In this paper, we address global motion planning aspects of dextrous manipulation of 3D objects by a robotic multi-fingered hand. We focus on an important sub-class of the general problem stated by Li, Canny, and Sastry in a paper appeared in ICRA'89, in which only the final object configuration is specified (as opposed to the complete final grasp in the general problem). Our aim is to move (manipulate) the object to the desired configuration through forces exerted by rolling and sliding contacts at the fingertips. We call this the re-configuration problem. A simple statement of the re-configuration planning problem is: starting from a given initial grasp of an object O (defined by the configurations of O and of the fingertips), find feasible trajectories (motions and contact forces) for the fingertips to move O to a desired final configuration}.
We have designed a comprehensive global motion planner that solves re-configuration of 3D smooth objects through pure rolling and pure sliding contacts. The novelty of our work is in applying key ideas (for organizing and structuring the search for feasible trajectories) from gross motion planning that integrate and combine these different contact modes for finding feasible motions and contact forces to achieve non-trivial 3D re-configuration tasks automatically. Our approach effectively deals with several aspects of dextrous manipulation that make it a difficult problem: high dimensionality of the search spaces, frictional contacts, pure rolling and pure sliding contacts, collision avoidance between the fingertips, and reachability and equilibrium of grasps. To the best of our knowledge, no other implemented planner accomplishes this.
The planner has been implemented and used in simulation for re-configuring smooth objects by an arbitrary number of fingertips. The simulation examples carried out, for three- and four-fingered idealized hand demonstrate the promise of our planner (see the illustration).
The planner has been also extended to incorporate the finger chain kinematics and to cope with the related kinematic aspects. Initial results in this direction have been reported in WAFR'98 and in the Advanced Robotics Journal.
Recent related publications: