The headlines from Beijing are about faster robots and broken records. Hidden inside the competition rules, however, is a more difficult question: when autonomy affects the result, how do we know where the decision really came from?

The scale of the second World Humanoid Robot Games is difficult to dismiss as a technological sideshow. From 22 to 26 August, 2,056 robots from 666 teams are competing across 51 events and 1,301 competition sessions at Beijing’s National Speed Skating Oval, according to the official Beijing Government overview. A separate China Daily report citing the organising committee says participants have come from 16 countries across six continents. The technological progress has been equally striking: in 2025, Tiangong Ultra completed the 100 metres in 21.50 seconds, while this year the winning robot ran 9.39 seconds, faster than Usain Bolt’s human world record of 9.58 seconds. Reuters also reports that more than 40 per cent of this year’s tests require full autonomy, reflecting a deliberate shift from spectacular demonstrations towards machines capable of acting independently in practical environments. Yet behind the records and rapid technological progress, Beijing is confronting a less visible problem: if autonomy itself becomes part of competitive performance, how can anyone be certain that the machine really acted alone?

The rulebook behind the spectacle

Days before the Games opened, organisers briefed teams on a document with an unusually revealing title: 《世界人形机器人运动会赛风赛纪管理办法》 — the World Humanoid Robot Games Competition Conduct and Discipline Management Measures. The official World Humanoid Robot Games document can be accessed here. At the teams’ meeting, organisers described a framework covering technical misconduct as well as behaviour around the competition, built around five principles that included prevention, full-process supervision and, most intriguingly, 技术溯源 — technical traceability. A separate fair-competition briefing required teams to rely on their own programmes and hardware and prohibited off-site remote control and human intervention. The Chinese account of the organisers’ 19 August briefing sets out these requirements in detail.

That language matters because autonomy is no longer simply an engineering preference at Beijing 2026. It now has competitive value. According to the official Beijing government briefing on the competition rules, with two obstacle-race exceptions, track events and other competitive events must be completed fully autonomously. In the scenario competitions, a fully autonomous performance carries a scoring coefficient of 1.0, while remote-controlled performance receives only 0.5. The official competition rules go further in individual events: intervention in a supposedly autonomous attempt can cause the performance to be rescored under the remote-control coefficient.

That changes the nature of the integrity problem. Once autonomy affects the score, it stops being merely a property of the technology and becomes a claim made through the competition result. Two robots may appear to complete the same task successfully, but the result deliberately assigns greater value to one if the machine is judged to have done so autonomously. The competition is therefore measuring something that cannot always be established simply by watching what happens in the arena.

When autonomy becomes a claim

A referee can see whether a robot falls, misses an object or fails to complete a task. The origin of its decisions is less visible. A machine may be physically present inside the competition area while parts of perception, inference or planning occur elsewhere in a wider computational system. Human involvement need not look like someone steering the robot with a joystick; the harder cases may concern subtler forms of assistance, external processing or high-level intervention that leave the visible performance unchanged.

Beijing has already developed a sophisticated approach to a related problem: identifying and tracing the machine itself. The organisers’ “one machine, one code” system gives robots an electronic identity linked to team information and equipment parameters, while the wider management platform can record operational data across the competition environment. In practical terms, organisers are increasingly able to establish which machine belongs to which team and which equipment has entered the controlled competition system.

But identifying the robot is not necessarily the same as identifying the source of the decision attributed to it. A useful way of describing that distinction is decision provenance: the ability, when necessary, to establish where the decisive instructions behind an autonomous performance originated and whether they were produced under the conditions allowed by the rules. It does not require making proprietary algorithms public or demanding that a referee understand every calculation inside an AI model. It requires something more limited but fundamental: if autonomy contributes to the result, there must eventually be a credible way of verifying the autonomy being rewarded.

From competition to evidence

This is where Beijing 2026 differs from the problem IPSEI examined in July in When robots enter the arena: the next frontier of sport integrity. That article considered what could happen when software, algorithms and digital infrastructure become part of the competitive system itself, including the implications for inspection, cybersecurity, digital evidence and accountability. The World Humanoid Robot Games now expose a more specific problem: a real competition has begun assigning competitive value to a technological quality whose origin may itself need to be proved.

The significance becomes clearer when the Games are viewed alongside Beijing’s wider robotics infrastructure. In April, the city launched a Humanoid Robot Training and Evaluation Base at the National Speed Skating Oval, designed to combine training, evaluation, pilot testing, certification, competition support and industrial services. At the same event, China’s Software Testing Center and leading universities introduced ET-Benchmark, intended to provide a common measurement framework for embodied-intelligence systems at a time when, as the organisers themselves acknowledged, the industry lacks unified evaluation standards.

This brings two functions unusually close together. Sport traditionally asks which competitor performed best under agreed conditions. Technical evaluation asks whether a system possesses the capability it claims to possess. The World Humanoid Robot Games increasingly do both: they produce winners, but their real-world tasks also expose whether technologies can perform reliably outside controlled laboratory demonstrations. The competition is therefore not merely displaying technical progress; it is beginning to produce evidence about it.

When medals become market signals

Chinese officials make that ambition explicit. Jiang Guangzhi, director of the Beijing Municipal Bureau of Economy and Information Technology, has said that rules developed through the competition are expected to evolve into practical technical standards, helping move robotics technology towards industrial application. In a phrase that captures the philosophy behind the Games, he said robots mastering these “last-metre” skills could “turn medals into orders” and move directly from the competition arena to real-world workplaces. People’s Daily Online reports the connection between competition rules, technical standards and industrial deployment here.

There is nothing inherently problematic about this model, and it would be wrong to describe the Games themselves as a certification system. Competition can be a powerful way to expose weaknesses, accelerate innovation and establish common benchmarks. The integrity question arises because the more economic meaning a result acquires, the more important it becomes to know exactly what that result proves. If “fully autonomous” is both a scoring category and, increasingly, a signal of technological maturity, confidence in the result depends not only on whether the robot completed the task, but on whether the capability being rewarded was genuinely present.

This may be the most important governance development hidden inside the spectacle in Beijing. The organisers are already creating rules against technical misconduct, distinguishing autonomous from remotely controlled performance, tracing individual machines and linking competition to a broader system of evaluation and industrial standards. The next challenge follows logically from the architecture they themselves are building: to move from tracing the machine to establishing the provenance of the performance it produces.

For human sport, a result usually tells us what happened. In autonomous competition, a credible result may increasingly have to tell us something more: where the performance came from. When medals begin to carry information about technological capability beyond the arena, integrity no longer protects only the fairness of the contest. It also protects the meaning of the claim that travels with the result — which is why the most difficult question emerging from Beijing may not be how fast the robots can run, but who, or what, really made the decision.