Out of nearly 1,400 booths at Japan's biggest robotics expo in 2026, zero Japanese-built humanoids were doing real work. The country that built the world's first full-scale humanoid in 1973, four decades before anyone else understood why the world would need one, didn't have a single one on a factory floor at its own show.
The one humanoid Japan did bring was playing basketball.
The Expo Where Japan Had Nothing to Show
Nagoya hosted Robot Technology Japan 2026 over three days in early 2026. More than 270 companies filled nearly 1,400 booths. A large share of the exhibits came from China, and the humanoids doing actual demonstrations belonged almost entirely to Chinese firms.
Dobot's Atom Max handed popcorn to visitors and demonstrated movement precision down to 0.05 millimeters, roughly the width of a human hair. Its real selling point was a learning system: an operator demonstrates a task once by remote control, the robot records the trajectory and joint positions, and that same motion data transfers to any other machine in Dobot's lineup, from humanoids to underwater manipulators.
At the next booth, Limx Dynamics showed off Ali, a robot that took a string of voice commands in sequence, remembered the task cue, located the right objects on its own, and executed the actions without stopping between them. That kind of unbroken, multi-step execution is rare. Most polished robot demo videos are stitched together from several takes.
Japan's own exhibitors stuck to classic industrial robotics. Kawasaki Heavy Industries showed three robots painting a single part as a coordinated team, one positioning it, one applying the coating, one checking the result with a camera. Impressive engineering, but none of it humanoid. Toyota's Q7 was the only human-shaped Japanese robot at the entire expo, and it spent the show sinking basketball shots from a moving two-wheeled platform.
Every practical humanoid at the expo, without exception, sat at a Japanese distributor's booth, already positioned for sale into local factories. Japan had, in effect, outsourced its own humanoid supply chain to China before the show even opened.
Why Japan Walked Away in 2018, Not 2026
The instinct is to read this as a country falling behind in real time. The timeline says otherwise. Honda ended development of Asimo, the robot that had represented humanoid robotics globally for twenty years, in 2018.
Honda didn't shelve the project quietly. It redirected the underlying work into a remote-controlled avatar robot with a multi-fingered hand gripping five times harder than Asimo's, built to assist a human operator rather than replace one, compensating for inaccuracies and holding grip balance steady. External trials began in 2023, with commercialization targeted for the 2030s.
That's the real starting point. Japan opted out of the general-purpose humanoid race eight years before China's manufacturing scale made the decision look inevitable. What it chose instead was harder to put in a highlight reel.
Where Japan's Robots Actually Went
In March 2026, exactly fifteen years after the disaster, TEPCO deployed a 22-meter robotic arm, roughly the height of a seven-story building, into the Fukushima Daiichi site to extract samples of melted nuclear fuel from three damaged reactors. Radiation levels there run between 4.7 and 9.7 sieverts per hour, a dose that would kill a person within minutes. Until 2024, engineers had managed to extract only a microscopic sample from the debris.
More than $40 billion has gone into the decommissioning effort so far. TEPCO has reaffirmed 2051 as its target for finishing the job, a goal it has held to publicly even as retrieval timelines keep slipping and independent experts question whether it's achievable at all.[1] Several generations of robots have failed here already, some stuck in wreckage within hours, others lost in flooded reactor water. This is precisely the kind of job with no audience and no tolerance for a polished demo reel.
A similar pattern shows up in space. Japanese startup GITAI has tested robotic arms both inside and outside the International Space Station and holds contracts with NASA and the US Space Force, including one to build a lunar rover. GITAI exists because of one of Japanese robotics' most public losses: in 2013, the startup Shaft beat teams running on Boston Dynamics hardware at the DARPA Robotics Challenge, only for the Japanese government to decline funding it. Google bought Shaft, later shut it down, and its founder went on to lead GITAI, explicitly choosing to avoid the cheap mass production race that he says killed his first company.
Since spring 2026, a humanoid called Maholo has worked overnight at the Tokyo Institute of Science, using standard lab pipettes and centrifuges without any retrofitting, running experiments in an empty lab through the night. Its project lead claims the approach accelerates the search for new medicines by tens or even hundreds of times, a figure that comes from the lab itself and hasn't been independently verified. Japan's Ministry of Science has since put close to $50 million toward a similar automated lab in Okazaki.
Then there's the quieter end of the list. The therapeutic robot Paro has worked in Japanese nursing homes for over fifteen years with no silicone skin and no hype cycle, just a long, unglamorous track record. And in Tohoku, a university startup called Kuiden Tech is preparing mass production of an autonomous farm platform whose rough-terrain mobility system was originally built for lunar rovers, repurposed from the moon to aging Japanese farmland.
Two Different Bubbles
The humanoid industry currently runs on two incompatible bets. The Western bet is on future valuation. Figure AI's valuation climbed from $2.6 billion to $39 billion in roughly eighteen months, backed by an estimated 150 robots actually in operation. That gap between price and product is close to the textbook definition of a bubble, one that pops the moment the market wants shipped units instead of demo footage.
China's bet is on volume. The country already had the component supply chain, motors, gearboxes, batteries, fine-tuned from years of building drones and EVs. Instead of one perfected prototype, Chinese firms ship thousands of simpler machines into warehouses and factories, let them fail in the field, and iterate faster than a competitor can finish a single polished release. Estimates put China's share of global humanoid production in 2026 at 80 to 90 percent. Unitree has already turned a profit, with gross margins reported as high as 60 percent.
| Player | Strategy | Signal |
|---|---|---|
| Figure AI (US) | Valuation-led | $39B valuation, ~150 robots deployed |
| Unitree (China) | Volume and margin | Profitable, up to 60% gross margin |
| UBTECH (China) | Industrial and consumer split | ~1,000 industrial units sold in 2025; silicone-skin companion robot outsold that in pre-orders within weeks |
Even within China, the picture splits in two. UBTECH, the world's largest humanoid manufacturer, sold only about a thousand full-size industrial robots across all of 2025, despite an Airbus contract to test its Walker S2 on an aircraft assembly line. Its companion robot with silicone skin then out-booked its entire industrial business in the first few weeks of pre-orders alone. China is running a real industrial rollout and a consumer hype cycle at the same time, and it's not yet clear how much of the second one will actually reach a buyer's living room.
Noetra: Japan's Bet on a Brain for Any Body
In early July 2026, Japan's government set a target of 10 million AI-equipped robots in operation by 2040, roughly one for every 10 to 12 residents, backed by government funding of up to $6.1 billion.[2] The plan is not to out-manufacture China. It's to build the intelligence layer, a foundation model that can in theory run on any robot body, industrial or humanoid.
That effort sits with a new consortium called Noetra, majority-owned by SoftBank, NEC, Sony, and Honda, formally commissioned by Japan's METI and NEDO. Training data is being collected inside ordinary Japanese homes and industrial plants to build a model grounded in real household and factory tasks, not just internet text.
This is also where Toyota's basketball-playing robot stops looking like a stunt. Sinking a shot from a moving two-wheeled platform requires connecting vision, balance, body control, arm strength, and the trajectory of a flying object in real time, the same whole-body control problem Boston Dynamics is tackling with Atlas. Toyota Research Institute trained a large behavior model that Boston Dynamics now uses in Atlas, letting the robot chain long sequences of actions without new code for each one. Basketball is a visual stand-in for a research problem, not the point itself.
The irony is that even this quiet strategy isn't fully Japan's own. Toyota trains its most capable robot in the US, on hardware from Boston Dynamics, a company that passed through Google and SoftBank before landing in Korean hands. The same chain of corporate deals that once cost Japan the Shaft team now runs through the hardware backing its own comeback plan.
My Take
Judged purely on units shipped, China has already won this round, and it isn't close. 80 to 90 percent of global humanoid production for 2026 isn't a lead, it's a different market entirely. But "who wins" depends on what you're measuring. If the measure is units on a factory floor, China wins outright. If it's valuation per robot deployed, American startups look increasingly disconnected from their own product. If it's whether a country built the one thing hardest to copy, an AI model that survives contact with real nuclear debris, real drug discovery, and real farmland, Japan hasn't lost anything yet because it never entered that particular race.
My honest read: Noetra is the more interesting bet of the three, and also the one most likely to underdeliver on its own timeline.
- Japan's only humanoid at the 2026 Nagoya expo, Toyota's Q7, was demonstrating whole-body control through basketball, not doing factory work.
- Honda ended Asimo development in 2018, redirecting its robotics work toward a remote-operated avatar robot rather than a general-purpose humanoid.
- China is projected to produce 80 to 90 percent of humanoids globally in 2026, with Unitree already profitable at up to 60 percent gross margin.
- Figure AI's valuation jumped from $2.6 billion to $39 billion with roughly 150 robots actually deployed.
- Japan's robots are concentrated in high-radiation, high-complexity niches: Fukushima debris removal, ISS and lunar rover work through GITAI, overnight pharmaceutical labs, and farm automation.
- Japan's Noetra consortium, backed by up to $6.1 billion in government funding, aims to deploy 10 million AI-equipped robots by 2040, betting on a shared AI model over humanoid mass production.
For a closer look at the Chinese side of this split, see our rundown of 7 of the most realistic Chinese humanoid robots right now, or how Mitsubishi is scaling humanoid production at its Kyoto plant on the Japanese side. UBTECH's pricing swings, mentioned above, are covered in more depth in our breakdown of the U1's announced-versus-shipped price.
FAQ
Why doesn't Japan make its own humanoid robots anymore?
Honda ended development of Asimo in 2018 and redirected its robotics investment toward a remote-controlled avatar system instead of a general-purpose humanoid. Japan's robotics industry has since focused on high-complexity niches like nuclear decommissioning, space robotics, and pharmaceutical automation rather than competing on humanoid mass production.
How many humanoid robots does China produce compared to other countries?
China is projected to account for 80 to 90 percent of humanoid robot production globally in 2026, driven by an existing component supply chain built for drones and electric vehicles.
What is Noetra?
Noetra is a Japanese robotics AI consortium majority-owned by SoftBank, NEC, Sony, and Honda, commissioned by Japan's METI and NEDO. It's building a foundation model intended to run across different robot bodies, backed by up to $6.1 billion in government funding, with a national goal of 10 million AI-equipped robots deployed by 2040.
What robots is Japan using at Fukushima?
TEPCO deployed a 22-meter robotic arm at Fukushima Daiichi in March 2026 to extract samples of melted nuclear fuel from damaged reactors, where radiation levels reach up to 9.7 sieverts per hour. TEPCO's stated target for completing decommissioning is 2051, though the timeline has slipped repeatedly.
Conclusion
If the humanoid bubble bursts in the next couple of years, Japan's absence from the race will look less like a failure and more like the only sober bet in the room. If it doesn't burst, Japan will have sat out a window that never reopens. Either way, while the rest of the industry argues over whose humanoid walks more smoothly on an expo stage, a Japanese robotic arm is still working, alone, inside a reactor no human can enter.
0 Comments