Towards introspective loop closure in 4D radar SLAM

TL;DR

This work tackles loop closure in 4D radar SLAM, where sparsity, noise, and directional sensing challenge traditional loop-detection methods. It adapts the TBV SLAM pipeline to 4D radar by integrating Doppler-Inertial odometry, a learned alignment verification, 360° ScanContext place recognition, and a multi-metric loop-candidate verification framework. The results show that same-direction loop closures can substantially improve trajectory accuracy, achieving an absolute trajectory error as low as 0.46 m over 1.8 km, though opposite-direction loops remain difficult. The study highlights the need for revised loop-definition criteria and suggests future work on submap-based approaches and cross-environment generalization to enhance robustness across diverse sensing setups.

Abstract

Imaging radar is an emerging sensor modality in the context of Localization and Mapping (SLAM), especially suitable for vision-obstructed environments. This article investigates the use of 4D imaging radars for SLAM and analyzes the challenges in robust loop closure. Previous work indicates that 4D radars, together with inertial measurements, offer ample information for accurate odometry estimation. However, the low field of view, limited resolution, and sparse and noisy measurements render loop closure a significantly more challenging problem. Our work builds on the previous work - TBV SLAM - which was proposed for robust loop closure with 360$^\circ$ spinning radars. This article highlights and addresses challenges inherited from a directional 4D radar, such as sparsity, noise, and reduced field of view, and discusses why the common definition of a loop closure is unsuitable. By combining multiple quality measures for accurate loop closure detection adapted to 4D radar data, significant results in trajectory estimation are achieved; the absolute trajectory error is as low as 0.46 m over a distance of 1.8 km, with consistent operation over multiple environments.

Figures (6)

• Figure 1: Top: Same-direction loop closures can be detected using multiple quality measures. Bottom: Opposite-direction loop closures cannot be detected.
• Figure 2: Architecture of the radar SLAM pipeline
• Figure 3: Classifier evaluation for alignment classification and loop closure evaluation
• Figure 4: Detected loops in Mine sequence using 5 keyframes for descriptor generation and top 3 retrieved candidates. Red (dangerous failure – false loop with undesired high confidence), orange (safe failure – false loop but with desired low confidence), blue (safe failure, correct loop with undesired low confidence), green (success, correct loop with desired high confidence). The failures in the upper part of the trajectory originate from the traversal in reverse direction.
• Figure 5: SLAM and odometry trajectories with estimated loop closures. Z axis is zoomed in to better visualize z drift.