Sequential Testing for Descriptor-Agnostic LiDAR Loop Closure in Repetitive Environments

TL;DR

This work tackles LiDAR loop closure in repetitive environments where perceptual aliasing undermines traditional descriptor-based matching. It introduces Seq-SPRT, a descriptor-agnostic, multi-frame verifier that accumulates short streams of descriptor distances and applies a truncated Sequential Probability Ratio Test with user-defined Type-I/II targets to decide accept/reject. The authors provide an evaluation protocol emphasizing segment-level precision (K-hit) and post-POSE-GRAPH-OPT metrics, validated on a five-sequence library dataset designed to stress aliasing. Results show that Seq-SPRT improves precision and reduces false positives compared with single-frame and heuristic baselines, and when combined with ICP, yields the most stable pose graphs in aliasing-heavy scenarios.

Abstract

We propose a descriptor-agnostic, multi-frame loop closure verification method that formulates LiDAR loop closure as a truncated Sequential Probability Ratio Test (SPRT). Instead of deciding from a single descriptor comparison or using fixed thresholds with late-stage Iterative Closest Point (ICP) vetting, the verifier accumulates a short temporal stream of descriptor similarities between a query and each candidate. It then issues an accept/reject decision adaptively once sufficient multi-frame evidence has been observed, according to user-specified Type-I/II error design targets. This precision-first policy is designed to suppress false positives in structurally repetitive indoor environments. We evaluate the verifier on a five-sequence library dataset, using a fixed retrieval front-end with several representative LiDAR global descriptors. Performance is assessed via segment-level K-hit precision-recall and absolute trajectory error (ATE) and relative pose error (RPE) after pose graph optimization. Across descriptors, the sequential verifier consistently improves precision and reduces the impact of aliased loops compared with single-frame and heuristic multi-frame baselines. Our implementation and dataset will be released at: https://github.com/wanderingcar/snu_library_dataset.

Figures (4)

• Figure 1: The book repository at Seoul National University Library, showing repetitive shelving structures. The mobile robot, equipped with a 360-degree LiDAR, stereo camera, and IMU, was used to collect the dataset.
• Figure 2: Samples from the dataset. LiDAR scans and camera images for frame 318 (a) and frame 677 in Sequence 04. Although the viewpoints differ by more than 15 m, repeated structures induce nearly identical visual and geometric patterns.
• Figure 3: Proposed Seq-SPRT loop-verification pipeline. Left: global-descriptor retrieval forms a candidate set for a query keyframe. Middle--right: descriptor distances are mapped to log-likelihood ratios using learned loop/non-loop distance densities and accumulated by a truncated SPRT to output ACCEPT/REJECT.
• Figure 4: Pseudo ground-truth trajectories (top) and single-frame precision-recall (PR) curves (bottom) for all five library sequences. The bottom row shows pairwise PR curves when each global descriptor (LiDAR Iris, ScanContext++, NDT-Map-Code) is used in the Single verification mode (one-shot decision, no multi-frame policy). Under strong structural aliasing, single-frame operation is confined to a very low-precision regime even before any multi-frame verification is applied.