Instance-based reinforcement learning for robot path finding in continuous space

This paper presents two methods of shaping autonomous mobile robots within a framework of instance-based reinforcement learning. First one is Instance-Bared Classifier Generator, which is used to learn primitive behaviors. Second one is reinforcement learning based on Behavior Sequence Memory, which is used to learn optimal path and to distinguish hidden states. Learning capability of the proposed methods is confirmed through a path-finding task of a mobile robot in continuous space. Simulation results demonstrate that the robot can acquire behaviors such as light-seeking, collision-avoidance and wall-following; and that it can find the optimal paths in the alternately changing environments.