Graph-Loc: Robust Graph-Based LiDAR Pose Tracking with Compact Structural Map Priors under Low Observability and Occlusion

TL;DR

Graph-Loc tackles long-term LiDAR pose tracking with compact onboard priors by representing both map priors and scans as lightweight point-line graphs. It solves scan-to-map associations globally via unbalanced optimal transport with a graph-context regularizer, and stabilizes pose updates in low-observability segments through a degeneracy-aware delayed optimization that defers corrections along weak directions. The approach supports heterogeneous priors (polygon outlines, CAD layouts) without contour splitting and demonstrates competitive accuracy on KITTI and ERPoT, plus strong robustness under occlusion and gradual scene changes in controlled and real-world tests. The combination of global graph matching, transport-based data association, and degeneracy-aware refinement enables scalable, occlusion-tolerant localization with significantly smaller priors, suitable for long-term autonomous operation.

Abstract

Map-based LiDAR pose tracking is essential for long-term autonomous operation, where onboard map priors need be compact for scalable storage and fast retrieval, while online observations are often partial, repetitive, and heavily occluded. We propose Graph-Loc, a graph-based localization framework that tracks the platform pose against compact structural map priors represented as a lightweight point-line graph. Such priors can be constructed from heterogeneous sources commonly available in practice, including polygon outlines vectorized from occupancy/grid maps and CAD/model/floor-plan layouts. For each incoming LiDAR scan, Graph-Loc extracts sparse point and line primitives to form an observation graph, retrieves a pose-conditioned visible subgraph via LiDAR ray simulation, and performs scan-to-map association through unbalanced optimal transport with a local graph-context regularizer. The unbalanced formulation relaxes mass conservation, improving robustness to missing, spurious, and fragmented structures under occlusion. To enhance stability in low-observability segments, we estimate information anisotropy from the refinement normal matrix and defer updates along weakly constrained directions until sufficient constraints reappear. Experiments on public benchmarks, controlled stress tests, and real-world deployments demonstrate accurate and stable tracking with KB-level priors from heterogeneous map sources, including under geometrically degenerate and sustained occlusion and in the presence of gradual scene changes.

Figures (8)

• Figure 1: System overview of Graph-Loc. Each LiDAR scan is converted into structural point/line features and organized as an observation graph. A pose-conditioned visible subgraph is retrieved from a compact prior (polygon outlines or CAD/model layouts). Scan-to-map association is solved by unbalanced optimal transport to obtain globally consistent soft correspondences, followed by a degeneracy-aware delayed update for stable real-time pose tracking under occlusion and low observability.
• Figure 2: Qualitative results of feature extraction and fusion on high-occlusion environments. (a) LOAM-style short features. (b) Sector-based structural long features. (c) Fused and filtered features used to build $\mathcal{S}_t$.
• Figure 3: Illustration of structural corner inference under different occlusion scenarios. (a) Corners/intersections are not directly observed. (b) Temporary occlusion by dynamic objects. (c) Viewpoint-induced structural missingness. (d) Sector-based reasoning parallelized across angular bins.
• Figure 4: Qualitative results of our method on KITTI and ERPoT dataset.
• Figure 5: Qualitative results of our method on CMU-EXPLORATION (20 people).