Robust Metric Localization in Autonomous Driving Via Doppler Compensation with Single-Chip Radar

Pengen Gao,Shengkai Zhang,Wei Wang,Chris Xiaoxuan Lu

DOI: https://doi.org/10.1109/tits.2023.3305487

IF: 8.5

2024-01-01

IEEE Transactions on Intelligent Transportation Systems

Abstract:Metric localization is vital to autonomous driving where it corrects cumulative errors in a long-term run. Such errors are inevitable in real scenarios where GPS signals or some other drift-free exteroceptive measurements are not available, e.g., when an automobile goes through a tunnel. Using FMCW-based mmWave radars is an attractive metric localization technique with improved robustness as RF signals can traverse small particles in harsh weather conditions like snowing, foggy, and storming, but it faces a fundamental challenge of Doppler distortion. Existing works take spatial constraints to mitigate the Doppler distortion of point clouds from mechanical radars with limited accuracy. Modern single-chip mmWave radars that provide dynamic estimates, i.e., radial velocities, bring new opportunities to develop more accurate approaches. This paper presents DC-Loc++ , a robust metric localization framework by compensating Doppler distortions using a single-chip mmWave radar. It consists of an explicit velocity-assisted Doppler compensation module for each radar sub-map, an uncertainty-aware metric registration algorithm, and a failure recovery method that validates measurement constraints to generate a more confident pose graph for optimizing vehicle poses. Extensive experiments on both nuScenes dataset and a synthetic CARLA dataset show the effectiveness of DC-Loc++ , achieving 99.2% success rate and more than 20.0%, 30.2% error reductions in terms of translation and rotation estimates, respectively, compared with existing approaches.