Dynamic manipulation: Nonprehensile ball catching

Most industrial robots nowadays still employ strategies that neglect or minimize the effects of task dynamics. Some tasks, however, are intrinsically dynamic and can only be accomplished by considering their dynamic aspects. We address ball catching as a prominent and widely studied example for such a task. The paper follows a special approach to accomplish the task: the nonprehensile catching, which means catching without a form- or force-closure grasp. Depending on the tracked ball velocity, two different catching methods are proposed: First, catching of the ball during the initial contact. Second, catching the ball after an initial rebounce during the subsequent contact. For both approaches, the ball trajectory is predicted with a recursive least squares algorithm. The dynamic manipulability measure is used for the contact point selection. Once a permanent contact between ball and end effector is established, a balancing control based on force/torque feedback is applied. Both methods are experimentally validated using a six DoF industrial robot.