Planning Grasps for a Robot Hand in the Presence of Obstacles

Abstract

A partial solution to the problem of fitting a robot hand to an object in a given environment is presented. Information provided by a set of fingertip contact regions or by a grasp prototype is exploited to efficiently compute the space of wrist configurations from which the grasp specifications can be met. The size and shape of this solution space reflect the constraints imposed by the environment geometry; they can be used to compare a grasp to others or to resolve grasp symmetries, e.g., in a wrap grasp of a cylinder. A quality metric computed over the solution space can be used to optimize a grasp. A general method is described for computing this solution space, and a parallel implementation is presented. Examples demonstrate grasp comparison and symmetry resolution.