Zhiyuan Robotics Wins 18 Golds at World Humanoid Robot Games, Showcasing Factory-Trained Bots
Zhiyuan Robotics topped the medal table at the World Humanoid Robot Games with 18 golds, proving its mass-produced robots can excel in both athletic and operational tasks.
The company's dexterous hands won 7 golds out of 8 specialized events, its operation track took 6 golds out of 12 events, and its competition track secured 5 golds. In one event, a Zhiyuan dexterous hand measured 20 grams of salt in 27 seconds using a small spoon; the Jingling G2 completed firefighting and library organization tasks, while the Yuanzheng A3 executed a 540-degree outward swing kick in tai chi.
All participating robots were standard mass-produced models, not custom-built for the games, according to the report. Zhiyuan leveraged its accumulated experience from real factory deployments, highlighting a shift from robots that can only demonstrate to robots that can complete real work.
Zhiyuan's success is underpinned by its full-stack AI models. The GO-2 VLA foundation model uses an action chain of thought to handle long-horizon tasks, and the GE-2 world model predicts environmental changes for safer trial-and-error learning. The company also used a sim-to-real approach to refine motion control for tai chi routines.
Zhiyuan founder Deng Taihua was quoted as saying that repetitive, physically harmful work should be done by robots, and that the humanoid form offers the best universality for existing infrastructure such as door handles, stairs, and vehicle seats. Zhiyuan has deployed robots in factories, with a reported 99.99% task success rate across 17,625 operations at a Longqi Technology plant in Nanchang.
While the games showcased impressive feats, the real goal is to move robots into everyday work environments. Zhiyuan's “deployment state” solutions aim to go beyond lab trials, enabling multi-unit, sustained operations in productivity scenarios such as production line loading, industrial handling, logistics sorting, and security patrols. The path to widespread adoption still requires solving issues like manual takeover, failure intervals, automatic recovery, and unit economics, but the company believes the turning point is approaching.