Two major races were won and lost in Beijing on August 26. On one track, a humanoid robot ran 100 meters in 8.64 seconds, nearly a full second under Usain Bolt’s world record. On another, the same robot slammed into the padded wall at the end of the race track.
Global headlines trumpeted the new robot record, but its dramatic crash after the finish line was arguably the more meaningful story.
Start with the speed. The winning 100 meters at last year’s first World Humanoid Robot Games took 21.50 seconds. This year, the record fell three times in five days: first 9.39, then 8.86, then 8.64 – all set by one sprinting machine, Tiangong Ultra, built by Beijing’s X-Humanoid.
To put that into perspective, robot sprint times more than halved in 12 months. Yet a machine that runs at record speed and can’t stop isn’t a finished product by a long shot.
Medicine has a word for such a gap. Efficacy is what something achieves under ideal conditions – in this instance, a clean track, a fully charged battery and the robot’s makers watchfully nearby.
Effectiveness is what something achieves in ordinary use, under less-than-ideal operating conditions, only partially charged, and maintained by people who didn’t build it. The first tells you very little about the second, which is why drug approvals come with registries and adverse-event reporting.
Chinese robotics is producing efficacy at a record-breaking clip and at the same time very little in effectiveness. An 8.64-second sprint tells a warehouse manager nothing about how a machine might behave and function at month 11, or more consequentially when someone steps into its path.
Jonathan Hurst of Oregon State University put it plainly at the Beijing Games: progress will be measured where humanoids “do work autonomously for hours and generate real economic value.”
The event’s organizers know that well. Alongside the sprints, 2,056 robots from 666 teams competed in 51 events and 1,301 contests covering warehousing, assembly and emergency response. Beijing says the 2027 Games will lean further in that practical direction.
Look at China’s robotics industry underneath the spectacle of racing robots, and it is strikingly lopsided. China installed 295,000 industrial robots in 2024, 54% of the world’s total, and now runs more than two million, roughly 4.5 times Japan’s stock and five times America’s.
For the first time, Chinese manufacturers outsold foreign suppliers in their own market, taking 57% of their own market. No other country is close.
The pressure behind China’s robotics drive is demographic. China’s over-60 population is set to pass 400 million by 2035 while well over 100 million people still work in its factories. Automate too fast and the wage shock will adversely hit millions of households. Too slow, and the economy will run short of hands.
Japan, Korea and Germany face the same dilemma. It is a demographic problem wearing a geopolitical costume. Which is why the most valuable thing Beijing produced last week was not the shattering of Usain Bolt’s sprint record.
On the Games’ closing day, the organizers released 2,500 hours of operational data from training and competition, covering 12 scenarios and more than 10,000 tasks, free to companies, universities and research institutes.
If humanoids are ever going to be safe and useful in a warehouse or a hospital ward, China is the country best placed to prove it. It has the deepest deployment base, the most complete supply chain, a home market buying homegrown machines, the demographic push to keep going and an annual arena that stress-tests robots in public and gives away the data.
Nobody else has that combination. The US has stronger disclosure institutions and a thinner industrial base. Europe has the safety-standards tradition and shrinking factory floors on which to apply it. That is, the capability to produce humanoid robots sits in one place, while the tradition of publicly publishing on development failures sits in another.
China publishes records, valuations and dataset volumes, but not how often a machine falls or how long one runs between faults. Neither does anyone else. The blind spot is industry-wide, and China is simply the country that could close it fastest.
But the money has moved without waiting for proof of concept. This month, China’s Unitree became the first humanoid maker to list on a mainland exchange, closing its Shanghai debut up 460% at roughly US$50 billion. The sector now has a public price without a public safety record.
Significantly, Washington is moving the other way. On July 28, the US Federal Communications Commission added foreign-made advanced robots to its Covered List, blocking new humanoid and quadruped models from the authorization needed to sell in the US with existing models grandfathered.
The stated reason was cybersecurity: a flaw that let a remote actor into thousands of foreign-made robots inside American homes, including cameras and microphones.
But while the US government and others have built real evidence-gathering for robot security, there is still nothing on safety. An import ban tells a hospital nothing about whether a 60-kilo machine is safe next to a nurse.
Three fixes would change that, and none requires a hard-to-get geopolitical agreement. For one, China should publish credible reliability data. Mean time between failures under load is what a buyer needs to know – and no Chinese or other manufacturer discloses it.
Two, build an incident registry, along the lines of drug safety monitoring or aviation accident investigation. Falls, collisions and burning batteries are adverse events. Sharing information on development failures made medicine and flying far safer, without slowing either down.
Three, measure retraining as carefully as we measure machines. How many workers move into robot maintenance and integration, and how fast, is a number nobody publishes either.
None of this is technically hard, and the pieces are already on the table. They just sit in different countries: China has the humanoid fleet, the arena and the data; regulators in the US, Europe and Japan have run safety registries for decades without killing the industries they oversee.
Insurers price risk for a living. Someone has to put these three together so the industry can race ahead, not slam into a protectionist wall.
Y. Tony Yang is an Endowed Professor at the George Washington University in Washington, D.C.
