Abstract: Localization and mapping of the underground ramps is one of the key technologies for achieving unmanned driving of underground mining trucks. The underground ramp is a typical unstructured environmental feature, and the road has a certain inclination angle. However, the traditional SLAM algorithm cannot obtain accurate odometer information, which makes the accuracy of positioning and mapping difficult to meet the driving requirements of unmanned mine trucks. To overcome above problems, this paper proposes a positioning and mapping method suitable for underground ramp environments in mines by studying the laser SLAM (Simultaneous Localization And Mapping) algorithm LeGO-LOAM. Firstly, in terms of the characteristics of smooth cement walls and sparse features on both sides of the underground ramp, a manual landmark-based auxiliary enhancement positioning method is designed, effectively increase the number of point cloud features, to optimize the pose estimation results and avoid the phenomenon of mapping drift. Then, in the feature preprocessing stage, a highly efficient algorithm for extracting ground points based on the fusion of laser point cloud height difference and slope information is proposed, by improving the selection strategy of ground points, which can effectively be identified for sloped potholes, to solve the problem of large inclination angle and error in underground ramps. Next, a slope point cloud surface voxel clustering algorithm based on the CVC (Curved-Voxel Clustering) clustering algorithm is designed, using the curvature voxels and hash-based data structures to segmentation of point clouds, which significantly improve the robustness of point cloud clustering in sparse and noisy environments underground. Finally, the Scan-To-Map is used for point cloud registration, while balancing the performance and speed of point cloud alignment. The on-site experiment in an underground ramp of Shandong, Sinosteel Group proved that the proposed method has improved the accuracy by 13.15% and reduced the Z-axis error by 22.3% compared with the original algorithm. The quality of the map has been significantly improved, which can effectively solve the problem of mapping and positioning in underground autonomous driving.