Capture Regions For Part Manipulation

Jean Ponce
Dept. of Computer Science and Beckman Institute
University of Illinois at Urbana-Champaign

Abstract

In this talk, I will present a brief overview of my research group's activities in computer vision, then focus on our current work in robotic planning. In particular, I will address the problem of moving a polyhedral part in a plane through a succession of contacts with simple mechanisms that may have discrete and/or continuous degrees of freedom. The proposed approach is based on constructing the capture region of each equilibrium of the part/robot system, i.e., the maximal subset of the part's configuration space such that any motion starting within this subset is guaranteed to end up in the equilibrium configuration. This reduces the problem of planning the overall part trajectory to constructing a sequence of atomic robot motions between equilibrium states. This approach does not assume that contact is maintained during the execution of the manipulation task; nor does it attempt to predict the object motion between equilibria. I will illustrate it with two examples: grasping and in-hand manipulation in the presence of large position uncertainties, and motion planning for mobile robots pushing an object while avoiding obstacles. I will present simulation results as well as preliminary experiments with actual robotic devices, including a reconfigurable gripper developed at the University of Illinois and a Nomadic Scout mobile platform. I will conclude the talk with a discussion of current limitations of our approach as well as plans for future work.
Eyal Amir
Last modified: Tue Nov 17 17:08:12 PST 1998