Abstract: This paper presents a new approach for mobile robot to avoid obstacles based on multi-sonar information in an unknown environment. The approach takes multi-sonar information as the ART-2 neural network's inputs and quickly achieves identification and classification for the currently perceptual environment. Then, a fuzzy controller for mobile robot obstacle avoidance in an unknown environment is designed. In order to demonstrate the effectiveness and real-time ability of the proposed approach, we report a number of experimental results of mobile robot obstacle avoidance in an unknown environment.