This paper describes a control method of chasing parson by a vision guided mobile robot in wholly unknown indoor environment. The robot obtains the environment information by repeating observing Stereo vision. By making a map using the information, it decides the optimal path to chase the parson avoiding obstacles. Since the observation by stereo may have an ambiguity, for example, recognizing nonexistent obstacles or not recognizing existent obstacle. Modeling this ambiguity, we compute the reliability of map by integrating multiple observation results.