A neural network approach to the elimination of road shadow for outdoor mobile robot

A new method of road tracking oriented environmental noise elimination is presented for implementing navigation and control of land autonomous vehicles (ALV). The concept of vision based environmental noise is firstly introduced for the purpose of road and/or obstacle edge detection. Then, a representation of pyramid is proposed for vision processing. Furthermore, a fuzzy neural network is designed and implemented to recognize the environmental noises such as shadow and water prints on the road. With structure optimization by genetic algorithm and special training by classified samples, we use the network to guide our THMR-III (Tsinghua University Mobile Robot, Model 3) in the outdoor real world. Experiments have shown good properties for the ALV's "perception-action" behaviors, including obstacle avoidance, road following, wandering, etc. Although the work is still going on, we can see from the present results the better quality, adaptability and robustness of the above approach.