1 Overview

Nonprehensile manipulation is manipulation without a form- or force-closure grasp. Examples include pushing, throwing, juggling, tapping, batting, and rolling (Mason [24]; Higuchi [11]; B hler and Koditschek [6]; Erdmann [8]; Huang et al. [12]; Zumel and Erdmann [40]; Aiyama et al. [1]; Trinkle and Zeng [37]). In each of these examples, the robot takes advantage of the natural task dynamics to help control the motion of the part. Nonprehensile manipulation occupies the majority of the manipulation spectrum, comprising everything between situations where the robot exerts complete control to situations where the natural dynamics exert complete control. During a baseball throw, the ball is at first held firmly in the hand, then is allowed to roll off the fingers, and finally follows a free-flight trajectory determined by gravity and air resistance. The nonprehensile manipulation problem is to arrange the rolling motion on the fingers such that the release state will allow the ball to reach the goal state.

Nonprehensile manipulation allows a simple, low degree-of-freedom robot to control more part freedoms through relative motion (slipping, rolling, and free flight) and to cause motion outside of its kinematic workspace. One cost of using nonprehensile manipulation, relative to pick-and-place manipulation, is that motion planning and control are harder. A nonprehensile manipulation planner must reason about dynamics. Pick-and-place is usually concerned...