Enhancing reliability of a vehicle steering algorithm by combining computer vision and neural vision

Super cruise control adds the feature of automatic steering or lateral control to the ordinary intelligent cruise control that implements only longitudinal control using brakes and accelerators. This paper addresses the problem of steering control needed for super cruise control in which automatic steering is effected in a rather limited driving environment, that is, driving on highways at high speeds. For maximum safety, a very robust real-time control algorithm is essential. To meet this objective, this paper proposes a fitness-based modular steering control architecture that ensures robustness, stability, and safety while at the same time meeting real-time constraints for high speed driving. Fitness here refers to the applicability of each expert module for the current input situation. Currently, three modules, namely edge, color, and neural modules are used for steering while the input to each module is the road image obtained by the CCD camera. The ultimate steering command solution is obtained by weighted combination of the outputs of the modules whose fitnesses are above a certain threshold. The proposed steering control algorithm has been verified through real experiments on the Postech Road Vehicle II.