With the deployment of the national deep-earth energy strategy and underground infrastructure projects in China, the demand for autonomous mobile robots in underground mines, engineering, and pipelines is growing rapidly. Underground autonomous robots have to bear troubles like satellite positioning signal denial and scene degradation which easily lead to serious error drift in robot pose estimation and distortion in environmental map construction. To address the problem of incomplete state estimation of underground degraded environment robots, an accurate and robust LiDAR-inertial SLAM framework and method is proposed. It combines the inertial odometer and the LiDAR-inertial odometer by the cascade optimization process. In addition, the intensity feature is introduced into LiDAR point cloud feature matching to reduce the matching error caused by sparse point cloud geometric features, and correct constraint direction is introduced through degradation detection to ensure the robustness and accuracy of pose estimation. The experimental results on public datasets and field tunnels show that the proposed method has excellent performance both in accuracy and robustness. The positioning accuracy in the degraded roadway reaches 0. 03 m, which can provide reliable state estimation and environment description for robots in underground degraded environments.