Humanoid robots used to be a research curiosity; at the second World Humanoid Robot Games they now race faster than the fastest human alive. At the opening in Beijing’s National Speed Skating Oval, the Tiangong Ultra humanoid ran 100 meters in 9.39 seconds, breaking Usain Bolt’s 9.58-second world record, with 2,056 robots from 666 teams across 16 countries competing in 51 events.
The Games
The second World Humanoid Robot Games opened on August 22 at the “Ice Ribbon” National Speed Skating Oval in Beijing, co-hosted by the Beijing municipal government, China Central Television, the World Robot Cooperation Organization, and the Asia-Pacific Robot Cup International Council. The numbers are a dramatic jump from the inaugural games: 666 teams and 2,056 robots from 16 countries across six continents, with team count up 138% and robot count quadrupled from last year. The event expanded from 26 to 51 events, comprising 30 competitive events and 21 scenario-based events, for a total of 1,301 matches. New events include long jump, weightlifting, tug-of-war, and table tennis, while scenario events cover industrial, logistics, hotel, home, and fire-rescue settings.
China’s domestic participation dominates: 641 teams and 1,975 robots from 157 companies and 200 universities and research institutes, with 27 of the country’s 985-project universities competing. Internationally, the United States, Germany, and Japan sent teams, and Brazil assembled a national squad from five robot football World Cup champion teams.
The Results That Broke Human Records
The opening ceremony’s 100-meter preliminary for the large category produced the headline result: the Tiangong Ultra, from the Beijing Humanoid Robot Innovation Center, crossed the line in 9.39 seconds, breaking Usain Bolt’s 9.58-second human world record. At last year’s games, the best 100-meter time was 21.50 seconds, so the improvement in one year is roughly 12 seconds, an absurd rate of progress.
The same robot took first in the 400-meter preliminaries with 39.70 seconds, also beating the human world record of 43.03 seconds. In the standing high jump demonstration, the Tiangong Ultra cleared 2.8843 meters, far beyond the human standing-jump record of about 1.6 meters and a massive improvement over last year’s champion mark of 95.641 centimeters.
Honor’s humanoid robot, Lightning, which won April’s humanoid half-marathon, finished the 400 meters in 41.95 seconds, also under the human record, and will compete in the 100-meter and 1500-meter events. The pattern across results is consistent: the gap between the best humanoid robots and the best humans in athletic events is closing, and in several disciplines it has already closed.
Fully Autonomous Operation
The organizers made a deliberate change this year: multiple competitive events removed manual remote control entirely, requiring the robots to run fully autonomously, with onboard large models handling perception, decision-making, and adaptation. That is the difference between a remote-controlled stunt and a genuine test of embodied AI.
The opening ceremony showcased the state of the art in coordinated autonomy. Eighty Booster T2 humanoids marched in step, each independently computing its path and adjusting its posture dynamically, and collectively rearranged their formation to spell “BEIJING.” Fifty Tiangong embodied-intelligence 3.0 robots navigated a bend in tight formation and performed the high-difficulty “Thomas flips” move. Twenty-five Galbot G1 wheeled robots completed long-horizon, fully autonomous tasks across home, retail, and restaurant scenarios, including the dexterous manipulation capabilities the company has been publicizing.
What This Means
The games are a benchmark the way benchmarks should be: directly observable, hard to game, and easy to compare year over year. A 100-meter sprint on a track is a cleaner measure of locomotion capability than any simulator score, and the year-over-year data, from 21.50 seconds to 9.39 seconds, is the fastest-improving athletic curve in history because it is a robotics curve.
The results also show where the frontier is moving: from locomotion to full-stack autonomy. Running fast matters less than running autonomously, perceiving the environment, and making decisions under competition conditions. The switch to autonomous operation this year signals that the industry has crossed a threshold where the models, not the actuators, are the bottleneck, which is exactly where AI capability connects to hardware.
The competitive picture is worth reading carefully. China’s dominance in participation is overwhelming, but the international entries, including Brazil’s robot football champions, show the field is globalizing. The robotics supply chain, motors, actuators, sensors, batteries, and the AI models that run them, is becoming a strategic industry in the same way that AI chips did, and events like this are as much industrial showcases as athletic competitions.
The autonomy results connect to the same embodied-AI frontier we have been tracking: NVIDIA’s open VLA model for vehicle control shows the perception side of robots acting in the physical world, covered in our NVIDIA Alpamayo 2 Super review, and the pattern of AI breaking human benchmarks has a mathematical cousin in the Astra proof breakthroughs we documented earlier. The games are the physical-world counterpart to those results: measurable, public, and improving at a pace that makes last year’s numbers look like a different era.
The Honest Caveats
The record comparisons come with asterisks. The 100-meter time of 9.39 seconds was run in a preliminary heat, not a final, and race conditions, track, and the robots’ battery states are not directly comparable to a human championship. The human records were set under different equipment rules, and robot running mechanics differ from human running, with different energy and stride constraints. The standing high jump comparison is similarly apples-to-oranges. The results are impressive as engineering, but “beats the human record” should be read as “beats the human record under robot-specific conditions.”
Autonomy claims also deserve scrutiny. “Fully autonomous” means no manual remote control, but the robots may still receive environmental cues, pre-programmed strategies, or wireless coordination, and the boundary between autonomous and assisted is defined by the organizers, not an independent auditor.
Who Should Care
Robotics companies should read the year-over-year results as the competitive baseline for embodied AI. AI researchers should watch the autonomy shift, since the games are becoming a public benchmark for perception and decision-making under physical conditions. Supply-chain watchers should note the strategic dimension of humanoid robotics. And anyone tracking AI’s physical frontier should treat the 9.39-second 100-meter as the number to beat, because a year from now it will be someone else’s headline.
The Bottom Line
The second World Humanoid Robot Games delivered the clearest signal yet that humanoid robots are exiting the demo phase: 2,056 robots competing autonomously, 51 events, and multiple human athletic records broken. The 100-meter result, 9.39 seconds against Bolt’s 9.58, is the headline, but the real story is the shift to fully autonomous operation, which moves the bottleneck from hardware to AI. The caveats about race conditions and autonomy definitions are real, but the direction is not in doubt: embodied AI is improving at a pace that makes year-over-year comparisons absurd, and this is the year the robots stopped being curiosities and started being competitors.