Grasping unknown objects based on 3d model reconstruction

Automatic grasping of unknown objects for multifingered robot hand is a very difficult problem because the location and model of the object are unknown and the possible hand configurations are numerous. In this paper, we propose a new strategy for modeling and grasping prior unknown objects based on powerful 3D model reconstruction. The whole system consists of a laser scanner, simulation environment, a robot arm and the HIT/DLR multifingered robot hand. The object to be grasped is scanned by a 3D laser scanner and reconstructed in simulation scene. After different grasping are evaluated within the simulation scenes, an accurate arm and hand configuration can be calculated to command the robot arm and multifingered hand. The experimental results strongly demonstrate the effectiveness of the proposed strategy