Bridging Reinforcement Learning and Optimal Control via Feasible Action Mapping

This research proposes FAOC, a control framework that combines reinforcement learning (RL) and optimal control (OC) to efficiently solve complex tasks while ensuring physical constraints and recursive feasibility. The method transforms RL actions into feasible parameters for the optimal control problem (OCP), enabling safe and flexible operation. Evaluated on robot table tennis, FAOC demonstrates superior sample efficiency and performance compared to state-of-the-art baselines, with the implementation open-sourced.