Unknown object grasping using force balance exploration on a partial point cloud

Reducing the computing time for unknown object grasping while maintaining grasp stability is the goal of this paper. Inspired by the camera sensor distribution of the PR2 and Baxter robots, as well as active exploration for unknown object grasping, a novel unknown object grasping algorithm is proposed. This algorithm is based on two 3D sensors distributed like the PR2 and Baxter robots. Using the inputs from the two 3D sensors, a partial point cloud is constructed. Series of virtual viewpoints are allocated at intervals surround the principal component axis to build interval virtual object coordinate systems, from which force balance computation is carried out. The force balance is examined both in the XOY plane and the XOZ plane to guarantee the grasping stability. The hand configuration with the best force balance is returned as the final grasp configuration. Simulations based on a Universal robot arm and a Lacquey fetch gripper demonstrated favorable performance. Our algorithm can quickly process the partial point cloud and output the final grasp within 1 or 2 seconds (varying according to the point sets). The simulations demonstrated the effectiveness of our grasping algorithm.