Autonomous acquisition of environmental oriented behaviour for a distributed autonomous swimming robot

This paper describes a distributed autonomous swimming robot which can acquire environment oriented behaviour through learning. We demonstrate the approach by implementing the robot on the water surface where the environment is similar to an unstable and uncertain real world. Since the robot control is difficult, and it takes many repetitions before the target behaviour can be acquired. We introduce oscillation action patterns suitable for providing the motion through water for fast learning. As a result, the robot can acquire the active target-approaching behaviour using only the local learning, and the behaviour is almost the same as that in the centralized control.