An intelligent grasping system for applications in prosthetic hands

This paper presents an intelligent grasping system for applications in developing advanced prosthetic hands. The system learns how to grasp various objects based on experiments, controlled by the user, between the prosthetic hand and the object. Two target functions are learned. The first maps the hand configuration, grasp quality and contact characteristics to the object type. The second maps the object, grasp quality and contact characteristics to a stable hand configuration. Once the system learns these two functions, it enable the prosthetic hand to grasp object with little or no user intervention. Two models of artificial neural networks were used to learn these functions. Testing on 8 everyday objects in a special simulation environment show very promising results.