1: \begin{abstract}

2: We consider a scenario where the aim of a group of agents is to perform the optimal coverage

3: of a region according to a sensory function. In particular, centroidal Voronoi partitions have to be computed. 

4: The difficulty of the task is that the sensory function is unknown and has to be reconstructed on line from noisy measurements.

5: Hence, estimation and coverage needs to be performed at the same time.  We cast the problem in a Bayesian regression framework, where the sensory function  is seen as a Gaussian random field. Then, we design a set of control inputs which try to well balance coverage and estimation, also discussing

6: convergence properties of the algorithm. Numerical experiments show the effectivness of the new approach.

7: \end{abstract}

8: