MARCH: Model-Assisted Reinforcement Learning for the Perceptive Control of Humanoids over Sparse Footholds
The framework addresses difficult bipedal locomotion over sparse terrain by combining model-based planning with reinforcement learning. It generates a safe reference trajectory using simplified models, trains a teacher policy guided by safety constraints (Control Lyapunov Function), and then distills this teacher into an efficient, vision-based student policy. This results in physically grounded, sample-efficient locomotion on hardware.