Automatic grasping: An optimization approach

A method is introduced for locating stable grasps with articulated mechanical hands. A metric called the “grasp goal function” represents the “distance” to an unstable grasp for a given position, external load, and set of joint torques of the gripper. The process of locating a stable grasp is treated as an optimization problem for which the “most stable” grasp is found by maximizing the grasp goal function. It is assumed for the stability criterion that point contacts with friction exist between the gripper and object and that the force system is statically determinate. Physical constraints, such as torque or motion limitations, are accounted for by the incorporation of potential functions in the objective function. Two- and three-dimensional simulations were undertaken for which objects were represented as polygons/polyhedral and gripper elements as ellipses/ellipsoids. Results from the simulations indicate that the method is particularly useful for estimating stable grasps when the initial estimate is close to being stable, but that it also can often perform adequately when this is not the case. In some instances, however, inadequate grasps are obtained due to the presence of local optima which represent unstable grasps. Regardless of the latter deficiencies, however, it is still felt that the method will have application in future automated systems as a “low-level” component.