Abstract: Aiming at the problem that there is no GNSS signal in the coal mine, the state-of-the-art LiDAR SLAM algorithm is prone to degenerate due to insufficient feature constraints, a tight coupling SLAM algorithm of LiDAR and IMU for the coal mine environment is proposed. First, we design a dynamic feature point extraction method, by detecting whether there is degradation in the underground environment of coal mine to dynamically adjust the number of feature points extracted, build a rich and good feature information constraint matrix, improve the accuracy of pose estimation; then, the factor diagram optimization is used to realize the robust and accurate SLAM in the coal mine. Finally, a wide range of experimental analysis is carried out through the measured data in the coal mine. The results show that the proposed laser SLAM algorithm performs well, the pose estimation error is reduced by 50.93% in the horizontal direction and 42.13% in the vertical direction compared with LIO_SAM, it can provide technical reference for intelligent perception and safety inspection of coal mine robots.