摘要:AbstractAs robots leave the simple and static environments to more complex and dynamic ones, they will have to improve their localisation abilities and to deal with heterogeneous and imprecise data. In this paper, we present a general cooperative framework designed to localize in an absolute way a fleet of heterogeneous vehicles. Depending on the sensors it embeds, each vehicle localize itself using a GNSS system (typically GPS), an orientation system (a compass for instance), the detection of the others robots in the neighbourhood (typically with a LIDAR) and the detection of visible geo-referenced features in the map (eg. wall, poles, etc...). These map features are often imprecise (as is typically the case with collaborative public maps such as OpenStreetMap). Our approach allows to update these features positions in the same framework. We first present the filtering approach we developed to solve the classical over-convergence problem using the SCI (Split Covariance Intersection) filter. Map feature relative detection being simultaneously the main information source as well as compute-time expensive, we show how in the same framework we optimize resource usage thanks to an entropy optimization strategy which avoids all sensor data fusion and instead selects the best one at each time step.