Performance limits and stiffness control of multifingered hands

Abstract

This paper investigates the fundamental performance limitations of robot hands in terms of their ability to modulate the stiffness and the center of compliance (CC) of a grasped object. We consider the operation of a Cartesian object stiffness controller in two stages. First, object position is determined from joint measurements. Sliding, rolling, and uncertainty in the initial grasp pose can produce errors in the object location calculated from these joint angles. Since the controller produces restoring forces proportional to the calculated object position, these errors can lead to inaccuracy in the commanded forces. Errors in object position estimation can also cause difficulties in the second stage of the controller operation, force generation, which we analyze in terms of the grasp kinematics errors. In practical terms, the lower limit to attainable stiffness may be set by friction in the robot hand, which increases “effective stiffness” until the frictional forces are exceeded. One upper bound to object stiffness is finger tip compliance. Maximum stiffness is also related to the geometry of the grasp configuration: due to controller stability limits, the maximum usable stiffness decreases as the CC is moved away from the fingers, and increases as the object width increases. These results are experimentally confirmed in a precision assembly task with a two-fingered robot hand.