Abstract:
Simultaneous localization and mapping (SLAM) is a critical technology for unmanned driving. Existing SLAM methods have the drawbacks of significant cumulative errors and drift in coal mine roadway environment. In this study, a roadway environment feature-assisted SLAM algorithm integrating inertial measurement unit (IMU) and LiDAR was proposed. IMU observation data was used to predict the motion state of point cloud and motion compensation was applied to reduce point cloud distortion caused by equipment movement. Pose transformation information from LiDAR odometry was obtained through point cloud registration, forming a LiDAR odometry constraint. Point clouds from roadway sidewalls and floor were extracted and fitted to planes, establishing environmental constraints. Using IMU pre-integration constraints, LiDAR odometry constraints, and environmental constraints, the algorithm applied factor graph optimization to achieve tight coupling between LiDAR and IMU, enabling high-precision 3D reconstruction of roadway scenes and accurate localization of autonomous vehicles. Simulation experiments showed that the absolute trajectory root mean square error (RMSE) of the roadway environment feature-assisted IMU-LiDAR integrated SLAM algorithm was 0.1162 m, and the relative trajectory RMSE was 0.0409 m, improving positioning accuracy compared to commonly used algorithms such as LeGO-LOAM and LIO-SAM. Based on the test results in a real environment, the algorithm provides excellent mapping performance with no drift or trailing, demonstrating strong environmental adaptability and robustness.