The interpretation of data coming from the real world may require different and often complementary uncertainty models: some are better described by possibility theory, others are intrinsically probabilistic. A logic for belief functions is introduced to axiomatize both semantics as special cases. As it properly extends classical logic, it also allows the fusion of data with different semantics and symbolic knowledge. The approach has been applied to the problem of mobile robot localization. For each place in the environment, a set of logical propositions allows the system to calculate the belief of the robot's presence as a function of the partial evidences provided by the individual sensors.

A mobile robot localization system based on sensor fusion is described. Data coming from various sensors can require di#erent and often complementary uncertainty models: some are better described by possibility theory, others are intrinsically probabilistic. A logic for belief functions is introduced to axiomatize both semantics as special cases. For each place in a map of the environment, a set of logical rules allows to calculate the belief of the robot's presence, as a function of the partial evidences provided by the individual sensors. Various experimental runs have shown promising results. 1 Introduction The aim of this work is to apply to robotics a logical framework where di#erent uncertainty models, like belief functions, possibilities (i.e. consonant belief functions) and probabilities (i.e. additive belief functions) can be uniformly axiomatized. 1.1 Uncertainty models Although probability theory has a leading position in the description of uncertainty, in th...