Towards Safe Lidar Positioning – Statistical Bounding of Lidar Pose-Error Covariance Using a Map-based Prediction

Hadi S. Wassaf, Jonathon Poage, Jason H. Rife

Peer Reviewed

Abstract:	Lidar systems are an attractive source of positioning information for automated-vehicle navigation because they provide accurate range and angular measurements while supporting other functions such as object detection and avoidance. Importantly, lidar positioning can operate in GNSS-challenged environments - e.g., urban and suburban areas where occlusion and multipath degrade GNSS accuracy, integrity, and availability. Prior research has shown that point-cloud scan matching can produce both odometry and absolute positioning with promising accuracy, in some cases approaching centimeter level in feature-rich outdoor settings. However, predicting and quantifying solution integrity (for example, position or pose protection limits) remains difficult. Our team has previously investigated an end-to-end lidar positioning pipeline and showed how sample data from a priori route surveys can be used to predict spatially varying (along-route) covariance for subsequent traversals. That approach was effective but insufficiently conservative in at least two major ways: it did not account for uncertainty in sample covariance estimates from limited data, and imposed the strict assumption that the offline measured residual bias vector is temporally stationary or persistent. In this paper, we address these limitations. Our approach is, first, to construct a statistical upper bound for the true covariance matrix that explicitly accounts for finite-sample effects and, second, to bound error bias for unknown bias-vector direction. We validate the approach using data collected from multiple dwells, a term we use to mean repeated driving passes along the same route. Estimating mapped-based bias and covariance parameters from one group of dwells and validating on separate dwells demonstrates that our proposed along-route protection-level predictions ensure conservatism better than our previous approach, while still producing reasonably tight error bounding. By improving conservatism and bias handling, the proposed approach represents progress toward solutions for safety critical automated navigation.
Published in:	Proceedings of the 2026 International Technical Meeting of The Institute of Navigation January 26 - 29, 2026 Hyatt Regency Orange County Anaheim, California
Pages:	268 - 287
Cite this article:	Wassaf, Hadi S., Poage, Jonathon, Rife, Jason H., "Towards Safe Lidar Positioning – Statistical Bounding of Lidar Pose-Error Covariance Using a Map-based Prediction," Proceedings of the 2026 International Technical Meeting of The Institute of Navigation, Anaheim, California, January 2026, pp. 268-287. https://doi.org/10.33012/2026.20558
Full Paper:	ION Members/Non-Members: 1 Download Credit Sign In