This paper describes autonomous acquisition of target approaching behavior for a distributed autonomous robotic system on a water surface. The robot is a distributed controlled swimming one which has redundant linkage structure. Being located under unstable and uncertain floating environment, it is difficult for the robot to control itself. However, in this paper, we propose a method being simpler than those already proposed. That is, oscillation action patterns are preset heuristically, and simple Q-learning is employed to realize the target approaching behavior. The robustness of the behavior acquired by this simple scheme is shown by the observed fact that the behavior is environmental oriented, and the behavior is used recursively by the swimming robot.