Prophet Actor-Critic-Based Deep Reinforcement Learning for Obstacle Avoidance in Robotic Arm Control

Abstract

For the current common obstacle avoidance trajectory planning tasks of robotic arms, the problem of insufficient versatility is usually encountered. The grasping performance of the manipulator is mainly constrained by obstacles. How to improve the obstacle avoidance ability of the manipulator to improve its grasping ability. In order to improve the intelligent control effect of the robotic arm, an intelligent AI control method of the robotic arm combined with deep reinforcement learning is proposed. Moreover, in order to solve the problems of low learning efficiency caused by low-quality empirical data in the early stage of training and low efficiency in obtaining expert empirical data, an EDS mechanism is proposed to improve training efficiency by expanding expert empirical data and adopting unbiased dual memory bank sampling rules. In addition, in order to enhance the obstacle avoidance capability of the robotic arm, a COR system is constructed to help quickly generate the optimal trajectory, and the end effector and fuselage of the robotic arm can simultaneously avoid obstacles in complex environments, and achieve a balance between obstacle avoidance and motion exploration. The results show that the success rate of grasping complex objects in both obstacle free and obstacle free environments can reach more than 80%. Compared with the existing models, it has faster convergence speed and learning effect, and has better model performance. This paper combines experimental analysis to verify the effectiveness of the AI control method proposed in this paper, which can effectively ensure the collision-free trajectory planning of the robotic arm in different scenarios and has strong adaptability to scene changes.