Humanoid Robot Breaks Usain Bolt's 100-Meter Dash Record at Beijing Games

A humanoid robot called Tiangong Ultra ran a 100-meter dash in 9.39 seconds during a preliminary heat at the second annual World Humanoid Robot Games in Beijing, breaking the 9.58-second human world record set by Usain Bolt in 2009. Engadget

The Tiangong Ultra, developed by the Beijing Humanoid Robot Innovation Center, overtook Honor's Lightning robot, which had finished the same event in 9.47 seconds. Chinese media reported a slightly faster time of 9.32 seconds in a practice race for the same model. DW The World Humanoid Robot Games, which kicked off August 22 and runs through August 26 in Beijing, features thousands of robot participants competing across track and field, soccer, table tennis, and boxing. Engadget

The pace of improvement is hard to ignore. At the inaugural Games the previous year, the Tiangong Ultra won the 100-meter dash in 21.5 seconds. A 12-second reduction in one year, in a sport where human records shift by hundredths of a second, is a sharp departure from the incremental progress familiar to anyone who tracks athletic performance. At that same inaugural event, Unitree posted a 33.71-second time in the 100-meter sprint. Engadget

The sprint record is one data point in a broader pattern of accelerating movement in humanoid robots. Reuters reported that another robot of the same model as the 100-meter champion ran 400 meters in 39.7 seconds, faster than Wayde van Niekerk's world record of 43.03 seconds. Reuters In a separate half-marathon for robots in Beijing, AP News reported a winning finish time of 2 hours, 40 minutes, and 42 seconds. AP News Beijing also hosted a half-marathon for Chinese companies to showcase their humanoid robots, where Honor's Lightning robot won gold with a finish time of 50 minutes and 26 seconds. Engadget

These results are not entirely without precedent in robotics. A soft cheetah robot with a flexible spine, developed at NC State, completed a 100-meter sprint in 5.95 seconds in 2020, though that platform was a quadrupedal (four-legged) design rather than a bipedal (two-legged) humanoid. Engadget The distinction matters: a two-legged robot maintaining dynamic balance — keeping itself upright while in motion — at sprint speed is a fundamentally different control problem from a four-legged robot with a spine that bends and absorbs impact.

The broader context here is not that robots have become faster than humans. That threshold was crossed years ago in non-humanoid form factors. What is notable is that bipedal humanoid platforms, constrained by the same balance and energy-efficiency challenges that govern human running, are now crossing it too, and doing so with a year-over-year performance curve that bears no resemblance to the plateau governing human athletics.

Reuters raised the question of whether these robots can perform tasks as basic as plugging in a cable, and the gap between locomotion (the ability to move from place to place) and manipulation (the ability to use hands or grippers to interact with objects) is real. Sprinting in a straight line on a prepared surface tests actuator power, gait control, and dynamic stability. It does not test perception, dexterous manipulation, adaptive planning, or the ability to operate in unstructured environments. The same robot that runs 100 meters in 9.39 seconds may struggle with tasks a toddler handles without thought.

Still, the locomotion results matter for practical deployment. A humanoid robot that can move at human-sprint speed has a larger operating range for applications that require traversing distances quickly, whether in logistics, emergency response, or last-mile delivery scenarios. The gait-control advances underlying these sprint times, particularly in dynamic balance recovery and energy-efficient bipedal locomotion, transfer directly to real-world terrain navigation where speed is less important than stability.

The competitive dimension is also worth noting. The World Humanoid Robot Games functions as a public benchmarking exercise for Chinese robotics companies, and the presence of multiple entrants, including Tiangong Ultra, Honor's Lightning, and Unitree, creates an incentive structure that pushes performance forward. The fact that the same company's robot competed in both editions of the Games provides a longitudinal data point on the rate of improvement, and the trajectory from 21.5 seconds to 9.39 seconds in twelve months is steep by any reasonable measure. Engadget

The natural question is whether that curve continues. Linear extrapolation from two data points is not projection; it is curve-fitting. The early gains in any performance domain are typically the easiest, as straightforward engineering problems, from actuator sizing to basic gait stabilization, yield to direct optimization. The next phase, where marginal gains require solving harder problems in materials, energy density, and control sophistication, will likely be slower. That said, the field has clearly entered a phase where the engineering questions are about optimization rather than feasibility, and that is a meaningful transition for any technology.