On-line Learning for Covert and Overt Perceptual Capability for Vision-based Navigation

In this paper, we propose a model to develop robots’ covert and overt perceptual capabilities using reinforcement and supervised learning. Covert perceptual behaviors are treated as actions selected by a motivational system. Overt perceptual behaviors can be learned using either supervised or reinforcement learning. We apply this model to the visionbased navigation system. Instead of dealing with problems in controlled environments with a low-dimensional state space, we test the model on images captured in non-stationary environments. Incremental Hierarchical Discriminant Regression is used to generate states on the fly. Its coarse-to-fine tree structure guarantees real-time retrieval in self-generated highdimensional state space. K-Nearest Neighbor strategy is adopted to further reduce training time complexity.