High Speed Dexterous Manipulation with High Speed Vision

Vision is one of most important sensors for dexterous manipulation. The sampling rate of usual vision system is 30Hz. However, it is not enough to control manipulation system in realtime. In order to solve this problem, a high-speed vision system has been developed in which the sampling rate is more than 1kHz. In this paper, some examples of dexterous manipulation based on the high-speed visual systems are proposed. First, tool manipulation by multifingered hand is proposed, in which the calibration errors among the hand, the tool, and the target are compensated for in realtime by visual feedback. Next, flexible rope manipulation by a multifingered hand is proposed, in which the change of shape of the rope is observed by vision in realtime. Finally, dynamic regrasping using fingertip rotation mechanism is proposed. Experimental results are shown.