Sasha

Learned whole-body policy (PPO) in a MuJoCo Warp replica of HIM's sim, with HIM's ball physics and opponent. HIM's grade, 3 seeds: 0.989 legal returns per receive (HIM's reference policy: 0.966), 0.0045 falls per 10 s. Video: the longest of 888 rallies (28.7 s, 20 returns). Simulation only; not hardware-validated.

Sasha·Unitree G1