Article

Cubic Doggo reinforcement learning standing stability training update

Cubic Doggo used Gymnasium and SB3's PPO in MuJoCo to train standing while minimizing pitch and roll, and deployed the ONNX file back to MuJoCo, raising resistance to horizontal push from 3.5N to 4N. The second phase added "push recovery training," using horizontal force as domain randomization under a curriculum schedule, but after two weeks of training the improvement was limited. Next steps are balancing on fixed and varying slopes, then deployment to Gazebo and a physical robot.

Part of the event: Cubic Doggo push-resistance training improves only 0.5N in two weeks

This page provides a summary and source link only.