LIO-EAC: gravity-referenced external altitude compensation for LiDAR-inertial odometry

Simultaneous localization and mapping (SLAM) using Light Detection and Ranging (LiDAR) sensors often exhibits degraded altitude estimation in environments with irregular terrain or weak vertical observability. We show through first-order error propagation of the point-to-plane update that pitch or ground-normal errors produce altitude errors scaled by the mean forward range of supporting ground points, making open terrain with distant ground returns and few vertical structures vulnerable to vertical drift. Based on this analysis, we propose LIO-EAC, an external altitude compensation module for Fast-LIO2 that does not modify its iterated extended Kalman filter state vector. The module derives a gravity-referenced pitch cue from simple moving average-filtered accelerometer measurements and estimates a lumped, slowly varying pitch bias from its residual with the iEKF-implied pitch. The pitch cue provides a proxy surface normal for point-to-plane updates, while the estimated bias realigns the stored map about the current position under an event trigger. LIO-EAC is evaluated against Fast-LIO2 in a level open field, a multi-level underground structure, and an indoor–outdoor round trip with sloped transitions. Four repeated trials, ablation studies, and geometric consistency metrics show reduced altitude drift without additional sensors or modification of the underlying estimator.