A 192k-parameter network trained with PPO in MuJoCo MJX drives the Unitree Go2 at 50 Hz. The physics is MuJoCo compiled to WebAssembly, running here at 500 Hz.
Loading…
WASD move · QE turn · Shift sprint · Space push
Trained on flat ground with random pushes, motor latency and friction changes; the same policy file runs on the robot.