Reorientation planning for a multifingered hand based on an orientation states network using regrasp primitives

摘要

This paper concerns the reorientative manipulation planning for a grasped object by a multifingered robotic hand. This manipulation is characterized as the one in which the grasped object is reoriented by the hand without being released, in order to achieve a certain desired final grasping orientation. Two primitives (rotation and pivoting) effectively perform this task by sequentially executing them. A planner which generates a sequence of the two primitives is presented. Two salient features of the planner are that it plans by considering the geometrical shape of the object to be manipulated and that based on a proposed states network it works in two phases. In its first off-line phase, executed once only for the given object, it constructs a structured states network composed of all those orientation states of the object at which at least one of the rotation or pivoting primitives is feasible. This states network serves as road-map for sequence planning. The second on-line phase of the planner can be repeatedly provoked to find the sequenced route on the network between any two given states—corresponding to the initial and goal orientation of the grasped object. A program has been developed for the proposed planner.