Robust Multi-Sensor Fusion SLAM Based on Road Network and Reflectivity Enhancement

Among the capabilities of an autonomous robot, map building and state estimation stand out as fundamental prerequisites. We propose a robust multi-sensor fusion Simultaneous Localization and Mapping (SLAM) framework for outdoor environments, enhanced by reflectivity and road network. This framework integrates geometric and textural data from LiDAR, along with inputs from IMU, wheel encoder, and GPS, incorporating additional road network data. When constructing LiDAR odometry factors within the factor graph, we not only leverage LiDAR point clouds to build geometric residuals in the form of point-edge and point-plane constraints but also estimate the reflectivity values of illuminated surfaces based on echoes intensity, generating reflectivity images. Subsequently, we employ a visual approach to construct photometric errors. To further enhance the robustness and accuracy, we jointly utilize IMU and wheel encoder to build pre-integration factors. Simultaneously, in the construction of GPS factors, we leverage road network information to assess the confidence of GPS measurements. Experimental evaluations conducted on datasets demonstrate that our proposed algorithm outperforms other SLAM systems in outdoor environments, providing accurate localization information and point cloud maps.