A humanoid robot ran 100 meters in 8.64 seconds this week in Beijing, almost a full second faster than Usain Bolt's 2009 world record. Another jumped 3.4 meters straight up from a standing start, breaking a human vertical-jump record set in 1993. A third faced off against a human Olympian in an exhibition table-tennis match. These were the highlight reels from the second annual World Humanoid Robot Games, which ran August 22 to 26 and drew more than 600 competing teams, mostly from Chinese universities, research institutions and robotics companies. [1]
The same games also featured robots trying to hammer nails into a corkboard — and mostly failing, even when human operators wearing motion-tracking gloves were steering their every move. [1] That gap between the highlight reel and the workbench is the actual story the games told this week, and it is a story state broadcasters covering the event have little incentive to lead with.
The sprint record came from a robot built by the Beijing company X-Humanoid, whose entry won gold with that 8.64-second time — a dramatic jump from the same robot's own winning time of 21.5 seconds in last year's competition. [1] A separate X-Humanoid robot broke the 1993 human vertical-jump record with its 3.4-meter leap, according to the state-run Chinese outlet Global Times, which covered both records as evidence that embodied artificial intelligence — AI systems that learn, plan and execute real-world physical actions — has crossed a meaningful threshold. [1] State broadcaster CCTV+ ran its own footage of the games, including the nail-hammering event, without drawing the same conclusion from the failures that it drew from the triumphs. [1]
Stanford computer scientist Karen Liu, who was not involved in organizing the games, explained why the split between spectacular athletic feats and mundane manual tasks isn't a coincidence. "If I ask you which is harder — folding laundry or doing a backflip — most people will say folding laundry is way easier," Liu said. "But for humanoids, doing a task that we do in our daily lives is actually way harder than a backflip." [1] Sprints and jumps, she explained, follow a prescribed physical trajectory that a robot's control system can calculate in advance and execute with mechanical precision. "Every single millisecond, it knows what to do and how much torque to generate," Liu said. "It doesn't need to 'think.'" [1]
Hammering a nail requires the opposite kind of intelligence: continuous sensory feedback, fine force control and a gripping mechanism that can mimic the strength and dexterity of a human hand. Liu said that remains one of the hardest unsolved problems in robotics, because robotic grippers generally cannot create enough sustained contact with a tool to apply consistent, strong force — the exact failure mode the games' corkboard event exposed in front of an audience. [1]
The 2026 games leaned deliberately into that gap. Unlike last year's event, this edition expanded beyond track-and-field spectacle into what organizers called "scenario-based events" — shelving books, making beds, and the nail-hammering task — scored specifically to reward robots that completed the work autonomously over those remotely operated by a human handler. [1] Cheng Hao, chief executive of Booster Robotics, which brought nearly 300 robots to the competition, framed the whole event as diagnostic rather than promotional. The games, he said, "are not simply a competition, but an important real-world testing ground for embodied intelligence." [1]
That framing found an unlikely ally outside China. Aya Durbin, who directs development of the Atlas humanoid robot at Boston Dynamics in Massachusetts — a company that did not send any robots to Beijing — agreed with the substance of the practical-task emphasis even from the sidelines. "Humanoids won't become a part of our everyday lives if they're always 'toys,'" Durbin said. "They need to be used to solve actual problems and make the world a better place." [1] Her comment lands as an implicit rebuke of exactly the kind of coverage that leads with a sprint record and buries the hammering failure — a critique from within the industry that state media summaries of the games did not carry.
None of this erases what the sprint and jump numbers actually show: humanoid hardware has advanced fast enough in a single year to beat a human world record that stood for 33 years and to nearly triple its own sprint speed since last year's games. That is real, measurable progress, and it is the kind of number that travels easily into a headline. But the corkboard sat right next to the track the whole time, and the nails tell a different story about how close these machines actually are to doing the boring, dexterous work — laundry, carpentry, dishes — that would make them useful rather than merely fast.
-- KENJI NAKAMURA, Tokyo