Active perception is probabilistic in nature and rests on an internal model of the world that makes predictions, and updates them through learning by experience/observation. In this axis, we aim at developing such models and confront them to behavioural responses, on the one hand, and to further identify the neural implementation of such computations, on the other hand.