Fig. 1. 3D model diagrams of the workpiece. (a) Top of workpiece; (b) bottom of workpiece

Fig. 2. Experimental platform scene point cloud

Fig. 3. Algorithm flowchart in this article

Fig. 4. Point cloud preprocessing process diagrams. (a) Downsampling; (b) remove platform point clouds; (c) remove container point cloud

Fig. 5. Feature map to be selected after regional growth segmentation

Fig. 6. Selection result of candidate faces

Fig. 7. Point cloud block 1 plane expansion process. (a) Minimum bounding rectangle in space; (b) found critical point cloud; (c) expansion result

Fig. 8. Point cloud block 2 plane expansion process. (a) Found critical point cloud; (b) expansion result

Fig. 9. Feature surface restoration result

Fig. 10. Prepositioning result of stacked workpieces

Fig. 11. Improved ICP algorithm flowchart

Fig. 12. Error curves of different loss parameters

Fig. 13. Template point cloud

Fig. 14. Source point cloud

Fig. 15. Registration effects of different algorithms on self-made source point clouds. (a) ICP; (b) T+ICP; (c) 3D-ISS+T+ICP; (d) proposed algorithm

Fig. 16. Workpiece 3 source point cloud

Fig. 17. Initial transformation effect

Fig. 18. Registration effects of different algorithms on actual point clouds. (a) ICP; (b) T+ICP; (c) 3D-ISS+T+ICP; (d) proposed algorithm

Fig. 19. Scenarios under different working conditions. (a) Case 1; (b) case 2; (c) case 3

Fig. 20. Positioning effect of case 1

Fig. 21. Positioning effect of case 2. (a) Prepositioning; (b) precision positioning

Fig. 22. Positioning effect of case 3. (a) Prepositioning; (b) precision positioning

Fig. 23. Different workpiece positioning results. (a) Rectangular concave convex parts; (b) circular concave convex parts; (c) convex components

Fig. 24. Robot grasping experimental platform

Fig. 25. Diagram of robot grasping

Fig. 26. Robot grasping process. (a) Grabbing; (b) after grabbing