Lidar-Level Localization With Radar? The CFEAR Approach to Accurate, Fast, and Robust Large-Scale Radar Odometry in Diverse Environments

Abstract

This paper presents an accurate, highly efficient, and learning-free method for large-scale odometry estimation using spinning radar, empirically found to generalize well across very diverse environments -outdoors, from urban to woodland, and indoors in warehouses and mines -without changing parameters. Our method integrates motion compensation within a sweep with one-to-many scan registration that minimizes distances between nearby oriented surface points and mitigates outliers with a robust loss function. Extending our previous approach CFEAR, we present an in-depth investigation on a wider range of data sets, quantifying the importance of filtering, resolution, registration cost and loss functions, keyframe history, and motion compensation. We present a new solving strategy and configuration that overcomes previous issues with sparsity and bias, and improves our state-of-the-art by 38%, thus, surprisingly, outperforming radar SLAM and approaching lidar SLAM. The most accurate configuration achieves 1.09% error at 5 Hz on the Oxford benchmark, and the fastest achieves 1.79% error at 160 Hz.

Are We Ready for Radar to Replace Lidar in All-Weather Mapping and Localization?

“…Another interesting result is that, when we evaluate our lidar odometry in SE(2), we observe a significant gap in the performance of lidar and radar odometry. This is somewhat contrary to what has been shown in prior work where radar odometry appeared to be getting close to the performance of lidar [35]. One important caveat here is that the underlying ground truth is in SE(3) whereas radar odometry is being estimated in SE(2).…”

“…Recently, there has been a resurgence of research into improving and refining radar-based localization. Currently, the state of the art for radar odometry with a spinning mechanical radar is CFEAR, which extracts only the k strongest detections on each scanned azimuth and subsequently matches the live radar scan to a sliding window of keyframes in a manner similar to ICP [35]. For a more detailed review of radar-based localization, we refer readers to the survey by Harlow et al [36].…”

Toward Precise Ambiguity-Aware Cross-Modality Global Self-Localization

“…Odometry estimation, for which the Oxford dataset has been predominantly used so far (cf. [58], [82], [83], [84], [85], [86]), only requires relative GT and map consistency which is satisfied by the dataset. In the case of MulRan, the maps and GT poses of the individual sequences are in itself consistent, however they are not globally consistent anymore when regarding the maps and GT poses from two traversals of the same scenario but different sequences (cf.…”

Two Key-Frame State Marginalization for Computationally Efficient Visual Inertial Navigation

“…This sensor combination is prevalent in autonomous driving architectures [7]- [9] where the control system requires both the surrounding map and the current pose relative to the surroundings to safely plan its maneuvers. Multisensor fusion algorithms that are popular in this domain include visual-inertial simultaneous localization and mapping (VI-SLAM) [10], radar [11], [12], visual inertial odometry (VIO) [13]- [15], and visual-inertial-LiDAR (VIL) odometry and mapping [16]- [18].…”