Achieving high success rate in dual-arm handover using large number of candidate grasps, handover heuristics, and hierarchical search

Graphical Abstract

Abstract

Many dual-arm handover systems suffer from low success rates since the grasps are limited to a small number of candidate grasps and handover poses. Large number increases success rate but significantly explodes combinatorics and lowers system performance. How to efficiently exploit the large number of grasps and handover poses is a difficulty faced by many dual-arm manipulation researchers. This paper uses handover heuristics and hierarchical search to. challenge the difficulty. For one thing, it samples large number of grasps and handover poses to ensure success handover. For the other, it employs the handover heuristics to reduce combinatorics and the hierarchical search to reduce search space. Leveraging these technique leads to an algorithm that has high computational efficiency and high success rate. We report the algorithm in this paper and examine its performance using both statistical simulation analysis and real-world experiments. Results show that the algorithm can deal with thousands of grasps within a few seconds using a standard PC.

Keywords:

• Handover
• manipulation planning
• dual-arm robots
• heuristic and hierarchical search

Acknowledgements

The authors would like to thank the members of the manipulation research group, intelligent systems research institute, national institute of advanced industrial science and technology (AIST) for their helpful discussions.

Notes

The authors declare that there is no conflict of interests regarding the publication of this paper.

1 Flipping an object with a single arm is not impossible. For example, some promising methods using single-arm regrasp,[1] single-arm regrasp with external fixtures,[2,3] dexterous manipulation [4] and extrinsic dexterity [5] can be found in state-of-the-art literature. Comparing the performance of these solutions with the dual-arm handover in this paper is one of our interest and will be in our future work.

2 Details of the Kawada Nextage robot can be found at http://nextage.kawada.jp/en

3 Not explicitly explained in literature, but probably less than 10 due to the hardware platforms at that time. The original word used was ‘several’.

4 The algorithm is also applicable to multi-arm handover. Once the order of handover is decided, the multi-arm handover problem can be divided into several dual-arm handover problems and solved, respectively.

5 The graph is named regrasp graph rather than handover graph since it is shared by single- and dual-arm cases. The dual-arm handover problem is essentially a dual-arm regrasp problem.

6 Each matrix is one end-effector orientation. The purple vectors are sampled using icosphere. They are pointing to the vertices on a unit icosphere. The green vectors are sampled around the purple vectors.