CFEAR Radarodometry -- Conservative Filtering for Efficient and Accurate Radar Odometry

Abstract

This paper presents the accurate, highly efficient, and learning-free method CFEAR Radarodometry for largescale radar odometry estimation. By using a filtering technique that keeps the k strongest returns per azimuth and by additionally filtering the radar data in Cartesian space, we are able to compute a sparse set of oriented surface points for efficient and accurate scan matching. Registration is carried out by minimizing a point-to-line metric and robustness to outliers is achieved using a Huber loss. We were able to additionally reduce drift by jointly registering the latest scan to a history of keyframes and found that our odometry method generalizes to different sensor models and datasets without changing a single parameter. We evaluate our method in three widely different environments and demonstrate an improvement over spatially cross-validated state-of-the-art with an overall translation error of 1.76% in a public urban radar odometry benchmark, running at 55 Hz merely on a single laptop CPU thread.

Radar style transfer for metric robot localisation on lidar maps

“…Among these studies, one topic is ego-motion estimation for odometry-based localisation, which has been studied in recent studies [21][22][23][24]. These methods were proposed to achieve radar data filtering first and then estimate the robot motion between two relative radar scans.…”

“…In Ref. [22], k strongest points in radar scan could be also used for odometry estimation. Additionally, data-driven-based methods could also achieve radar data filtering.…”

“…In summary, compared to radar odometry-based pose tracking [21][22][23][24], the proposed radar-based localisation can improve pose estimation for long-distance travelling. As for long-term running, localisation on maps [26][27][28][29] is a more efficient and applicable solution compared to radar SLAM [25].…”

“…where s i is a particle of estimator, and u is the robot motion between the two timestamps t − 1 and t. Several recent methods [21][22][23][24] could be used to estimate the radar egomotion between two consecutive radar scans. Considering the computation complexity of handcraft or learning-based feature extraction, we propose to use point-to-plane-based iterative closest points (ICP) [39] for motion estimation.…”

“…The ground truth in RobotCar dataset has been used for odometry evaluation in many recent papers [21][22][23][24]. However, starting positions are not provided and different sessions were collected with different start positions, which means the provided ground truth ego-motion could not be used for mapbased localisation directly.…”