Path planning has always been a key challenge in the field of robotic arm task execution and is a prerequisite for robotic arms to successfully complete specified tasks. This paper begins with the spatial pose and kinematic model of the robotic arm represented by DH, and solves the kinematic model of the robotic arm through forward and inverse kinematics. Starting from Cartesian space trajectory planning, the paper constructs a path optimization model for robotic arm task execution with the objective of minimizing the path execution time, while using kinematic metrics as constraints. Based on the DPPO algorithm in reinforcement learning, the paper introduces the CMA-ES mechanism to construct the DPPO-CMA algorithm and designs corresponding state-action and reward functions. Research shows that the average path length of the DPPO-CMA algorithm is 581.58 mm, which is 158.18 mm shorter than the average path length of the P-RRT* algorithm. The path search time decreases from the average of 163.25 seconds in the P-RRT* algorithm to 29.16 seconds. Additionally, the dynamic response results of the reward value are higher in this algorithm, and the task execution path planning results of the robotic arm exhibit higher stability and positioning accuracy. Reinforcement learning can better learn the task execution status of the robotic arm, thereby improving its efficiency during task execution and ensuring industrial production efficiency.
