Planning of rapid grasp operations in unstructured scenes

Grasping of unmodelled objects in unstructured scenes is a significant task in numerous robot applications. In order to reach short cycle times, grasp planning must make the best of the image information acquired all at once, which comprises more or less visual occlusions and thus unknown regions. For time reasons, it is not acceptable to inspect in further steps all unknown regions by means of local sensing before grasp planning. We present a grasp planning system, which determines explicitly the fractions of grasp operations in known and unknown regions. Using that crucial information, a planning system may not request a tool change, but prefer a grasping operation with the mounted gripper, which moves mostly in known and free regions. Then, only a small part has to be executed slowly using local sensing with, for examples, proximity or force torque sensors