Credibilistic IMM likelihood updating applied to outdoor vehicle robust ego-localization

This paper deals with the ego-localization problem for an outdoor vehicle equipped with proprioceptive sensors. In a current way, the ego-localisation consists in two stages (prediction and updating of the vehicle state) in order to provide a new vehicle state with its associated uncertainty. Previous works have proved the efficiency of several type of algorithms. Among them, we can quote the IMM (Interacting Multiple Model) approach. IMM is based on the discretization of the vehicle motion space modeled by simple manoeuvres, such as Constant Velocity (CT) or Constant Turn (CT) models. This yields a very adapted approach for heavily manoeuvring vehicles. Unfortunately, without GPS data, this approach becomes difficult to use. The idea is to combine different kind of proprioceptive information by using the IMM (Interacting Multiple Model) approach in a credibilistic way for the likelihood model updating in the prediction stage. In our case, data are coming from dead reckoning sensors. This kind of sensors complement the GPS for updating the vehicle state as well as the model likelihood.