To program an autonomous robot so that it acts reliably in a dynamic environment is a very hard task. Towards a promising approach to this problem, we have developed a genetic fuzzy controller for a mobile robot, and showed the possibility by applying it to a simulated robot called Khepera. The robot gets input from eight infrared sensors and operates two motors according to the fuzzy inference based on the sensory input. This paper attempts to analyze the adaptive behaviors of the controller by using automata, which indicates the emergence of several strategies to make the robot to navigate the complex space without bumping agains the walls and obstacles.