News · 2026-08-17

Beijing's robot 100-meter dash goes fully autonomous this month

The 2nd World Humanoid Robot Games opens August 22 in Beijing with a rule change that matters more than any result will: the 100-meter race is now restricted to fully autonomous robots. Beijing's municipal government says the five-day event at the National Speed Skating Oval runs 32 events across competitive and scenario-based categories, and that the scenario competitions leave the purpose-built test arenas used last year for real factories, hotels and model homes.

Key facts

• The games run August 22 to 26, 2026 at the National Speed Skating Oval in Beijing, with 32 events, per Beijing's official announcement.
• The 100-meter race is upgraded to fully autonomous robots only; scenario-event rules encourage fully autonomous positioning, recognition and operation.
• New dexterity tasks include folding clothes in a home setting, extinguishing fires in emergency rescue scenarios, and preparing cooked food in retail settings.
• Unitree's own company timeline states that on April 11, 2026 its H1 humanoid "exceeded a top running speed of 10 m/s, once again setting a new world record."

The autonomy requirement is the substantive change and it is easy to miss underneath the spectacle. Last year's games permitted manually operated robots in most events, with full autonomy required only for freestyle gymnastics and dance. A teleoperated robot sprinting 100 meters is a demonstration of actuators and a skilled human pilot. An autonomous one has to handle its own perception, balance recovery and lane-keeping at speed, with nobody to catch a stumble. Those are different engineering problems, and only one of them is on the road to a useful machine.

The venue change does similar work. Beijing's announcement is explicit that scenario events move into "real-world settings such as factories, hotels, and model homes," where robots "must autonomously complete complex, long-duration tasks," and it states the intent directly: the change "aims to transform humanoid robots from mere demonstration tools into effective, real-world helpers with tangible outputs." Anyone who has followed robotics knows why this is the hard version. A test arena has controlled lighting, known floor friction, no unexpected obstacles and a layout the team measured in advance. A hotel corridor has none of those, which is the entire content of the sim-to-real gap. Moving the competition into buildings that were not built for it converts the games from a showcase into something closer to an evaluation.

The task list reinforces that. Folding clothes, extinguishing a fire and preparing cooked food are not chosen for drama. They are chosen because each requires sustained fine manipulation of deformable or unpredictable objects over minutes rather than seconds, which is precisely where current humanoid systems fall apart and where vision-language-action models are being aimed.

The context is a national industrial push rather than a sporting one. Beijing's announcement notes that the city hosts nearly a quarter of China's whole-robot manufacturers and describes humanoid robotics as a priority track, with application as the leading driver of industrial development. The games are effectively a public benchmark run by the state that also happens to be the largest customer for the results, which is worth keeping in mind when reading any outcome.

That brings up the thing this story most needs, which is calibration about the numbers. Humanoid robot performance claims travel further and faster than the evidence behind them, and the primary sources are usually narrower than the headlines. Unitree's H1 product page is a specification sheet -- moving speed of 3.3 meters per second, potential mobility above 5 meters per second, and a note that the product keeps iterating so the shipped version may differ. The 10 meters per second figure appears on the company's own timeline as a company-claimed test result, and Beijing's government separately reported the same figure from a test run ahead of the humanoid half marathon. That is a company claim corroborated by a government report, which is meaningfully better than a press release and meaningfully worse than an independently instrumented record.

It is also, for scale, slightly below the human 100-meter world record pace of roughly 10.4 meters per second averaged across the full distance -- and a robot's top speed in a test is not the same measurement as a human's average over a timed race from blocks. This is the value of the competition format: standing long jump and high jump events at these games explicitly prohibit elastic propulsion and launch devices, and the 100 meters now requires autonomy, which turns loose marketing figures into results produced under published rules that everyone can read. Regulation of the claim, not just of the machine.

The honest caveat is that a preview is not a result. Everything above describes rules, venues and event counts announced in advance; nothing here reports what any robot actually did, because the games have not happened yet. It is also worth noting that the policy conversation around these machines is moving in parallel and not always toward them -- the proposed humanoid robot ban we covered never left committee, but the fact that it was written at all says something about how fast the demonstrations are outrunning the public's read on them.

Primary source, verified: read the paper →

Ground Truth. (2026, August 17). Beijing's robot 100-meter dash goes fully autonomous this month. Ground Truth. https://groundtruth.day/news/beijings-robot-100-meter-dash-goes-fully-autonomous.html

@misc{groundtruth:beijings-robot-100-meter-dash-goes-fully-autonomous,
  title  = {Beijing's robot 100-meter dash goes fully autonomous this month},
  author = {{Ground Truth}},
  year   = {2026},
  month  = {aug},
  url    = {https://groundtruth.day/news/beijings-robot-100-meter-dash-goes-fully-autonomous.html}
}

Topics: robotics · humanoid-robots · embodied-ai · china · benchmarks