Hundreds of humanoid robots compete on real workplace tasks at Beijing's World Humanoid Robot Games
Beijing's second World Humanoid Robot Games drew 666 teams and over two thousand robots from sixteen countries to compete in workplace-style tasks like hotel bed-making and a new tweezer-dexterity test, per Xinhua and Interesting Engineering. Every scenario event now scores fully autonomous performance twice as high as remote-controlled runs. Unitree founder Wang Xingxing says robots still struggle once real-world conditions change.
Beijing's second World Humanoid Robot Games ran August 22 through 26, 2026, drawing 666 teams and 2,056 robots from 16 countries across six continents, a 138 percent increase in team participation from the inaugural edition. The event expanded to 51 competition items (up from 26) and 1,301 matches (up from 487), with 21 new scenario-based contests built around real workplace duties: hotel bed-making, luggage handling, supermarket restocking, and library book sorting. A new Dexterous Hand Challenge tested fine motor control directly, with robots weighing powder, opening bottle caps, connecting cables, and picking up beans with tweezers. Every scenario event now scores fully autonomous performance at a 1.0 coefficient versus 0.5 for remote-controlled runs. Unitree founder Wang Xingxing said limited generalization is the industry's biggest bottleneck: robots trained on enough data can approach 100 percent success in controlled settings, but performance falls sharply once the object or environment changes. He predicted a so-called 'ChatGPT moment' for embodied AI, where a robot could complete about 80 percent of everyday tasks in an unfamiliar setting through simple voice or text instructions, could arrive within two to three years, or five to ten years at the latest, per Xinhua. Organizers said strong competition results are meant to translate directly into commercial orders. Sources: Xinhua (Aug 23, 2026), Interesting Engineering (Aug 14, 2026).