Generating dextrous manipulation for multifingered robot hands by combining motion planning and teaching

Proposes a scheme that combines motion planning with teaching to efficiently generate dexterous manipulation for multifingered robot hands. In this scheme, a human operator wearing an instrument glove first demonstrates the task of manipulating an object. Then, a motion planner is employed to compute a contact motion between the human hand and the object from the initial grasp to the final one. The demonstration is to limit the search region of the algorithm to the neighborhood of the trajectory of the human hand measured by the glove so as to improve the efficiency. The motion planner is used to overcome uncertainties in measurement and modeling of human hands. The proposed approach has been implemented and validated by experiments.