truction. Based on visual informations, the system must select motions and view points in order to build a map of its own environment. An hierarchical decomposition of the pro-blem is proposed. The ˝rst step is dedicated to the object search in order to inventory all the objects of the scene. Our approach is based on a probabilistic description of the scene occupancy. Our search strategy consists in generating a sequence of observations such that all the probabilities will reach1everywhere an object is present and 0 elsewhere. The next step deals with the exploration of each particular object in order to improve its description. We present an object modeling as a mixture of stochastic and set membership models allo-wing to coarsely approximate the objects envelope while taking localization uncertainties into account. For this particular model, we develop an estimating algorithm and elaborate an optimal exploration process based on the localization uncertainty minimization. At last, we focus on servoing aspects that make the system able to track an object while moving around it. This problem is solved thanks to visual servoing technics whose performances are studied from the eye-in-hand/eye-to-hand cooperation point of view.
