Author Affiliations
School of Internet of Things Engineering, Jiangnan University, Wuxi, Jiangsu 214122, Chinashow less
Fig. 1. Flow chart of pose estimation algorithm based on point pair features
Fig. 2. Diagram of point pair feature
Fig. 3. Diagram of building a Hash table
Fig. 4. Diagram of local coordinates
Fig. 5. Diagram of voting process
Fig. 6. Diagram of consistency adjustment of point cloud normal direction. (a) Original point clouds; (b) point cloud normal direction before adjustment; (c) point cloud normal direction after adjustment
Fig. 7. Diagram of gripping pose strategy adjustment. (a) Objects to be gripped; (b) top 5 proposed poses; (c) output poses
Fig. 8. Diagram of adjustment of angle deviation caused by rotational symmetry
Fig. 9. Diagrams of gripping system. (a) Simulation environment; (b) real environment
Fig. 10. Diagrams of pose estimation. (a) Simulation environment; (b) real environment
Fig. 11. Line charts of pose estimation results. (a) Simulation environment; (b) real environment
Statistical results | Simulation environment | Real environment |
---|
Total number of experiments | 240 | 240 | Number of successes | 233 | 182 | Number of failures | 7 | 58 | Success rate /% | 97.1 | 75.8 | First reason for failure is that range of robot is not enough | 7 | 12 | Second reason for failure is that suction of gripper is not enough | 0 | 46 |
|
Table 1. Statistics of gripping success rate in experiments
Statistical results | Original algorithm | My algorithm |
---|
Number of successes | 17 | 20 | Number of failures | 3 | 0 | Success rate /% | 85 | 100 | Fitness /% | 85.79 | 98.19 | Inlier RMSE (root-mean-square error) /mm | 0.00068 | 0.00032 |
|
Table 2. Statistics of point cloud matching results of simulation data