As the application of mechanical manipulators has grown in the fields of industrial automation, prosthetics and remote manipulation, the need for more versatile and adaptable end effectors has become increasingly apparent. Current manipulation practice is severely limited by our inability to adapt to a variety of parts and the lack of fidelity in force control. It has become clear th at articulated end effectors, or mechanical hands, can be used to extend manipulation capability in terms of cost effectiveness and in terms of the overall complexity of tasks th at may be performed. Coupled with more intelligent control systems, articulated hands promise to be of major importance in the future of robotics.

This thesis deals with three issues central to extending our use and understanding of articulated hands in manipulation. First we establish a rational basis for analyzing the kinematics or geometry of articulated hand designs and grasps upon objects. We define acceptable designs as those which may completely restrain a grasped object as well as im part arbitrary forces and small motions to it. A group of 600 potential hand designs was analyzed and the subset of acceptable designs identified. Secondly we introduce a systematic method for formulating force, position and stiffness control matrices for articulated hands. This includes controlling internal as well as external forces on the grasped object. Finally we look more closely at the effect of kinematic and structural design on basic limits to the accurate exertion of forces. For two- and three-link fingers we identify loci where forces may be most accurately exerted.

As an application of the ideas developed herein, a three-finger articulated hand, suitable for computer control, was designed and constructed. The description of this device, known as the Stanford/JPL hand, includes details on its active force sensing capability as well as its unique tendon actuation system.