Humanoid robot games in Beijing: 500+ robots compete, but most still clumsy and far from useful

Summary

Over 500 humanoid robots competed in Beijing this weekend. Most remain clumsy and inefficient at basic tasks. The gap between what robots can do in controlled settings and what they can do reliably in the real world spans years, not months.

Robots at the competition attempted basic real-world work: moving boxes, delivering luggage, cleaning rooms. One robot spent more than 17 minutes throwing away nine pieces of trash in a mock hotel room. In a pharmacy simulation, another spent nearly five minutes picking up three boxes of medicine. A third took about two minutes placing two containers on shelves. Tasks trivial for humans remain hard for robots. The core problem is force sensitivity. Picking up and folding a newspaper without tearing it, pulling a phone from a pocket without damage, or assembling food on a plate without crushing delicate items requires precisely calibrated force that robots cannot yet execute reliably.

A robot trained by the Beijing Institute for General Artificial Intelligence in collaboration with Unitree Robot, a leading Chinese robot maker, competed in a hotel reception scenario. It used a three-fingered hand to drag a suitcase to a designated door. The robot moved in small stomping motions and froze at times. A researcher at the event said: "Many of our humanoid robot algorithms are still in the lab demo stage."

China declared it wants to lead the world in humanoid robots by 2027. By that metric, China has arguably already won. The country is running the only large-scale public robotics competition of this kind, and the competitors and infrastructure signal serious state and commercial investment. The U.S. is already discussing how to prevent China from doing to humanoids what it did to drones: flood the market with cheap, capable units that crowd out American competitors. Yet the low technical bar in Beijing suggests the risk is overblown. If robots are still struggling with basic dexterity and force control, the path to mass-market deployment and price compression is longer than geopolitical framings suggest.

Figure AI founder Brett Adcock has been testing market appetite for more capable humanoids. When AI researcher Noam Brown at OpenAI challenged a Figure robot video by asking what would happen if the standing desk were raised six inches, Adcock re-ran the task with the table elevated. The robot adapted in real time, satisfying Brown's test. Whether the adaptation was scripted, teleoperated, or genuinely autonomous remains unclear. The response was enough to signal credibility to a skeptical observer. That kind of real-time task adaptation is precisely what the Beijing competitors could not reliably demonstrate.