Some approaches to context based grasp planning for a multi-fingered gripper

Dexterous manipulations of a-priori unknown objects by a multi-fingered gripper require a vision assisted planning of suitable grasps. The selection of the optimum grasp is based on context information specifying the goal of manipulation given by the user, constraints due to the gripper geometry and object properties to be perceived by a vision system. The integration of vision into a hand-arm system allows planning and performance for grasping even unregularly shaped objects. The paper presents a context level for classification of specific grasp situations and a planning level for determination of gripping positions and forces. Especially strategies searching for suitable contact points for the gripper are discussed and compared to each other