This paper presents an Underground Simultaneous Localization and Mapping (uSLAM) method to localize an autonomous underground robotic system and map its surroundings. A Rao-Blackwellized Particle Filter (RBPF) with the information provided by a Ground Penetrating Radar (GPR) system installed in the robot and odometry data is described. RBPF generates possible trajectories, where each one of them has its 3D occupancy grid map. A scan matching method based on groups of GPR measurements to improve the proposed trajectories is also described.