The hardware behind the run is new: compact motors strong enough to hold a humanoid balanced at speed. Its limits showed at the same games, where robots burned out motors trying to stop and several runners never got going at all.
At the second World Humanoid Robot Games in Beijing, a humanoid called Tiangong Ultra covered 100 meters in 9.39 seconds, faster than Usain Bolt's world record of 9.58. The time came from CCTV, the state broadcaster that supplied timing and results for the sprints and jumps at the games; a press release from its maker, X-Humanoid, claimed a slightly quicker 9.32.
It was not the first machine at the games to manage it: Lightning, built by the phone maker Honor, had run the distance in 9.47 seconds a few days earlier.
The same model raced again the next day, covering 400 meters in 38.15 seconds and cutting almost five seconds off the previous record at that distance. It won the long jump too, at 7.97 meters.
The speeds rest on two recent pieces of engineering: compact motors with high power output, which can hold a humanoid balanced while it moves quickly, and control algorithms trained in simulation. That training is likely why several of the robots run in a way that plainly does not look human — the algorithm explores until it finds the best way to move rather than copying the gait people evolved.
At the games the hardware was being pushed to its physical limits. Robots struggled to stop, sometimes blowing out their motors in the attempt, and several of the runners either never started or moved at no more than a walking pace.
Speed is the simpler problem in any case. Several contests added this year would have given a human competitor no trouble at all, yet they were designed to probe the machines' onboard intelligence: fitting cables into sockets, dropping a piece of rubbish into a bin in a staged bedroom. A roboticist at Brown University told Wired that handling objects inside a complicated setting is a far harder problem than running fast or jumping — an example of Moravec's paradox, where a backflip looks more impressive than plugging in a cable and is the easier of the two for a robot to do.
In some photographs and footage from the games, people appear to be steering machines by remote control, an indication that those robots could not carry out the tasks on their own.
One entry stood apart from that remote driving: Galbot's robot played tennis against a human opponent with no one at the controls, a task that demands advanced perception and high-speed reflexes.