We have developed a sensor to detect position and posture of an object in a short-range, called a "proximity sensor". The proximity sensors are mounted on fingertips of a robot hand, and it is realized that the fingertip surface and an object surface face each other before contact in grasping process by the robot. We call such postural adjustment "preshaping". When grasping an object by a robot hand system with the combination of the proximity sensor and a visual sensor, the visual sensor has only to guide the robot hand so that the proximity sensor detects the object and pre-shaping is performed appropriately. In this paper, we investigate the information which the visual sensor should acquire to perform pre-shaping appropriately.
