Fast and Robust Bin-picking System for Densely Piled Industrial Objects

Objects grasping, that is, bin-picking, is an area that almost all operating robots in the industry are ought to perform in. Although a lot of related approaches in this field have been proposed, due to the poor robustness or high resource consumption of many existing works, grasping densely piled objects still faces some huge challenges. In this work, we develop a bin-picking system for safely and adaptively grasping intensively piled objects. For the challenge of the occlusion scene, the system leverages Improved DBSCAN to first segment the point cloud of the objects, which is improved by the combination of region growing algorithm and Octree algorithm to accelerate calculation. After that, the system utilizes PCA and ICP algorithms for coarse and fine registration respectively. At the same time, we also developed the GRS, which evaluates the grasping risk through the possibility of collision and the stability of the object as well as the entire pile of objects. We also conduct actual experiments on the Anno robot, verifying the speed and robustness of the system.