Enhancing Robotic Arm Path Planning Efficiency Through a Phased Reward Configuration Mechanism

Abstract

This paper addresses the challenges of inefficiency and randomness in deep reinforcement learning (DRL) for robotic arm path planning and obstacle avoidance. To tackle these issues, we propose a Phased Reward Configuration Mechanism (PRCM), which divides the learning process into distinct phases. Each phase employs a customized reward function designed to optimize the robotic arm's movements in various operational regions, such as fast approach, stationary approach, and obstacle regions. By dynamically adjusting the reward based on the arm's position and task progress, PRCM guides more efficient exploration and improves path planning performance. We integrate this approach with several DRL algorithms and conduct simulation experiments to evaluate its effectiveness. The results show that PRCM significantly enhances the stability and efficiency of the training process, improving both the quality of path planning and the arm's ability to work in complex environments.