LAUSR

Robust localization is essential for emergency rescue operations, infrastructure inspection, and autonomous exploration in complex underground spaces. However, the factors, such as global navigation satellite system (GNSS)-denied conditions, semi-structured environments, and complex terrain in underground spaces, pose significant challenges to robotic localization. To overcome these issues, we propose a light detection and ranging (LiDAR)-based localization method that uses Gaussian-blurred normal distribution transform (NDT) maps to achieve accurate and robust localization for robotic systems in complex underground spaces. First, we propose a point cloud density ratio index model to quantitatively evaluate the reliability of localization methods in underground spaces, providing a foundation for the adaptive parameter adjustment of localization algorithms. Second, we propose the Gaussian-blurred NDT map construction method to improve the robustness of classical NDT registration. This method fuses the distribution information from neighboring cells to represent the distribution of the current cell using the Gaussian kernel function. Finally, we present an optimization-based localization method using the unscented Kalman filter (UKF) and NDT. The initial robot pose is estimated in the prediction phase of the UKF, while the observed pose derived from NDT is used to correct the initial pose in the update phase, thereby achieving accurate localization. To validate the performance of our proposed method, comprehensive experiments are undertaken using both public and field test datasets. Satisfactorily, the absolute position root mean square error (RMSE) is less than 0.069 m, which provides a valuable reference for long-running, accurate, and robust localization in complex underground spaces.