SUMMARYAn autonomous vehicle to carry parts, etc., in an industrial plant must run on an orbit composed of straight lines and curves, all the while determining its positions.This paper considers the case where the orbit is constructed using lan-ks which are relatively easy to recognize in deterrrrrmn g the run distance. Also, an autonomous vehicle system is developed which selects one of the landmarks to note and navigates on the orbit toward the object point.. .In the autonomous vehicle based on the image processing, if there exists an error due to the quantization of the image, the orbit may not be determined accurately. A case may also exist where the vehicle goes out of the orbit due to an external disturbance such as a slip.This paper also proposes a method which accurately determines the orbit for the vehicle and guides the run of the vehicle smoothly. In the proposed method, the range of the error in the image quantization is utilized positively and the object point is determined on the image with the accuracy of 0.5 pixel to determine the orbit on which to a p p m h the next object point.As the vehicle runs, the locus of the object point on the image is predicted based on the determined orbit; and the result is compared to the achral locus on the image. By this comparison, the orbit is corrected for each frame 45 when the vehicle goes out of the orbit. In addition, when there exists an obstacle, the method is applied to guide the run so that the vehicle avoids the obstacle. The usefulness of the method is demonstrated by the result of an experiment.