Interactive Learning of Direct Mappings between Visual Percepts and Actions

We introduce flexible algorithms that can automatically learn direct mappings from visual stimuli to actions. The learning process is carried on in a fully autonomous fashion, through interactions with the environment. Our algorithms work by introducing an image classifier in front of a Reinforcement Learning algorithm. The classifier is ∗Aspirant du Fonds National de la Recherche Scientifique Belge (F.N.R.S) founded on the local-appearance paradigm: It partitions the visual space according to the presence or absence of highly informative local descriptors. The image classifier is incrementally refined by selecting new local descriptors whenever perceptual aliasing is detected. Thus, we reduce the visual input domain down to a size manageable by Reinforcement Learning, permitting us to learn direct percept-to-action mappings. Experimental results on a continuous visual navigation task illustrate the applicability of the framework.