Selective Point Sampling for LiDAR-Inertial Odometry

The accuracy of LiDAR-Inertial Odometry (LIO) is highly sensitive to low-quality LiDAR measurements, which can substantially degrade state estimation. In this paper we propose Selective Point Sampling (SPS), a lightweight strategy that enhances LIO performance by discarding as outliers LiDAR measurements that are statistically inconsistent with their local map neighborhood. Consistency is quantified using the Kullback-Leibler (KL) divergence between each measurement's distribution and the averaged distributions of its neighboring map points, and only measurements below a KL threshold are retained to form residuals for state updates. SPS is implemented as a modular plug-in compatible with existing EKF-based LIO frameworks, requiring minimal integration effort. Extensive experiments on public datasets-NCLT, M2DGR, Botanic Garden, and Newer College-and additional custom-collected sequences demonstrate consistent localization accuracy gains across multiple baselines with negligible computational overhead. Code will be released at <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://github.com/BlueAkoasm/SPS</uri>.