Dextrous manipulation for multifingered hands: Planning and control

The main purpose of using robot hands is to provide dexterity to the grasp of an object, so that tasks involving manipulation of the object are easier to perform. In this work, we study the problems of modelling and control of articulated hands with a view to enhancing the dexterity of these hands. We also examine the problem of planning grasps of fingers on an object, allowing the contact points to change by rolling or sliding of the fingers on the object.; A framework is developed and presented for analyzing multifingered hands, taking into account the contact specifications. These provide constraints which can then be used to determine the kinematic and dynamic equations of motion for multifingered hand systems. The grasp matrix for a hand-object system is defined, and the concepts of force closure and prehensility are defined, and shown to be equivalent. An index for measuring the manipulability of a grasp is also provided.; We then proceed to define control laws for a multifingered hand manipulating an object, under two different contact conditions. The first is for fingertips which roll on the object surface, and the second is for controlled sliding of the fingertips on the object surface. These control laws are based on the idea of computed torque control of a single manipulator, suitably extended to take into account the relevant contact conditions.; For both of the above control laws to be implementable, it is necessary that the grasp maintain force closure over the course of manipulation. Thus, in the framework of planning, we provide a quality measure on a grasp to indicate how good the grasp is, in terms of achieving force closure. This measure is used to analyze and compare grasps. During manipulation of the object, collisions between the finger links and object must be avoided at all times, thus we explore the problem of grasp choice from this perspective. The control schemes developed and presented here are all tested and illustrated by dynamic simulation. These schemes have a hierarchical structure which should allow for parallel implementation.