A method for loading and unloading stacked workpieces on a coating line based on the SVD-ICP algorithm

During the production process on the coating line, workpieces are often stacked disorderly, resulting in significant mutual obstruction. This creates considerable challenges for robotic loading and unloading operations. Conventional vision algorithms struggle to meet production requirements for recognition success rates and pose estimation accuracy when faced with weakly textured, reflective workpiece surfaces and complex occlusion environments. To address this issue, this paper proposes a robotic loading and unloading method that integrates Singular Value Decomposition (SVD) and Iterative Closest Point (ICP) algorithms. First, point clouds within the storage bin undergo preprocessing and segmentation. Subsequently, feature matching using Fast Point Feature Histograms (FPFH) and SVD decomposition is employed to compute a coarse pose estimation of the workpiece. This approach overcomes the ICP algorithm’s sensitivity to initial values and its tendency to converge to local optima. Finally, using this coarse pose as the initial estimate, a KD-Tree accelerated ICP algorithm performs fine registration, yielding high-precision six-degree-of-freedom pose estimation. Robotic simulations and physical experiments were conducted. In the simulated environment, the workpiece loading/unloading success rate reached 97%, while the experimental platform achieved an 85.41% success rate. The total runtime of the component recognition and pose estimation algorithms is less than 1.26 s, meeting the real-time requirements of industrial applications. This paper provides a reliable solution for robotic component handling in scenarios such as coating lines.