China's Tiangong Omni robot wins race with self-evolved face-covering running style

A humanoid robot developed by Beijing's X-Humanoid Innovation Center won first place at a robotics athletics event in Beijing with an unusual running posture — arms raised to cover its face. Engineers confirmed the quirky gait was never manually programmed; it emerged spontaneously through reinforcement learning inside a simulated training environment. The robot's primary training stack uses NVIDIA's Omniverse platform, Isaac Sim for high-fidelity physics simulation, and Isaac Lab for large-scale parallel RL training, topped by the team's own Tien-Kung-Lab algorithm layer. MuJoCo was used as a secondary cross-validation tool to check that the emergent behavior was not an artifact of PhysX physics before deploying the policy to the real robot. AI researcher David Ha noted the achievement on X, comparing it to novel locomotion strategies previously discovered in MuJoCo environments but highlighting that this one transfers successfully to the real world.