Exchange of location and sensor data among connected and automated vehicles will demand accurate global referencing of the digital maps currently being developed to aid positioning for automated driving. This project explored the limit of such maps’ globally-referenced position accuracy when the mapping agents are equipped with low-cost Global Navigation Satellite System (GNSS) receivers performing standard code-phase-based navigation, and presented a globally-referenced electro-optical simultaneous localization and mapping pipeline, called GEOSLAM, designed to achieve this limit.

The key accuracy-limiting factor was shown to be the asymptotic average of the error sources that impair standard GNSS positioning. Asymptotic statistics of each GNSS error source were analyzed through both simulation and empirical data to show that sub-50-cm accurate digital mapping is feasible in the horizontal plane after multiple mapping sessions with standard GNSS. GEOSLAM achieves this accuracy by (i) incorporating standard GNSS position estimates in the visual SLAM framework, (ii) merging digital maps from multiple mapping sessions, and (iii) jointly optimizing structure and motion with respect to time-separated GNSS measurements.

At the conclusion of this project, it was shown that the asymptotic average of position errors due to thermal noise, satellite orbit and clock errors, and tropospheric modeling errors are negligible. However, the position error due to inaccurate ionospheric modeling may lead to persistent dm-level biases in the horizontal position if the corrections are sourced from the IGS global ionospheric model (GIM), but other recent models such as the Fast PPP IONEX GIM perform better in this regard.

Multipath errors persist with multiple mapping sessions through the same urban corridor and may not be zero mean. With adequate multipath exclusion, persistent multipath biases may be reduced below 50 cm on average. Simulation results showed that sub-50-cm accurate digital mapping was feasible in the horizontal plane after multiple mapping sessions with code-phase-based GNSS, but larger biases persisted in the vertical direction.

A globally-referenced electro-optical SLAM pipeline, termed GEOSLAM, was detailed and demonstrated to achieve sub-50-cm horizontal localization accuracy on real data collected in a moderate urban environment by incorporating code-phase-based GNSS position estimates in the visual SLAM framework and jointly optimizing maps merged across time-separated sessions.

Simulated Corridor
Iono Bias
Multipath Mean
Geoslam Loop