Outlier-Aware LiDAR–IMU–UWB Integration with NLOS Mitigation and Anchor Self-Calibration in Complex Scenarios

Journal article, 2026

LiDAR–inertial odometry (LIO) is susceptible to long-term drift in global navigation satellite system (GNSS)-challenged environments, limiting accurate state estimation for unmanned aerial vehicles (UAVs), mobile robots, and autonomous ground vehicles. Ultra-wideband (UWB) infrastructure can provide ranging constraints to mitigate this drift. However, its practical utility is limited by two critical bottlenecks: the labor-intensive requirement for precise anchor pre-calibration and measurement degradation caused by non-line-of-sight (NLOS) propagation. To address these challenges, we propose a UWB-aided continuous-time fusion framework that tightly couples LiDAR, IMU, and UWB data using non-uniform B-splines on Lie groups within a factor graph optimization (FGO) structure. First, to reduce the reliance on precise anchor positions, we formulate UWB anchor positions as static anchor states and jointly optimize them with the continuous-time vehicle trajectory under rough prior initialization. This design enables asynchronous range measurements to jointly constrain the spline trajectory and anchor position. Second, a map-aware NLOS mitigation strategy based on geometric penetration depth (GPD) is developed. GPD estimates physical obstructions along UWB propagation paths from the LiDAR-generated map and converts them into continuous measurement uncertainty, thereby avoiding residual-based robust kernels that suppress measurements only after large residuals emerge. The proposed method is evaluated on a public vehicle-infrastructure urban dataset and multiple self-collected datasets covering complex urban and controlled indoor environments. Experimental results demonstrate that the proposed method outperforms LiDAR–inertial, UWB–inertial, and discrete-time UWB–LiDAR–inertial baselines in both localization accuracy and robustness, while achieving long-term drift-bounded localization in GNSS-challenged environments. To facilitate further research, our implementation and datasets are available at https://github.com/PolyU-TASLAB/LIUO. A video demonstration is available at https://youtu.be/pGqk48h-OBc.