We present a formal method for data fusion, based on possibilistic logic. The method has been applied to a real-world problem of noisy sensor-data fusion: the position estimation of an autonomous mobile robot navigating in an approximately and partially known office environment using a topological map. Each place in the map is characterized by a set of logical formulae axiomatizing both symbolic knowledge and uncertain information from the sensors. At each time instant during navigation, the necessity for each place is calculated using a function generated by a proof system based on sequent calculus. Several test runs using a real robot have shown the adequacy of the approach in interpreting and disambiguating the information coming from independent perceptual sources, in combination with symbolic knowledge.