Experiments in robotic sensorimotor control during grasp

A series of experiments is presented, using a robot manipulator, which attempts to reproduce human sensorimotor control during grasping. The work utilizes a multifingered, dextrous robot hand equipped with a fingertip force sensor to explore dynamic grasp force adjustment during manipulation. The work is primarily concerned with the relationship between the weight of an object and the grasp force required to lift it. Too weak a grasp is unstable and the object will slip from the hand. Too strong a grasp may damage the object and/or the manipulator. An algorithm is presented which reproduces observed human behavior during grasp-and-lift tasks. The algorithm uses tactile information from the sensor to dynamically adjust the grasp force during lift. It is assumed that there is no a priori knowledge about the object to be manipulated. The effects of different arm/hand postures and object surfaces is explored. Finally, the use of sensory data to detect unexpected object motion and to signal transitions between manipulation phases-with the coincident triggering of new motor programs-is investigated