A Reinforcement Learning-Based Trajectory Generation Method for Polishing Complex Surfaces

In the field of mechanical processing, parts grinding is a critical step to enhance the smoothness and precision of part surfaces. However, existing technologies face limitations in terms of informatization and intelligence, which hinder the optimization of grinding efficiency and quality. To address the technical challenges of trajectory generation for complex surface polishing, this paper proposes a reinforcement learning-based method. The approach utilizes a Deep Q-Network to optimize trajectory planning strategies, achieving adaptability and efficiency. Experimental results demonstrate that, compared to traditional manual teaching and offline programming methods, the proposed method significantly enhances dynamic adaptability and trajectory planning accuracy for complex workpieces, especially in large-scale tasks, showcasing superior performance and scalability. The findings highlight that reinforcement learning provides a novel approach for robotic polishing trajectory generation, laying a foundation for the advancement of intelligent manufacturing.