Automatic Extraction Method of Machining Trajectories Based on Point Cloud Registration

In the application of industrial robots, automated production is a core goal for improving production efficiency and product quality. However, the extraction of complex spatial machining trajectories has always been a challenge for industrial robots. This paper proposes an automatic extraction method for machining trajectories based on an improved Deep Closest Point (DCP) point cloud registration network and B-spline interpolation. By leveraging the Linked Dynamic Graph Convolutional Neural Network (LDGCNN), the feature extraction module of the DCP network is enhanced, and Transformer and SVD modules are introduced, combined with the ICP point cloud registration algorithm for iterative optimization. Initially, the machining trajectory data points of the template point cloud are manually selected, and then the improved DCP point cloud registration network is used to register the target point cloud with the template point cloud, obtaining a transformation matrix. Using this matrix, the machining trajectory data points on the target point cloud are solved, and these data points are further subjected to B-spline interpolation to obtain precise and smooth machining trajectories. Finally, the degree of overlap between the machining trajectories extracted using this method and the actual machining trajectories is above 99%, and the entire process takes less than 0.2 seconds.