Dynamic Obstacle Avoidance for Robotic Arms Using Deep Reinforcement Learning with Adaptive Reward Mechanisms

Abstract

To address the challenges of robotic arm path-planning in dynamic environments, this study proposes a reinforcement learning-based dynamic obstacle avoidance method. The study concerns a robot with six rotational degrees of freedom when moving outside of singular configurations, enabling more flexible and precise motion-planning. First, a dynamic exploration guidance mechanism is designed to enhance learning efficiency and reduce ineffective exploration. Second, an adaptive reward function is developed to enable real-time path-planning while avoiding obstacles. A simulation environment is constructed using CoppeliaSim software, and the experiment is performed with two cylindrical obstacles that move randomly within the workspace. The experimental results demonstrate that the improved method significantly outperforms traditional algorithms in terms of convergence speed, reward value, and success rate.