Continuous Verification and Safe Localization in Semantic High Definition Maps for Automated Driving

Abstract

This work proposes a system for direct localization in HD planning maps and their continuous verification far ahead of the vehicle using only on-board sensors and processing, offering a new level of reliability for semantic HD maps. By detecting map elements with human-like accuracy in up to 180 m distance and achieving globally probabilistically optimal data association in real time, it achieves a localization accuracy of about 2 cm and 0.02° even in outdated maps.