Radar-only ego-motion estimation in difficult settings via graph matching

Abstract

Radar detects stable, long-range objects under variable weather and lighting conditions, making it a reliable and versatile sensor well suited for ego-motion estimation. In this work, we propose a radar-only odometry pipeline that is highly robust to radar artifacts (e.g., speckle noise and false positives) and requires only one input parameter. We demonstrate its ability to adapt across diverse settings, from urban UK to off-road Iceland, achieving a scan matching accuracy of approximately 5.20 cm and 0.0929 deg when using GPS as ground truth (compared to visual odometry's 5.77 cm and 0.1032 deg). We present algorithms for keypoint extraction and data association, framing the latter as a graph matching optimization problem, and provide an in-depth system analysis.

What Could Go Wrong? Introspective Radar Odometry in Challenging Environments

“…We demonstrated and evaluated our system on an outdoor dataset which challenges the *RMSE values while under failure, and on the overall performance for this particularly challenging dataset. Please see [2,3,4] for performance metrics on other less challenging datasets. We remind the reader that datasets used in this paper were intentionally selected to expose where our previously existing RO system fails and highlight the benefits of the introspection addition.…”

“…• Operation in off-road environments where the robot's motion is not truly planar causes large swathes of the scene to disappear and reappear intermittently while the robot is moving [4]. Fig.…”

Milli-RIO: Ego-Motion Estimation With Low-Cost Millimetre-Wave Radar

“…[11] proposes a data association method that requires powerrange spectra and the extracted landmarks to match similar geometries within two radar scans. Cen et al [27] proposes a gradient-based, one-parameter keypoint extraction algorithm, revising the work in [11]. The technique requires dense radar scans that already contain visible shapes and patterns, and is not practical for sparse radar measurements.…”

kRadar++: Coarse-to-Fine FMCW Scanning Radar Localisation

“…For geometric verification, we rely on the radar pose refinement pipeline described in Reference [2,3], which we summarise here. This algorithm reflects the understanding that real, correctly identified landmarks are the same distance apart in any two radar scans.…”

“…Indeed, there is a burgeoning interest in exploiting FMCW radar to enable robust mobile autonomy, including ego-motion estimation [2][3][4][5][6][7], localisation [7][8][9][10][11], Simultaneous Localisation and Mapping (SLAM) [12], and scene understanding [13][14][15]. However, despite radar's promise to deliver such capabilities, the study of these tasks is only mature for cameras and Light Detection and Rangings (LiDARs), and relatively little attention has been paid to radar for the same application.…”