Research on Deep Learning-Based Path Planning Algorithm for 3D Polishing Robots

This paper proposes a novel path planning method for 3D polishing robots based on deep learning. Firstly, the 3D polishing environment is modeled and represented as a voxelized three-dimensional grid structure. Then, the path planning problem is formalized as a constrained optimization problem. Subsequently, an end-to-end deep neural network model is designed to directly learn the optimal path planning strategy that satisfies various constraints from 3D environment data. The proposed deep learning method is evaluated by comparing it with classical RRT and BBT-RRT algorithms on a test set containing 10 complex 3D polishing environments. Experimental results demonstrate that the proposed method not only generates collision-free, fully covered, and smooth paths of high quality but also achieves shorter path lengths and higher computational efficiency, especially in highly complex environments, enabling millisecond-level online path planning.