Active exploration using Bayesian models for multimodal perception

In this text we will present a novel solution for active perception built upon a probabilistic framework for multimodal perception of 3D structure and motion - the Bayesian Volumetric Map (BVM). This solution applies the notion of entropy to promote gaze control for active exploration of areas of high uncertainty on the BVM so as to dynamically build a spatial map of the environment storing the largest amount of information possible. Moreover, entropy-based exploration is shown to be an efficient behavioural strategy for active multimodal perception.