Multi-finger “hugging”: a robust approach to sensor-based grasp planning

Considers the problem of planning the grasp operation for a multi-finger hand. The hand is expected to be able to handle three-dimensional objects of arbitrary unknown shapes as long as they are of “reasonable” size so as to make the grasping operation meaningful. The object geometry is not known beforehand. To provide input information for the interaction necessary, a whole-sensitive hand is assumed-every point of its surfaces possesses tactile sensing. The important property of the problem formulation is that in the grasp produced the hand and the fingers are expected to “hug” the object in a manner a human hand holds an apple-with the maximum contact with its surface for a firm, comfortable grasp. A novel, powerful strategy for real-time grasping is suggested which makes use of techniques for highly redundant sensor-based planar motion planning. An example is given that illustrates the performance of the approach