To address the problems of insufficient dynamic obstacle avoidance capability of robotic arm control based on deep rein-forcement learning,and the multi-task conflict that exists in the operation process,a multi-modal hybrid control method based on dual-role and regularized critic algorithm(DARC)was proposed.The task was decomposed into multiple obstacle avoidance modes,and the reward function was designed by using the repulsive-attractive idea of artificial potential field method and was trained separately.The multiple modes that had been preliminarily trained were switched by distance threshold or reward accu-mulation threshold to eliminate the conflict existing in hybrid control.A planar simulation model of multi-link robotic arm with soft body characteristics was designed by combining the kinematic characteristics of robotic arm unit.The experiment verifies that the proposed method can effectively improve the dynamic obstacle avoidance capability of robotic arm and avoid the conflict between multiple tasks.
维普
万方数据