Evaluating Localization Accuracy of Automated Driving Systems.

Area of Science:

• Robotics and Intelligent Systems
• Geomatics Engineering
• Automotive Engineering

Background:

• Automated driving systems require precise localization (<0.1m accuracy, >95% confidence).
• Existing localization techniques lack a unified validation methodology against ground truth.
• Accurate vehicle positioning is critical for safe and reliable autonomous operation.

Purpose of the Study:

• To establish a common methodology for validating vehicle localization accuracy.
• To evaluate different ground truth datasets for localization accuracy assessment.
• To identify the strengths and weaknesses of various validation methods.

Main Methods:

• Evaluation of four distinct localization validation methods: static driving path, HD map lane-centerline, HD map lane-boundary overlap, and stop point accuracy.
• Utilized two datasets: one from an automated vehicle on a static path with advanced localization systems, and another from manually-driven connected vehicles.
• Comparative analysis of validation methods against different ground truth references.

Main Results:

• Demonstrated broad applicability of the proposed validation approach across diverse datasets.
• Revealed specific advantages and disadvantages of each validation method and ground truth.
• Confirmed the feasibility of achieving sub-decimeter localization accuracy (up to 99.9% confidence) with high-quality systems.
• Highlighted the ongoing challenges in consistently achieving these high accuracy levels.

Conclusions:

• The developed methodology provides a robust framework for evaluating vehicle localization performance.
• High-definition maps and precise ground truth data are essential for reliable localization validation.
• While advanced systems show promise, achieving consistent, high-accuracy localization remains an active research area.