The nuclear reactor coolant pump (NRCP) is the heart of the nuclear power plant. This paper focuses on robot automation grinding processing for NRCP, which includes scanning, point cloud processing, grinding trajectory generation, and quality evaluation system based on reverse engineering. In this work, firstly, the point cloud of NRCP is obtained by robotic scanner system of hand-eye calibration. Secondly, the research proposes a novel method for point cloud simplification, denoising, and boundary extraction base on k neighborhood octree structure. More important, the efficient trajectory generation of grinding relies on transforming point cloud into adaptive triangular mesh. Lastly, quality evaluation system can calculate the deviation between point cloud and qualified workpiece. And the further path is generated according to the deviation. Experiments show that the accuracy of “246” hand-eye calibration method is less than 0.02 mm. The method of point cloud processing has obvious efficiency advantages over other researchers’ algorithms. The final results indicate that the error of grinding is less than 3 mm and efficiency can be improved by 2.5 times compared with manual grinding.