Robust multimodal LiDAR-inertial odometry based on adaptive residual fusion in natural environments

Simultaneous localisation and mapping (SLAM) forms the foundation of autonomous perception and navigation in mobile robots. In geometrically degraded scenarios such as tunnels, corridors and open areas, traditional laser SLAM is prone to reduced accuracy or tracking failure due to insufficient geometric constraints. To mitigate this issue, this paper introduces LiDAR return intensity information into the LiDAR-inertial odometry framework, establishing joint constraints between geometric residuals and intensity photometric residuals. First, reliable image patches are selected based on intensity confidence and geometric complementarity; subsequently, a reliability-driven adaptive residual fusion strategy is employed within an Iterative Extended Kalman Filter (IEKF). The adaptive weights are estimated jointly from geometric degradation indicators, photometric confidence, and a robust reliability term, rather than being determined directly by the magnitude of the residuals. Consequently, large residuals arising from outliers or model mismatches are suppressed by a robust kernel and are not misclassified as highly reliable observations. Experiments on public datasets demonstrate that this method improves trajectory accuracy and estimation stability in typical geometric degradation scenarios.