A potential field method-based extension of the dynamic movement primitive algorithm for imitation learning with obstacle avoidance

Abstract

This paper proposes an extension of the original Dynamic Movement Primitive (DMP) algorithm proposed by S. Schaal to imitation learning for object avoidance in a dynamic environment. A potential field was incorporated into the original DMP algorithm by using a virtual goal position which is calculated using a potential field. A humanoid robot ISAC was trained in simulation to learn how to generate movements similar to the demonstrated movements when an obstacle is placed in the environment. This proposed extension provides robots more robust and flexible movement generation when an obstacle exists. Simulations were performed to verify the effectiveness of the method.