In past decades, there are many vision algorithms. In practice, the environment in which robot stayed is filled with stochastic noise such as shadow, puddle, and fallen leaves. Those factors damage the robustness of the past vision navigation algorithm. A new method to eliminate shadow during mobile robot navigation is presented in this paper. The presented method gives a detailed description on how to use fuzzy network to recognize and eliminate the shadow in low-resolution image during robot navigation, and optimize the structure of fuzzy neural network by genetic algorithm. Also, the experiment results obtained on THMRIII Mobile robot test-bed are given in this paper.