Abstract: This study addresses the challenges of low positioning accuracy and poor stability caused by insufficient illumination, dust interference, and communication limitations in the unmanned construction of underground powerhouse cavern groups. We develop a LiDAR-IMU fused localization and mapping method through constructing a multi-sensor fusion framework. A tightly-coupled approach is implemented using the ESKF algorithm to integrate LiDAR point cloud data with IMU motion parameters. Specifically, this system leverages LiDAR for 3D spatial feature extraction to overcome low-light constraints, while utilizing six-degree-of-freedom IMU motion parameters to compensate for data loss during rapid equipment movement or occlusion. The framework is further enhanced through synchronous integration of keyframe matching, video pose optimization, and loop closure detection mechanism to improve system robustness. Simulation tests conducted on the M2DGR dataset demonstrate that this LiDAR-IMU fusion method increases scene coverage by 40% and reduces the average positioning error down to 16 cm, showing its significant accuracy improvement over single LiDAR solutions. Practical engineering applications confirm its effectiveness in overcoming dust interference and dynamic obstacles in complex underground cavern environments, and demonstrate it has achieved a positioning accuracy and mapping stability meeting the construction requirements.

Key words: underground cavern group, high-accuracy positioning, multi-sensor fusion, LiDAR-inertial odometry, error state Kalman filter