2020/06/22 by Michael Hegedus, Hegedus, Michael, Kamal Gupta +3
Engineering · Computer Science · #Robot Manipulation and Learning #Robotics and Sensor-Based Localization #Robotic Path Planning Algorithms
paper · pdf · doi:10.48550/arxiv.2006.12676
We present a generalized grasping algorithm that uses point clouds (i.e. a\ngroup of points and their respective surface normals) to discover grasp pose\nsolutions for multiple grasp types, executed by a mechanical gripper, in near\nreal-time. The algorithm introduces two ideas: 1) a histogram of finger contact\nnormals is used to represent a grasp 'shape' to guide a gripper orientation\nsearch in a histogram of object(s) surface normals, and 2) voxel grid\nrepresentations of gripper and object(s) are cross-correlated to match finger\ncontact points, i.e. grasp 'size', to discover a grasp pose. Constraints, such\nas collisions with neighbouring objects, are optionally incorporated in the\ncross-correlation computation. We show via simulations and experiments that 1)\ngrasp poses for three grasp types can be found in near real-time, 2) grasp pose\nsolutions are consistent with respect to voxel resolution changes for both\npartial and complete point cloud scans, and 3) a planned grasp is executed with\na mechanical gripper.\n