GNSS Feature Maps – Robust Lane-level Accurate GNSS Navigation in Urban Trenches

authored by: Fabian Ruwisch
supervised by: Steffen Schön
Abstract: The demand for high accuracy and high integrity positioning using the Global Navigation Satellite System (GNSS) sensor is on the rise, as GNSS is the only observation system capable of providing absolute positioning information. However, all GNSS positioning strategies are sensitive to the operating environment, posing a substantial challenge in fulfilling the localization requirements of autonomous vehicles, particularly in dense urban areas. The primary error source for GNSS-based vehicle positioning in these areas is the reception of multipath signals – combinations of direct and reflected signals – and NLOS (Non-Line-of-Sight) signals, which are only reflected signals that reach the antenna. These signals can cause significant inaccuracies in vehicle position estimation regardless of the GNSS positioning technique used. To address multipath and NLOS errors, two primary strategies have been developed. One is 3DMA (3D-Mapping-Aided) GNSS, which improves urban GNSS navigation by utilizing 3D city models. Another involves using robust estimation strategies that include all observations but reduce the impact of erroneous observations on the positioning solution through various robust loss functions. However, these techniques have limitations, such as an overly conservative down-weighting of observations, lack of robustness for highly contaminated data, the need of additional 3D city model information or computationally intensive algorithms.

In this thesis, two innovative strategies are proposed to improve GNSS-based navigation in urban trenches, building upon existing multipath mitigation strategies for single, static stations (i.e., utilizing the ground-track repeatability of ranging errors) with the objective of generating a GNSS Feature Map tailored for automotive applications. The thesis discusses all critical aspects of the map generation process in detail, including the coordinate information serving as foundation of the map, its resolution in longitudinal, lateral and vertical direction, and an in-depth evaluation of the GNSS ranging error similarity. The final product is a GNSS Feature Map consisting of satellite visibility information or pseudorange residual information for all satellite positions in a regular grid along a selected trajectory. The performance of employing various robust loss functions for computing the observation weights based on map information is theoretically evaluated through a Monte-Carlo simulation. In this context, the HG-estimator, an adapted robust estimator, is introduced. Simulation results for multi-GNSS SPP (Single Point Positioning) demonstrate that when prior knowledge of ranging errors is applied to compute observation weights, a 3D position error of around 2 m is achieved even if 70 % of the observations have a standard deviation of 100 m. The map information is further incorporated into an extended Kalman filter (EKF) framework for GNSS RTK (Real-time Kinematic) positioning, allowing either the exclusion of potential NLOS satellites or the adaption of robust estimation techniques. The evaluation and validation of these strategies are carried out based on two kinematic automotive experiments, located in a medium and deep urban trench, respectively. The impact of GNSS Feature Map information is assessed by means of typical GNSS performance parameters, such as accuracy, integrity and ambiguity resolution. Improvements of 54 % and 79 %, and 60 % and 64 % in horizontal and vertical accuracy for the medium and deep urban trenches, respectively, are achieved when applying the HG-estimator with map information. Consequently, lane keeping and lane determination accuracy requirements are met. The integrity and reliable ambiguity resolution are significantly enhanced, which leads to an overall more robust state estimation. By combining the map information with raw data from different receiver grades, the hardware independence is successfully proven. Finally, the results are compared to receiver-internal RTK solutions, yielding a significant improvement in the deep urban trench.
Organisation(s): Institute of Geodesy
Type: Doctoral thesis
No. of pages: 139
Publication date: 08.10.2025
Publication status: Published
Electronic version(s): https://doi.org/10.15488/19718 (Access: Open)
: Details in the research portal "Research@Leibniz University"

BibTeX

@phdthesis{744983f21c90413d9f566c44e0571d23,
title = "GNSS Feature Maps – Robust Lane-level Accurate GNSS Navigation in Urban Trenches",
abstract = "The demand for high accuracy and high integrity positioning using the Global Navigation Satellite System (GNSS) sensor is on the rise, as GNSS is the only observation system capable of providing absolute positioning information. However, all GNSS positioning strategies are sensitive to the operating environment, posing a substantial challenge in fulfilling the localization requirements of autonomous vehicles, particularly in dense urban areas. The primary error source for GNSS-based vehicle positioning in these areas is the reception of multipath signals – combinations of direct and reflected signals – and NLOS (Non-Line-of-Sight) signals, which are only reflected signals that reach the antenna. These signals can cause significant inaccuracies in vehicle position estimation regardless of the GNSS positioning technique used. To address multipath and NLOS errors, two primary strategies have been developed. One is 3DMA (3D-Mapping-Aided) GNSS, which improves urban GNSS navigation by utilizing 3D city models. Another involves using robust estimation strategies that include all observations but reduce the impact of erroneous observations on the positioning solution through various robust loss functions. However, these techniques have limitations, such as an overly conservative down-weighting of observations, lack of robustness for highly contaminated data, the need of additional 3D city model information or computationally intensive algorithms.In this thesis, two innovative strategies are proposed to improve GNSS-based navigation in urban trenches, building upon existing multipath mitigation strategies for single, static stations (i.e., utilizing the ground-track repeatability of ranging errors) with the objective of generating a GNSS Feature Map tailored for automotive applications. The thesis discusses all critical aspects of the map generation process in detail, including the coordinate information serving as foundation of the map, its resolution in longitudinal, lateral and vertical direction, and an in-depth evaluation of the GNSS ranging error similarity. The final product is a GNSS Feature Map consisting of satellite visibility information or pseudorange residual information for all satellite positions in a regular grid along a selected trajectory. The performance of employing various robust loss functions for computing the observation weights based on map information is theoretically evaluated through a Monte-Carlo simulation. In this context, the HG-estimator, an adapted robust estimator, is introduced. Simulation results for multi-GNSS SPP (Single Point Positioning) demonstrate that when prior knowledge of ranging errors is applied to compute observation weights, a 3D position error of around 2 m is achieved even if 70 % of the observations have a standard deviation of 100 m. The map information is further incorporated into an extended Kalman filter (EKF) framework for GNSS RTK (Real-time Kinematic) positioning, allowing either the exclusion of potential NLOS satellites or the adaption of robust estimation techniques. The evaluation and validation of these strategies are carried out based on two kinematic automotive experiments, located in a medium and deep urban trench, respectively. The impact of GNSS Feature Map information is assessed by means of typical GNSS performance parameters, such as accuracy, integrity and ambiguity resolution. Improvements of 54 % and 79 %, and 60 % and 64 % in horizontal and vertical accuracy for the medium and deep urban trenches, respectively, are achieved when applying the HG-estimator with map information. Consequently, lane keeping and lane determination accuracy requirements are met. The integrity and reliable ambiguity resolution are significantly enhanced, which leads to an overall more robust state estimation. By combining the map information with raw data from different receiver grades, the hardware independence is successfully proven. Finally, the results are compared to receiver-internal RTK solutions, yielding a significant improvement in the deep urban trench.",
keywords = "St{\"a}dtische GNSS-Navigation,, NSS-Merkmalskarte, ehrwegeabschw{\"a}chung, Robuste Sch{\"a}tzung, Autonome Fahrzeuge, Urban GNSS navigation, GNSS Feature Map, multipath mitigation,, robust estimation, autonomous vehicles",
author = "Fabian Ruwisch",
note = "A major part of this work was carried out within the framework of a project named „Korrektur von GNSS-Mehrwegeffekten f{\"u}r die zuverl{\"a}ssige Eigenlokalisierung von hochautomatisierten Fahrzeugen in innerst{\"a}dtischen Bereichen (KOMET)“ and funded by the Federal Ministry for Economic Affairs and Climate Action (BMWK). I extend my sincere thanks to the responsible people at BMWK, as well as to my colleagues from project partners Bosch and Geo++ for the excellent collaboration and insights.",
year = "2025",
month = oct,
day = "8",
doi = "10.15488/19718",
language = "English",
series = "Wissenschaftliche Arbeiten der Fachrichtung Geod{\"a}sie und Geoinformatik der Leibniz-Universit{\"a}t Hannover ",
publisher = "Gottfried Willhelm Leibniz Universit{\"a}t Hannover",
school = "Leibniz University Hannover",
}