Light-weighted and accurate map matching localization for high-speed autonomous racing

Abstract

Autonomous racing competitions, With the advancement of autonomous driving technology, have emerged highlights the importance of precision localization. The autonomous challenges have held in open-sky environments, where GPS-based approaches have powerful advantages. However, to enhance scalability across diverse conditions and ensure system reliability that is not compromised by single-sensor failures, heterogeneous sensor integration for global positioning is essential. LiDAR, in particular, is vital for its rich geometric information, contributing to the sophistication of autonomous driving systems. Nonetheless, adopting LiDAR comes with the challenge of processing vast amounts of data it generates. This research introduces a framework to elevate the scalability and reliability of high-speed autonomous systems by refining LiDAR map matching. Techniques like ground extraction significantly reduce computational load by approximately 40%, while addressing LiDAR sensor distortions and time-delay compensation, crucial in high-speed settings. Lastly, the research presents a graph optimization method to create georeferenced point cloud maps essential for map matching, combining LiDAR data with INS. The effectiveness of framework was validated using two distinct datasets acquired from real-world environments.