World Humanoid Robot Games: what two Beijing editions show about robot mobility

Beijing has now hosted two World Humanoid Robot Games. The first, in August 2025, drew 280 teams and produced a 100-metre winner in 21.50 seconds. The second, in August 2026, brought more than 2,000 robots and a reported 100-metre time of 8.86 seconds. Together they show fast progress in straight-line running, while braking, balance, autonomy and practical tasks remain clear limits.

For anyone booking a robot for a live event, the games are a useful public test. Unlike a staged product video, results, falls and the rules on remote control were all visible. This article looks at what happened, how the rules shaped the results, and what the events reveal about where humanoid mobility stands today.

The 2025 games: a first public benchmark

The inaugural games ran for three days at the National Speed Skating Oval in Beijing. According to Xinhua's report on the 100-metre final, published on 17 August 2025, 280 teams from 16 countries and regions took part.

The 100-metre race was won by Tiangong Ultra from the Beijing Humanoid Robot Innovation Center, with a result of 21.50 seconds. Second and third places, at 22.08 and 24.53 seconds, went to two Unitree H1 robots entered by other companies. Xinhua noted that the same Tiangong robot had earlier won a humanoid half-marathon in Beijing in April 2025.

Unitree also said it took four golds in the 1,500 metres, 400 metres, 100-metre obstacle race and 4×100-metre relay, a claim reproduced in IEEE Spectrum's Video Friday round-up of 22 August 2025. In the same round-up, PNDbotics stated that its Adam robot was the only full-size humanoid to finish the 100-metre obstacle race. These are the companies' own statements rather than independently compiled results.

Remote control was part of the rules

The most revealing detail of 2025 was not a time but a scoring rule. Xinhua explained that in the 100 metres, a robot run by remote control had its final time multiplied by 1, while a fully autonomous robot had its time multiplied by 0.8. In other words, autonomy earned a 20 per cent time bonus.

Xinhua also reported that Tiangong Ultra was not the first robot to physically cross the finish line. It won because it navigated fully autonomously and therefore received the weighting. The report did not state whether the published times were raw or weighted, or how the second- and third-placed robots were controlled.

Two points follow from this. First, remote control was an accepted way to compete in 2025, and organisers felt the need to reward autonomy explicitly. Second, a robot that looks like it is running on its own may well have a human operator. For event audiences, this is the same question covered in our guide on teleoperated versus autonomous robots.

The 2026 games: more robots, stricter autonomy rules

The second edition opened on 22 August 2026 and ran for five days. A Reuters report carried by Al Jazeera put participation at more than 2,000 robots from 666 teams across 51 events.

The official competition guide for the 2nd games splits these into 30 competition events and 21 scenario-based events, and assigns each an autonomy rule:

• The 100 m, 400 m, 1,500 m and 4×100 m relay, all football formats, floor exercise, dance, wushu, tug of war, pitch-pot and table tennis are listed as fully autonomous only. By our count, that is 20 of the 51 events.
• The 100 m and 400 m obstacle races allow either fully autonomous or remote-control mode.
• In the 21 scenario-based events, such as retail service, packing, hotel guest service, emergency response and cable connection, a fully autonomous run carries a weight coefficient of 1 and a remote-controlled run 0.5.
• Jumps, weightlifting and kickboxing have no autonomy coefficient listed in the guide.

The direction is clear: sprinting and other athletic events moved from "autonomy is rewarded" in 2025 to "autonomy is required" in 2026. Practical tasks still allowed remote control, but at half value. Kickboxing, which tends to dominate social media clips, carried no autonomy rule at all in the guide.

Records, and the problem of stopping

Athletic results rose sharply. CBS News reported that a robot from Beijing-based X-Humanoid ran 100 metres in 9.39 seconds on the opening day and that the same company's robot cleared 2.88 metres in the high jump. Three days later, according to Reuters via Al Jazeera, Tiangong Ultra ran 8.86 seconds in a large-robot semifinal. The Beijing centre also claimed times of 38.15 seconds for 400 metres and 2:21.64 for 1,500 metres.

The headline comparison with 2025 needs care. The 2025 times may have included autonomy weighting, and robots of different sizes ran in separate groups in 2026. The improvement is real, but the exact margin is not a like-for-like figure.

The weakness was stopping. NBC News reported that organisers placed a thick mat several feet beyond the finish line and that robots slammed into it to halt. CBS News described robots having to hit a cushion and fall over to stop, and noted that some competitors tripped, fell, broke apart or caught fire. Reuters reported that after its 8.86-second run, Tiangong Ultra collapsed on impact with the mat and a small fire appeared in its torso.

What the games actually tested

A sprint on a flat, cleared track with a crash mat at the end is a narrow test. It shows that legged robots can now generate high speed in a straight line. It does not show that they can slow down safely, turn in a crowd, or recover from a stumble near people.

The scenario events were closer to real work. NBC News described tasks such as EV charging, restaurant work, emergency response, industrial assembly and cable connection, and called the cable task a test of vision, alignment and force control. NBC did not report how well robots performed in it, and we have not found published results to cite.

Industry voices quoted by NBC made the same point. Robotics investor Xu Qiaosheng said the intensity was higher than in 2025 and put heavy demands on battery management and heat dissipation. Lumos Robotics chief executive Yu Chao argued that running and jumping alone do not make robots more useful.

What the games do not tell you

The events had purpose-built surfaces, controlled spectator areas, technical crews and the option to swap or repair robots between heats. Few corporate events offer those conditions. Many entries were also research or racing builds, not the standard models that suppliers bring to venues.

Results also come mostly from organisers, manufacturers and state media. Independent, event-by-event data on falls, failures and operator involvement has not been published, so the true failure rate is unknown. For a wider look at how staged performances can differ from everyday ability, see our analysis of what China's gala and robot games really showed.

What this means for event organisers

• Speed is not the same as safety. Even record-setting robots needed a crash mat to stop. Plan walking routes, barriers and distances to guests as described in our safety planning guide for public events.
• Ask who is in control. Remote control was openly allowed in 2025 and is still allowed at reduced value in 2026 task events. Ask your supplier which parts of the show are autonomous and which are operated, and put it in your brief using our guide on what to tell a robot supplier.
• Expect falls and plan for them. Falls, breakages and even a fire happened under expert supervision. Agree a fallback in advance, as covered in what happens if the robot fails on the day.
• Heat and battery matter. High-intensity routines strain batteries and cooling. Shorter, scheduled performance slots are more realistic than continuous action.
• Judge by your use case, not by headlines. A sprint record says little about greeting guests on a crowded stand. Ask for footage of the exact model doing the exact task in a comparable venue.

Frequently asked questions

When and where were the World Humanoid Robot Games held?

Both editions took place in Beijing. The first was a three-day event in August 2025 at the National Speed Skating Oval, with 280 teams from 16 countries and regions according to Xinhua. The second opened on 22 August 2026, ran for five days and, according to Reuters, involved more than 2,000 robots from 666 teams across 51 events.

Were the robots at the games remote controlled?

Some were. In the 2025 100 metres, remote-controlled robots were allowed but autonomous runs had their times multiplied by 0.8, giving them an advantage. For 2026, the official competition guide made sprints, football, dance and several other events fully autonomous only, while scenario tasks allowed remote control at half the weighting of an autonomous run.

Did a robot really run 100 metres faster than Usain Bolt?

Reuters reported that Tiangong Ultra from the Beijing Humanoid Robot Innovation Center ran 8.86 seconds in a semifinal at the 2026 games, below the human record of 9.58 seconds. The run was on a prepared track in a large-robot category, and the robot stopped by hitting a mat, then collapsed and a small fire appeared in its torso.

What do the robot games mean for hiring a humanoid robot for an event?

They show that humanoid robots can move quickly in controlled conditions but still struggle to brake, balance and handle practical tasks reliably. For events, this means planning safe distances and barriers, asking suppliers which parts of a show are operated by a person, agreeing a fallback if the robot fails, and judging a model by footage of the task you actually need.

How this article was made: written by the humanoid4rent editorial team with AI assistance; every factual statement was checked against the sources listed below on 2026-10-31. Spotted an error? Write to info@humanoid4rent.com.