Landmark-based Vehicle Self-Localization Using Automotive Polarimetric Radars

TL;DR

The work tackles radar-only ego-localization for automated driving by proposing a landmark-based framework that leverages polarimetric scattering information without requiring polarimetric maps. It integrates PreCFAR covariance gridmaps, point- and line-shaped landmarks, landmark consensus, and a sliding-window pose-graph optimization to estimate the vehicle pose in SE$(2)$ relative to a map. Real-world experiments across diverse environments demonstrate RMS trajectory errors as low as $0.12\text{ m}$ and heading errors of $0.43^\circ$ with full polarimetry, and show robustness benefits as more polarimetric channels are used, especially in challenging scenes. The results highlight polarimetry as a key enabler for reliable landmark discrimination and high-precision radar-only localization in GNSS-denied settings, with future work extending to SLAM-based map construction.

Abstract

Automotive self-localization is an essential task for any automated driving function. This means that the vehicle has to reliably know its position and orientation with an accuracy of a few centimeters and degrees, respectively. This paper presents a radar-based approach to self-localization, which exploits fully polarimetric scattering information for robust landmark detection. The proposed method requires no input from sensors other than radar during localization for a given map. By association of landmark observations with map landmarks, the vehicle's position is inferred. Abstract point- and line-shaped landmarks allow for compact map sizes and, in combination with the factor graph formulation used, for an efficient implementation. Evaluation of extensive real-world experiments in diverse environments shows a promising overall localization performance of $0.12 \text{m}$ RMS absolute trajectory and $0.43 {}^\circ$ RMS heading error by leveraging the polarimetric information. A comparison of the performance of different levels of polarimetric information proves the advantage in challenging scenarios.

Figures (14)

• Figure 1: The proposed localization framework estimates the ego-vehicle's pose by optimizing a sliding window pose graph. An association (green) is shown for each pose from which a map landmark (white) is observed. The resulting localization trajectory is shown in red, the ground truth trajectory in cyan.
• Figure 2: The block diagram of the proposed approach contains the mapping (upper) as well as the localization part (lower). Lighter colored blocks and connections symbolize the multiple drives through the same environment for map creation, which are merged in the last mapping step.
• Figure 3: By the proposed Nyquist zone augmentation, the ego-motion estimation also provides correct results beyond the unambiguous Doppler velocity.
• Figure 4: PreCFAR covariance gridmaps combine a low-level radar data utilization with a meaningful accumulation of polarimetric information for incorporation into a gridmap. As a result, the superposition of scattering mechanisms and extended objects are well represented.
• Figure 5: The point- and line-shaped landmark candidates of the proposed extractors are shown for two drives in the same environment with an underlying PreCFAR gridmap of one of the drives. Comparing the non-polarimetric with the fully polarimetric experiment, the improved robustness of the point-shaped landmarks can be observed for the latter. The blue and green rectangles show the ego-vehicle positions where the optical images were taken.