Abstract: To address the poor robustness of traditional simultaneous localization and mapping (SLAM) algorithms in coal mine roadway environments, this paper proposes a LiDAR/IMU tightly coupled localization and mapping method. Considering the unstructured and highly dynamic characteristics of coal mine roadways, a point cloud clustering and dynamic-feature removal strategy based on the Breadth-First Search (BFS) algorithm is designed. In the front end, LiDAR feature points are fused with inertial measurement unit (IMU) data to perform time synchronization and motion compensation for motion distortions induced during LiDAR scanning. An improved iterative Kalman filter is employed to suppress noise, while the Levenberg–Marquardt (LM) method is used to refine the predicted covariance matrix, forming a tightly coupled IEKF–LM estimation framework that integrates LiDAR features and IMU pre-integration. In the back end, IMU pre-integration constraint factors and loop-closure factors are constructed with iterative initialization to perform global factor graph optimization. The proposed method is evaluated on the public KITTI dataset and through simulation experiments. Experimental results demonstrate that the proposed algorithm significantly outperforms FAST-LIO2 and A-LOAM in terms of global map consistency, and is capable of removing dynamic features while preserving a complete traversable area map. This method can provide a valuable reference for precise navigation of underground coal mine robots.