Deep Reinforcement Learning for Model Predictive Controller Based on Disturbed Single Rigid Body Model of Biped Robots

This paper modifies the single rigid body (SRB) model, and considers swinging leg as disturbances to centroid acceleration rotational of SRB model. proposes deep reinforcement learning (DRL)-based model predictive control (MPC) resist leg. The DRL predicts swing disturbances, then MPC gives optimal ground reaction forces according predicted disturbances. We use proximal policy optimization (PPO) algorithm among methods since it is a very stable widely applicable algorithm. It an on-policy based on actor–critic framework. simulation results show that improved PPO-based method can accurately predict disturbance, making locomotion more robust.