Vehicle localization in mountainous gravelled paths

This paper presents an approach for vehicle localization in the mountainous non-paved roads of a quarry. First, the elaboration of a 3D elevation map using data from dead reckoning, RTK-GPS and a laser sensor is described, where an extended Kalman filter was used for state and covariance estimation. Terrain traversability analysis is performed and scan points are voted in an elevation map which is probabilistically built and updated. Secondly, the vehicle's pose is estimated using a particle filter correcting the position towards the elevation map where GPS information is not used. Finally, experiments were performed in the mountainous gravelled paths of a quarry. Estimated localization results compared towards centimeter precision GPS show that the system offers sub-meter precision.