Emerging Workforce Disruptor Last Sunday (April 19, 2026), a humanoid robot named "Lightning" ran a half-marathon in Beijing in 50 minutes — beating the human world record by over six minutes. Millions of people suddenly realized this technology is no longer science fiction. But the reality is more nuanced: humanoid robot capability is accelerating at a logarithmic pace but at the same time is currently more limited than the hype suggests. What's actually happening right now: ·       Figure AI robots are working shifts at BMW in South Carolina ·       Amazon's Agility Robotics unit handles warehouse logistics ·       Tesla's Optimus performs material handling in Fremont ·       China is already shipping 5,000+ units annually What's NOT happening yet: Robots still struggle badly outside controlled environments. A home, a hospital corridor, a hotel room — novelty breaks them. Goldman Sachs sees mass adoption as a 2030s story, not today's.  But that’s not too far off…. The workforce question nobody wants to think about: The jobs in the crosshairs aren't executive roles. They're hotel housekeeping, warehouse picking, hospital orderly work, and domestic cleaning — physically demanding, lower-wage positions, performed by workers with the fewest resources to adapt. McKinsey estimates physical tasks make up over half of working hours for ~40% of the US workforce. Humanoid robots don't threaten one industry. They threaten a category. The geopolitical dimension: China leads in production volume. The US leads in AI and software. The company that wins humanoid robotics at scale gains a manufacturing cost advantage that ripples through every industry it touches. Including the military. Elon Musk says work will be "optional" within 20 years and predicts an end to poverty as a result of humanoids. Whether that means shared abundance or concentrated displacement depends almost entirely on policy choices no major economy has seriously designed yet. Whether mass adoption of humanoid robots will be good thing societally is a highly charged, controversial question and, like most questions about AI implementation, the answer depends on your time horizon.

Last Sunday, Beijing hosted a humanoid "foot" race in which a robot made in China beat the existing human record for a half marathon. This event was heavily covered by the media - the implications were not. For a condensed assessment, see our blog: https://lyrixis.com/blog

Robot 'interns' hit the factory floor in S China's Guangxi Industrial humanoid robots have begun "interning" on the factory floor at Dongfeng Liuzhou Motor Co., Ltd. in Liuzhou, south China's Guangxi Zhuang Autonomous Region. At a command from their instructor, Lu Baichun, the 11 robots instantly activate their visual navigation systems and head straight to their assigned stations. The factory has set up a 200-square-meter materials handling area as a training ground for UBTech's Walker S1 robots, where they practice moving bins, sorting parts and collecting empty containers. Liuzhou, one of Guangxi's key industrial centers, offers a wide range of real-world scenarios for testing and deploying AI. The robots train here to adapt to on-site conditions and optimize their workflows before eventually joining regular production lines. "The robots rely heavily on visual recognition," Lu said. "Changes in lighting, humidity or the surrounding setup can all affect how they perform. Real factory environments are far more complex, so learning to handle the variables is one of the key challenges they need to overcome." Although the robots are still "apprentices," Lin Changbo, general manager of Dongfeng Liuzhou Motor, is confident about the future of human-machine collaboration. "We expect them to keep improving efficiency, lowering costs and taking on physically demanding and high-risk tasks," he said. "They'll inject new intelligence into traditional manufacturing." To overcome the data challenges in expanding embodied AI for industrial use, Liuzhou has set up an embodied AI data collection and testing center. The center functions as a robot training school, recreating full-scale production environments from key local industries, including automobiles, construction machinery, pharmaceuticals and river snail rice noodles, or Luosifen, a local delicacy. It generates tens of thousands of training data points each day. At the center, 120 industrial humanoid robots train alongside their instructors, honing their skills for factory work. https://lnkd.in/gADRpPhX