Speed records and flying sparks at the World Humanoid Robot Games

The second edition of the World Humanoid Robot Games in Beijing delivered a remarkable spectacle. Humanoid robots broke Usain Bolt's human 100-meter sprint record, but slammed straight into the boarding right after the finish line because they couldn't slow down. Some units broke at the hips or spontaneously burst into flames.

The event, organized by the Chinese government and state media, underscores the enormous investment in China's fast-growing robotics sector. Alongside spectacular sprint races, kung fu, and kickboxing, this year the machines also had to perform everyday tasks, such as doing laundry and tidying rooms.

Specialists rather than generalists

The athletic performances show major progress in "whole-body control" — coordinating all joints and limbs at once — according to experts, but the machines are far from multifunctional. The sprinting robots were programmed solely to run straight ahead as fast as possible, with no regard for what happened after the finish line.

One robot from Beijing developer X-Humanoid drew attention by winning the 400 meters with a bizarre posture, arms held up in front of its face. The machine had discovered, through reinforcement learning in simulation, that this was the most efficient way to move its hips.

The limits of practical tasks

While the athletic events went viral, the everyday challenges were arguably more interesting substantively. Robots were given tasks in simulated living rooms, libraries, and hotels. The results showed that robots' processing speed on practical tasks still lags far behind that of humans.

During a complex firefighting and rescue test, robots had to identify hazardous substances, close off shutoff valves, and put out a fire. Only three of the twelve participating teams managed to complete the assignment successfully within the half-hour time limit.

Autonomy remains a challenge

The games also exposed the technology's limits. Many robots still relied on human teleoperation — remote control — while performing physical tasks. Participants received penalty points for human intervention, but the fact that it was needed at all shows how difficult autonomous decision-making in arbitrary situations remains.

On top of that, the required AI models often don't run on the robot's own hardware. Many machines send data over 5G modules to cloud servers to receive instructions. While investors are pouring billions of dollars into the sector and companies like Boston Dynamics and Agility Robotics are preparing commercial rollouts, a truly broadly capable humanoid robot remains, for now, a thing of the future.