Abstract | ||
---|---|---|
In real application scenarios, objects are usually in complex unstructured environments, so object sorting operations based on machine vision are essential. To solve it, we propose a sort planning method for multi-fingered dexterous hand to handle multiple target objects by using the two-stage object detection network and the sort order decision algorithm. Firstly, the two-stage object detection network based on RGB images is designed, in which the first-stage network is used to search candidate regions where objects may appear while the second-stage network evaluates each region to detect the positions of all target objects in the RGB image. Secondly, the global depth information of the objects and the relative positions between them are obtained from the depth image. Thirdly, the sorting order of the target objects is determined according to the human sorting behaviors, and the grasp planning method is applied to the segmented depth image which contains the chosen target object. Finally, the above steps are repeated until all sorting operations for all visible target objects are finished, and the performance of the sort planning method is verified on the actual robot platform with multi-fingered dexterous hand and the success rate of sorting is up to 82%. |
Year | DOI | Venue |
---|---|---|
2022 | 10.1109/CACRE54574.2022.9834103 | 2022 7th International Conference on Automation, Control and Robotics Engineering (CACRE) |
Keywords | DocType | ISBN |
sorting planning,dexterous hand,deep learning,grasp planning | Conference | 978-1-6654-6669-1 |
Citations | PageRank | References |
0 | 0.34 | 5 |
Authors | ||
8 |
Name | Order | Citations | PageRank |
---|---|---|---|
Xiangyu Dong | 1 | 0 | 0.34 |
Haoyuan He | 2 | 0 | 0.34 |
Peng Du | 3 | 67 | 13.00 |
Xinsheng Tang | 4 | 0 | 0.34 |
Teng Li | 5 | 0 | 0.34 |
Ruijing Yang | 6 | 0 | 0.34 |
Jie Huang | 7 | 0 | 0.34 |
Weiwei Shang | 8 | 77 | 13.89 |