Model Adaptive Gait Scheme Based on Evolutionary Algorithm

To let a robot learns to create behaviors without recourse to a certain model, or an action which is designed manually is always a problem. This paper proposes a novel method of autonomous locomotion scheme for legged robot. This method using biological evolutionary algorithm makes robots more robust for their walking. The evolutionary algorithm and a simulation environment is applied to find some actions, with which the robot could walk on. Moreover, the robot knows nothing about how to walk at the beginning and its model can change. Simulation and physical experiments are conducted based on the multi-legs robot platform. Finally the robot learnt to walk and the experimental result validates the algorithm.