Abstract: The colony of autonomous micro mobile robot will run in some unknown and unstructured environment. There are many limits to autonomous micro mobile robot. These limits make the robot only know the part information about the environment and the robot can’t remember many past statements. So it is very difficult to design the autonomous robot based on information. It is also difficult to ensure the robot behavior is very validity to its environment and its task when we design the autonomous micro mobile robot based on behavior. This paper presents a gene algorithm based on behavior. We design three kinds of base behavior: Obstacle-avoidance behavior, fault-repair behavior, and fault-scout behavior. At the beginning of the gene algorithm,the coders of these behavior are produced at random. At the last the behavior of the colony and the robot are very good to the environment and the task when the colony runs in the simulation environment. The result shows this method is very validity.