Detecting Grasping Positions and Postures in 3D Point Clouds by Geometric Constraints

In this paper, a method of positioning grasping points and grasping posture on strange objects by using geometric constraints is proposed. The input to the algorithm is point clouds of the objects and the geometric parameters of the gripper of the manipulator. The output of the algorithm is a set of grasping points and grasping posture, which can be expected to be the best target for the manipulator to grasp the object. The algorithm first determines the grasping geometric parameters of the gripper, and then sets a series of necessary conditions for grasping the objects successfully. After that, a large number of grasp point samples are carried out on the point cloud and the necessary conditions are used for filtering to obtain a set of potential grasping points. Finally, a kind of weighted calculation method is used to obtain the best grasps. The algorithm does not need to know the objects in advance, and it can get a better grasping effect on some objects with special shape (such as ring). The algorithm is provided as a ROS package for download at https://github.com/ZSM2019/Geometry_Grasp.