Kullback Leibler control approach to rubber band manipulation

Abstract

In this paper, we study robot manipulation of rubber bands and focus on a task where a robot winds a rubber band around objects since such a task is very common in daily life. Rubber bands are highly flexible and deformable objects whose shape is changed by a force; thus, it is not easy for a robot to autonomously operate such tasks with a rubber band due to the difficulties of accurate shape modeling and shape measurement. A key insight for overcoming these difficulties is based on observing human skills; we generally pull and stretch a rubber band by applying certain force. The following are the potential benefits of stretching and pulling a rubber band: 1) reduction of the shape diversity and 2) increased friction generated on the contact points for strengthening the contact with the object. In this paper, we propose a manipulation planning method for winding rubber bands around objects that maintains the rubber band that is being stretched during the manipulation to avoid shape modeling and shape measurement difficulties. Such a method is formulated by a stochastic optimal control problem called Kullback-Leibler control. Real-world experiments were conducted with a dual-arm Baxter robot to verify the effectiveness of our proposed method. Our robot manipulated a rubber band and wound around an object with different numbers of windings and magnitudes of applied forces by changing the parameters in the cost function.