Continual Robot Learning with Constructive Neural Networks

In this paper, we present an approach for combining reinforcement learning, learning by imitation, and incremental hierarchical development. We apply this approach to a realistic simulated mobile robot that learns to perform a navigation task by imitating the movements of a teacher and then continues to learn by receiving reinforcement. The behaviours of the robot are represented as sensation-action rules in a constructive high-order neural network. Preliminary experiments are reported which show that incremental, hierarchical development, bootstrapped by imitative learning, allows the robot to adapt to changes in its environment during its entire lifetime very efficiently, even if only delayed reinforcements are given.