Abstract: A fault self-repair method based on reinforcement learning algorithm is proposed for the robust design issue of lattice modular robot. The method transforms the self-repair process with the goal of filling empty spaces into a self-reconfiguration process by means of the movements of meta-modules containing empty nodes. Based on the reinforcement learning algorithm, a discrete path planning for the movement of empty nodes is proposed, and the empty nodes are accordingly guided to travel through the system. The simulation test results show that the effectiveness of this fault self-repair method is verified on the lattice modular robotic system and can be widely applied to other isomorphic modular robotic systems

Key words: modular robot, self-reconfiguration, reinforcement learning algorithm, fault self-repair