SHOES: Sim-to-real Hardware-adaptation via On-policy Expansion of Skills for Legged Robots

SHOES is a sample-efficient framework for adapting locomotion policies to footwear-induced morphology changes in legged robots. It freezes a simulation-trained nominal policy and fine-tunes lightweight modulation modules on the robot, while introducing SafeBox for constrained online adaptation. Simulation and real-world experiments on a Unitree Go2 demonstrate significant performance recovery after morphology changes.