Application of Reinforcement Learning to Switched Control of Hybrid Systems

Abstract In this paper, we try to apply learning automata with reinforcement learning (RL) algorithm to realize switched control of hybrid systems. The learning automata were used to optimize the generation probabilities of behaviors in order to achieve a satisfactory result. At first we apply RL to an experiment proposed in other's previous work, then we do some reproduction in order to compare two different methods. Since there are differences in targets, we generalized former RL algorithm. Simulation results demonstrate the feasibility and advantage of applying RL to switched control of hybrid systems.