Planning and Teaching Compliant Motion Strategies

Abstract

This paper describes a planning program that synthesizes compliant motion strategies, in which an object in the grasp of a robot slides along obstacles, in an attempt to reach a goal region. The input to the planner is a model of the task geometry, a start region, and a goal region. To make the planning problem tractable, we reduce the task geometry to a finite state space, whose states are collections of vertices, edges, and faces from the configuration space of the robot. Strategy synthesis is complicated by uncertainty in the start-ing configuration of the robot and in robot sensing and con-trol. The planner synthesizes compliant motions that are guaranteed to perform state transitions despite uncertainty. Using best first search, compliant motions are synthesized until a strategy is found from the start state to a goal state. A strategy may require that the robot stop in an intermediate state, using sensors to determine the next commanded motion. 1.