Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Advancements in Radar Odometry

Matteo Frosi, Mirko Usuelli, Matteo Matteucci

TL;DR

This work addresses robust radar odometry under adverse conditions by extending the CFEAR-3 framework with a full radar-only pipeline that includes filtering, motion compensation, oriented surface-point computation, kernel-based smoothing, and both multi-sweep registration and ICP-based pose refinement. The key contributions are two smoothing variants for oriented surface points, a local map with keyframes in a sliding window, and an ICP refinement stage that improves temporal consistency. Extensive experiments on the Oxford Radar RobotCar, MulRan, and Boreas datasets show that while some smoothing configurations may degrade performance in certain scenes, ICP-based registrations consistently enhance accuracy, with kernel smoothing and refinement yielding the best results in many cases. The results demonstrate improved localization accuracy across diverse weather and urban conditions, and the authors provide a ROS-independent C++ implementation to facilitate adoption and integration into SLAM systems.

Abstract

Radar odometry estimation has emerged as a critical technique in the field of autonomous navigation, providing robust and reliable motion estimation under various environmental conditions. Despite its potential, the complex nature of radar signals and the inherent challenges associated with processing these signals have limited the widespread adoption of this technology. This paper aims to address these challenges by proposing novel improvements to an existing method for radar odometry estimation, designed to enhance accuracy and reliability in diverse scenarios. Our pipeline consists of filtering, motion compensation, oriented surface points computation, smoothing, one-to-many radar scan registration, and pose refinement. The developed method enforces local understanding of the scene, by adding additional information through smoothing techniques, and alignment of consecutive scans, as a refinement posterior to the one-to-many registration. We present an in-depth investigation of the contribution of each improvement to the localization accuracy, and we benchmark our system on the sequences of the main datasets for radar understanding, i.e., the Oxford Radar RobotCar, MulRan, and Boreas datasets. The proposed pipeline is able to achieve superior results, on all scenarios considered and under harsh environmental constraints.

Advancements in Radar Odometry

TL;DR

This work addresses robust radar odometry under adverse conditions by extending the CFEAR-3 framework with a full radar-only pipeline that includes filtering, motion compensation, oriented surface-point computation, kernel-based smoothing, and both multi-sweep registration and ICP-based pose refinement. The key contributions are two smoothing variants for oriented surface points, a local map with keyframes in a sliding window, and an ICP refinement stage that improves temporal consistency. Extensive experiments on the Oxford Radar RobotCar, MulRan, and Boreas datasets show that while some smoothing configurations may degrade performance in certain scenes, ICP-based registrations consistently enhance accuracy, with kernel smoothing and refinement yielding the best results in many cases. The results demonstrate improved localization accuracy across diverse weather and urban conditions, and the authors provide a ROS-independent C++ implementation to facilitate adoption and integration into SLAM systems.

Abstract

Radar odometry estimation has emerged as a critical technique in the field of autonomous navigation, providing robust and reliable motion estimation under various environmental conditions. Despite its potential, the complex nature of radar signals and the inherent challenges associated with processing these signals have limited the widespread adoption of this technology. This paper aims to address these challenges by proposing novel improvements to an existing method for radar odometry estimation, designed to enhance accuracy and reliability in diverse scenarios. Our pipeline consists of filtering, motion compensation, oriented surface points computation, smoothing, one-to-many radar scan registration, and pose refinement. The developed method enforces local understanding of the scene, by adding additional information through smoothing techniques, and alignment of consecutive scans, as a refinement posterior to the one-to-many registration. We present an in-depth investigation of the contribution of each improvement to the localization accuracy, and we benchmark our system on the sequences of the main datasets for radar understanding, i.e., the Oxford Radar RobotCar, MulRan, and Boreas datasets. The proposed pipeline is able to achieve superior results, on all scenarios considered and under harsh environmental constraints.
Paper Structure (15 sections, 9 equations, 3 figures, 5 tables)

This paper contains 15 sections, 9 equations, 3 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Related works
  3. Radar odometry estimation
  4. Radar pre-filtering
  5. Motion compensation
  6. Oriented surface points computation
  7. Local mapping
  8. Registration
  9. Evaluation
  10. Setup
  11. Qualitative and Quantitative Results
  12. Oxford Radar RobotCar
  13. MulRan
  14. Boreas
  15. Conclusion

Figures (3)

  • Figure 1: Map obtained by testing the proposed method on partial sequence 10-14-35 of the Oxford Radar RobotCar dataset barnes2020oxford.
  • Figure 2: Our pipeline enhances radar odometry by incorporating a series of steps: filtering polar radar scan, motion compensation, calculating oriented surface points, smoothing, registering multiple radar scans at once, and refining the pose through ICP. The process leverages smoothing techniques to supplement additional information and aligns successive scans for further refinement after the bulk registration.
  • Figure 3: Odometry trajectory comparison among our pipeline settings with respect to CFEAR-3* and the ground truth. Starting from left to right, the first image is Oxford 18-15-20; the second image is Boreas 01-26-11: and the last image is DCC01.