Rescue robots usurally work in unknown complex enviroment with mess obstacles caused by the disaster. So the model based robot navigation methods are of less feasiblity. Moreover, in most cases, the resure targets are unvisualable. When there is a call for help from a survival wouned man blocked behind a wall in a earthquake, the rescue robot can not find the target via the visual, infrared, laser or utralsonic senors. As the sound can bypass obstacles, an audio navigation system is developed to guide the rescue robot run to the target in complex unknown and unvisalable enviroment, which can remedy the inadequacy of visual, utalsonic and infrared sensors. As the moudle based methods can not deal with the complicated conditions in the ruins, so a behavir-based audio navigation and obstacle avoidance systems is proposed. And ideal effects are obtained in experiments.