Evolutionary Motion Synthesis for a Modular Robot Using Genetic Algorithm

An evolutionary motion synthesis method using genetic algorithm (GA) is presented for selfreconfigurable modular robot M-TRAN designed to realize various robotic motions and threedimensional structures. The proposed method is characterized by its capacity to derive feasible solutions for complex synthesis problem of M-TRAN through natural genetic representation. For this purpose, the behavior of the robot is described using a motion sequence including both the dynamic motions and configuration changes of the robot. It is a series of segments each of which can specify simultaneous motor actuations and self-reconfiguration by connection/disconnection, starting from a given initial configuration. This simple description can be straightforwardly encoded into genetic representation to which genetic operations can be applied in a natural manner. We adopt traveling distance achieved by the evolved motion as the fitness function of GA. To verify the effectiveness of the proposed method, we have conducted simulations of evolutionary motion synthesis for certain initial configurations. Consequently, we confirm various adaptive motions are acquired according to different initial configurations and fitness functions. We also verify the physical feasibility of the evolved motions through experiments using hardware module M-TRAN II.