A method for driving a robot in a manner of imitation by watching (non-contact manner) based on the movement of a moving object, which has a complicated shape often causing self-occlusion, is provided. A plurality of image data of the robot is associated with pre-arranged operation commands and stored in an image corresponding operation command storing means 11. In order to have the robot perform a movement, the moving objecting caused to perform a desired movement, and at the same time, image data of the moving object are obtained as robot operational image data in time-series. The image data specifying and operation command generating means 14 specifies image data corresponding to the operational image data included in the robot operational image data in time-series among the plurality of image data stored in the image corresponding operation command storing means 11, and provides a pre-arranged operation command corresponding to the specified image data to the robot as an operation command to drive the robot. Owing to this, such problems as complicated shape and self-occlusion of the moving object are eliminated, and the robot performs the movement in an imitating manner by watching.