Planning regrasp operations for a multifingered robotic hand

Abstract

Regrasp operations consist of a sequence of pick-and-place operations, which are very useful to avoid collisions with the environment and to overcome the kinematic limitations in some cases. In this work, we propose an automatic planning system for regrasp operations for a multifingered robotic hand. The stable planes of the object to be grasped are automatically found. The minimal angle allowed to rotate the object without tipping it over is expressed using solid angles to quantify the stability of each stable plane. The grasp for the multifingered hand is generated in the simulation using hand preshapes and approach directions. The stable planes are placed in the simulation as obstacles to plan grasps that do not collide with such stable planes. After a grasp is found, collisions between the grasp and each stable plane are checked. This information and the found grasps computed off-line are saved in a grasp database. During on-line execution, regrasp operations are planned by a breadth-first search using the grasp database. The collision and kinematic feasibility are also considered by the planner.