Two thousand humanoid robots hold their own Olympics

The second World Humanoid Robot Games opened in Beijing on 22 August, at the National Speed Skating Oval, the venue built for the 2022 Winter Olympics, and the numbers say as much as the events: 2,056 robots from 666 teams representing sixteen countries, competing in 51 disciplines over five days, with the number of entries up 138 per cent from the first edition and the robot count quadrupled. What began last year as a curiosity has become, in one edition, the standard annual benchmark for embodied artificial intelligence.

The opening programme picked the two events that make the point best. The 100-metre sprint and the standing high jump, both already beyond what human athletes have achieved, showed exactly what has improved in twelve months: locomotion, balance and the coordination that makes a robot run without falling over. A machine winning a footrace against the human record is not a sporting upset, it is a unit of measurement, and the gap between this year’s results and last year’s is the real headline, because it quantifies how much faster humanoid locomotion is progressing than the industry’s own projections.

The rest of the programme is designed to test the harder skills. Alongside the track events are scene-based competitions drawn from daily life: folding laundry, putting out a fire, cooking food in a retail setting, sorting items at home. These are the tasks that matter commercially, because they require perception, manipulation and planning together, the combination that separates a robot that can run from a robot that can work. The organisers’ choice to weight the programme toward household and emergency tasks is a statement about where the industry thinks the value is, and it is the same statement every robotics company has been making to investors: the prize is not the Olympics, it is the housework.

The ceremony around the competition matters more than a novelty would suggest. A thousand robots performed in the opening show, thirty-six played instruments alongside human musicians, and eighty held formation changes on the floor, demonstrations of the fleet-level coordination that is the least discussed and most commercially relevant capability in humanoid robotics. The event also imported the Olympic tradition of a torch relay, with a robot carrying a ‘flame’ on a journey that began at the Milan Winter Games and included picking tea on a Fujian hillside and paddling a dragon boat, a deliberate piece of theatre linking robotics to the world of sport.

The international participation is the part that should be watched. Sixteen countries entered, from the United States, Germany and Japan to Brazil, and the competition is organized with global robot bodies and broadcast in dozens of languages. China’s domestic robotics industry has made humanoid development a national priority, and the games function partly as a showcase for that ecosystem, but the format has been designed as an international one, which is how it will be judged in the years ahead. A world championship that only one country takes seriously is a trade show; one that others enter and win is a field.

The honest measure of the games is not the spectacle but the rate of change. The robots that ran in Beijing were, by the industry’s own standards, crude in ways their makers freely admit, with power budgets, failure rates and manipulation limits that would not pass a workplace safety review. What the sprint times and the high jumps demonstrate is that the rate of improvement is steep, and that the gap between demonstration and deployment is closing along a curve that the previous edition made visible. Two thousand robots competing in fifty-one events is a strange way to track a technology, but it may be the most honest one available.