A humanoid robot fielded by the Tianzhuo team ran the 100 meters in 9.39 seconds at the second World Humanoid Robot Games in Beijing on the evening of August 22, 2026, finishing a body length clear of the field. The time is below the 9.58-second human world record, and it was set in a race the organizers had upgraded this year to admit fully autonomous machines only, with no human steering the robot down the track.
Key facts
- Winning time: 9.39 seconds, in a preliminary heat run before the opening ceremony.
- Date and place: August 22, 2026, at Beijing's National Speed Skating Oval, the venue nicknamed the Ice Ribbon.
- Scale: 51 events, split into 30 competitive events and 21 real-world scenario challenges, running through August 26.
- Primary source: China National Radio's report, August 23, 2026.
The number that matters here is the gap between two records that are not quite the same thing. World Athletics credits Usain Bolt with 9.58 seconds over 100 meters, set in Berlin in 2009 and unbeaten since. A machine covering the same distance in 9.39 seconds is, on the stopwatch, faster. It is also running on a track by itself, under rules written for robots, in a heat rather than a final, and against no human at all. Both facts are true, and the honest version of the story keeps them together rather than picking one.
What makes this edition different from last year's is the autonomy requirement. Beijing's own preview of the Games says the 100 meters was changed to a fully autonomous robot-only race -- the machine has to perceive the track, keep its balance, and steer itself. That is a much harder engineering problem than raw actuator speed. A wheeled or teleoperated platform can be made to go fast; a bipedal robot that runs a straight line at sprint pace while stabilizing its own gait, without a human in the loop, has to solve balance and control at the edge of what its hardware can physically do. We covered the rule change when the schedule was announced; this is the first result under it.
Think of it as the difference between a remote-controlled car and a self-driving one. Both can be quick. Only one of them has to know where it is.
The organizers are candid that the point is not athletics. Liu Weiliang, deputy director of Beijing's Municipal Bureau of Economy and Information Technology, framed the event this way in the state broadcaster's report: "Robots are not just a technology, they are an industry, and moreover a future we all have to face together." That is an industrial-policy sentence, not a sporting one, and it explains the program's shape. Alongside the sprint and the football, this edition added tug-of-war and weightlifting, plus table tennis and tennis, and revived two traditional Chinese sports -- tai chi and pitch-pot -- as control and precision tests. Twenty-one of the 51 events are scenario challenges rather than races: tasks staged in something closer to a factory or a home than a lab.
Why it matters: track events are a legible, hard-to-fake benchmark for locomotion, and legible benchmarks are how a field learns whether it is actually improving. Chinese manufacturers dominate the entry list, and the Games double as a shop window -- Unitree, one of the field's best-known names, has been describing itself as a rare profitable humanoid maker. A public leaderboard for balance, gait, and autonomous control is worth more to that industry than any demo video.
The caveat is that a sprint is one of the narrowest capabilities a robot can have, and the surrounding claims are much weaker than this one. Comparisons circulating alongside the sprint result -- a robot "standing jump" set against the human high jump record, or a top-speed figure set against a full-race average -- are category errors, measuring different events entirely. And the 9.39 came in a preliminary, on a prepared indoor surface, over a straight line. Nothing about it says a humanoid can navigate a warehouse, and the organizers' own decision to spend 21 of 51 events on real-world scenarios suggests they know it.
Originally published on Ground Truth, where every claim is checked against the primary source.
Top comments (0)