Abstract
We aim to construct an object manipulation database from grasping demonstrations of human. We consider that when a human grasps an object, there are several grasping motion patterns that are unique to the object. This variation of grasping patterns comes from their manipulation purposes. However, traditional researches do not consider such purposes but consider the physical or the geometrical stability of the object. In this paper, we develop a method to construct an object manipulation database by observing grasping motions performed by humans under a certain manipulation purpose. For this purpose, we developed a sensor grove that can measure the pose of human's hand, the force of each fingertip, and the relative position and orientation from a grasping object. We also measure the 3D information of the object.
