Stable Trotting of Quadruped Robot with Learning Algorithm

Abstract

In this paper, a learning algorithm is used for stable trotting motion of quadruped robot. The trajectory of the end point leg of 4-legged animals has a pattern of an ellipse. Each of the desired leg trajectories is designed to have an elliptical shape to express trotting motion of quadruped robot. If the 4-legged animals walk, the shape of the end point trajectory of the legs in flight phase is differnt from in stance phase. Because contact force exists in stance phase but doesn’t exist in flight phase. Therefore, animals modify the trajectory of the leg in stance phase to minimize ground reaction force. The end point trajectories of legs of quadruped robot are set to a half elliptical trajectory in flight phase and stance phase. If the leg trajectories are combined, it is one elliptical trajectory. The two half elliptical trajectories are parameterized to be able to be modified the trajectory based on learning algorithm. The quadruped robot is obtained score of the each parameter through learning algorithm and then each parameter is modified by the obtained score. The trajectory is modified by these parameters. The score is inverse of angle measured by a gyro sensor. If the obtained score in current state is lager than previous state, the motion of the quadruped in current state is better. Learning algorithm is evaluated by using RecurDyn dynamics simulator.