Dynamic Path Planning of Robotic Arm Based on Repulsive Potential Field and Deep Reinforcement Learning

To endow robotic arms with enhanced autonomous obstacle avoidance and path planning capabilities in complex dynamic environments, deep reinforcement learning (DRL) methods have been widely employed for robotic arm path planning tasks in recent years. However, existing DRL-based approaches suffer from high state space dimension and limited environment awareness, which leads to insufficient path planning performance and generalization performance of the algorithms in dynamic environments. In this paper, we propose an environment state representation method based on repulsive potential field (RPF) and incorporate it with proximal policy optimization (PPO) to develop a novel path planning algorithm, RPF-PPO. The algorithm takes the end of the robotic arm as the reference point and uses the RPF to model the obstacles uniformly, forming a fixed-dimension state representation, which effectively improves the algorithm's path planning performance and generalization ability in dynamic environments. Experimental results demonstrate that the proposed RPF-PPO algorithm achieves higher path planning success rates and superior generalization ability compared to existing methods, thereby validating the effectiveness of the proposed approach.