Pith. sign in

REVIEW 5 major objections 5 minor 34 references

A graph-based LiDAR tracker can hold accurate pose with map priors of only a few kilobytes, even under occlusion and featureless corridors.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · deepseek-v4-flash

2026-08-03 03:16 UTC pith:6TLPYF5O

load-bearing objection A solid systems paper with a novel OT-based graph matching idea, but it needs to explain initialization and solver details before the robustness claims fully land. the 5 major comments →

arxiv 2602.08417 v2 pith:6TLPYF5O submitted 2026-02-09 cs.RO

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

Wentao Zhao , Yihe Niu , Zikun Chen , Rui Li , Yanbo Wang , Tianchen Deng , Jingchuan Wang This is my paper
classification cs.RO
keywords LiDAR localizationcompact structural mappoint-line graphunbalanced optimal transportgraph matchingdegeneracy-aware optimizationocclusion robustnesspose tracking
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper sets out to prove that map-based LiDAR localization does not need dense point clouds or artificially inflated polygon maps. Instead, it represents both the prior map and each online scan as sparse point-line graphs and solves scan-to-map association as an unbalanced optimal transport problem with a local graph-consistency regularizer. The unbalanced mass relaxation absorbs missing, spurious, and fragmented features caused by occlusion and dynamic objects. A degeneracy-aware delayed optimizer detects weakly constrained motion directions from the eigenvalue structure of the normal matrix and postpones corrections along them until observability returns. If correct, the result is accurate and stable tracking with maps two to three orders of magnitude smaller than typical dense priors, validated on public benchmarks, controlled simulations, and real deployments.

Core claim

The central claim is that a compact structural prior, represented as a point-line graph, is sufficient for reliable LiDAR pose tracking if the association is solved globally and the update is conditioned on real-time observability. The pipeline converts each scan into an observation graph, retrieves a pose-conditioned visible subgraph of the prior by simulating LiDAR rays, and computes soft correspondences via unbalanced optimal transport with a graph-context regularizer. This avoids hard nearest-neighbour decisions, which are brittle in repetitive layouts, and tolerates missing, spurious, and fragmented structures under occlusion. In low-observability segments, the method detects weak eigen

What carries the argument

The central object is the point-line graph representation of both map and observation, with nodes carrying geometric attributes (points and line anchors) and edges encoding k-nearest-neighbour relations. The mechanism that carries the argument is unbalanced optimal transport over graph nodes, regularized by second-order pairwise consistency terms (distance consistency for points, angle consistency for lines), which computes a global soft assignment without strict mass conservation. A second mechanism is the degeneracy-aware delayed optimizer, which eigendecomposes the refinement normal matrix, detects modes with eigenvalue below a threshold, masks updates along those modes, and buffers the l

Load-bearing premise

The pipeline assumes the pose prediction before each scan is accurate enough that ray-cast retrieval keeps all correct map elements inside the candidate subgraph; if the prediction error exceeds the gating radius or top-K restriction, the right correspondences are pruned before matching begins.

What would settle it

Run the tracker down a long straight corridor with no intersecting features and no curve within sensor range; if the along-track position error grows without bound, the delayed-optimization buffer is not sufficient to prevent drift under sustained, non-recovering degeneracy.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

Share X Bluesky LinkedIn Reddit HN

If this is right

  • LiDAR localization can reduce onboard map storage from tens of megabytes to hundreds of kilobytes without sacrificing accuracy, by relying on structural graphs instead of dense point clouds.
  • Global unbalanced-optimal-transport matching removes the need for offline contour splitting, which had been used to improve matchability at the cost of map inflation and extra processing.
  • Degeneracy-aware delayed updates allow a tracker to pass through long straight corridors and parking aisles without accumulating along-track drift, because weak directions are not updated until constraints reappear.
  • The same matching and optimization machinery works with priors from different map sources (occupancy-grid outlines, CAD models, floor plans), making the method portable across environments with different mapping pipelines.
  • The pipeline can run in real time on GPU with lower mean latency than a split-based polygon pipeline, suggesting feasibility for online deployment.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The pose-conditioned retrieval step couples the method to a constant-velocity or otherwise accurate pose predictor; a natural extension would be to add a recovery mode that re-expands the candidate subgraph when the tracked pose becomes uncertain.
  • The delayed-optimization buffer may also help when observability degrades due to sensor noise or map error, not just geometric degeneracy; this could be tested by varying the degeneracy threshold in scenes with intermittent quality.
  • The unbalanced transport formulation could be reused in other graph-matching localization tasks, such as place recognition or loop closure, where partial overlap and outliers are the norm.
  • Adding semantic labels or inertial measurements would likely further stabilize the weakly constrained directions, but the paper does not explore this.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

5 major / 5 minor

Summary. Graph-Loc proposes a LiDAR pose-tracking pipeline that localizes against compact point-line graph priors derived from polygon outlines or CAD/layout sources. Each scan is converted into an observation graph; a pose-conditioned visible subgraph is retrieved from the prior by ray casting; correspondences are obtained by unbalanced optimal transport with a graph-context regularizer; and pose refinement is stabilized by a degeneracy-aware delayed update that masks weakly constrained directions. The authors evaluate on KITTI, ERPoT, a CMU-EXPLORATION-based controlled occlusion study, and real indoor/outdoor deployments, reporting competitive accuracy with sub-MB priors and ablation results supporting the roles of unbalanced OT, dynamic filtering, and delayed optimization.

Significance. If the claims hold, this is a useful practical contribution: it shows that compact, unsplit structural priors can support accurate LiDAR tracking under occlusion and low observability, and it provides evidence that global OT-based association can replace contour splitting in polygon-map pipelines. The evaluation is largely well designed: public benchmark map/tracking splits are disjoint, the controlled CMU-EXPLORATION study varies occlusion systematically, real-world experiments include independent mapping days, and the ablation experiments separate the contributions of the three main components. These strengths make the paper a credible candidate for publication, but several load-bearing aspects of the method are underspecified or internally inconsistent, as detailed below.

major comments (5)
  1. [Sec. III-B, Fig. 1, Algorithm 1] The paper never specifies the 'Global Pose Initialization' shown in Fig. 1. Algorithm 1 assumes P_{t-2} and P_{t-1} are already available, and Eq. (3) uses a constant-velocity prediction to retrieve the visible subgraph. If the initial pose or the prediction error exceeds the gating radius / top-K sparsification in Sec. III-C, the correct map elements are excluded before the UOT matching sees them, so no robust association or delayed optimization can recover. This is especially relevant to the paper's central claims about sustained occlusion, low observability, and scene change, where drift or prediction failure is most likely. Please describe the initialization procedure, report retrieval gating parameters, and provide a sensitivity test with perturbed initial poses or artificially injected prediction errors.
  2. [Sec. III-C, Eq. (4)-(8)] Eq. (4) is the core optimization, but the solver is only sketched as a 'Sinkhorn-style iterative scaling algorithm'. A standard Sinkhorn iteration applies to a linear cost plus entropy and linear marginal penalties; it does not directly handle the quadratic context term Omega_rel in Eq. (8). The paper says the coupling is evaluated sparsely and is an 'efficient approximate second-order coupling', but it does not give the actual fixed-point or alternating update, convergence conditions, or complexity. Since the central claim is that this global association is what enables robust tracking, the solver must be specified precisely enough to be reproduced and to justify the claim that Eq. (4), rather than a different surrogate, is being solved.
  3. [Table IV vs Fig. 5] There is a direct internal inconsistency. Table IV reports for the Gazebo-model prior under 20 people: Max, Mean, RMSE all '✘', i.e., no result. Yet Fig. 5(d) is captioned 'Trajectory estimated by Ours using the Gazebo-model prior under heavy occlusion (20 people).' Either the table row is missing data or the figure caption is wrong. This matters because the paper claims support for heterogeneous priors, and the Gazebo-model prior failing under exactly the heavy-occlusion condition would be an important limitation that the text does not discuss. Please correct the discrepancy and explain.
  4. [Table VII] In the ablation table, 'w.o. O.T.' has '✘' for CMU 5 and CMU 20. The text states that removing unbalanced OT causes the largest degradation, but it is never stated whether these entries indicate divergence/tracking failure or are missing for another reason. Since the controlled occlusion study is central evidence for the method's robustness, the ablated variant's behavior in these conditions must be reported explicitly (e.g., 'diverged' with a divergence time) and discussed.
  5. [Sec. III-C, Sec. IV] Several hyperparameters that directly control the method are not reported or analyzed: UOT weights beta, rho, epsilon; line cost weights w_theta, w_perp, w_parallel; degeneracy threshold tau_lambda and damping lambda_r; gating radius / top-K; and the shared total mass m. No sensitivity study or parameter-selection protocol is given. Given that the paper's robustness claims are qualitative and the method has many interacting parameters, some evidence that the results are not sensitive to reasonable changes in these values is needed for the reported numbers to be convincing.
minor comments (5)
  1. [Table II] Several entries are malformed due to missing separators, e.g., KITTI00-2 ERPoT row '149.6023.6634.01' and '7.728.36 9.61'. Please re-typeset all tables.
  2. [Table IV] The distinction between 'ERPoT' and 'ERPoT*' in Table IV should be clarified in the table caption or text. The text says the modified ERPoT is trackable only with 0 pedestrians, which matches ERPoT*, but the row labels are not explained in the table.
  3. [Sec. III-B, Fig. 3(c)] The 'weak orthogonal line' hypothesis relies on a Manhattan-world assumption. The paper correctly phrases it as soft evidence, but this assumption and its failure modes should be stated as a limitation, especially for environments with non-orthogonal dominant structures.
  4. [Eq. (9)-(10)] The context regularizer uses distance consistency for point-point pairs and angle consistency for line-line pairs. It would be helpful to state explicitly that no distance/position consistency is enforced for line-line context, and to justify that design choice.
  5. [Sec. IV-C] The indoor reference trajectories are produced by an offline LIVO pipeline. Please state the expected accuracy of this reference or report any validation, since the indoor ATE numbers are only as trustworthy as the reference.

Circularity Check

0 steps flagged

No load-bearing circularity; Graph-Loc's pose updates and evaluations are self-contained against fixed prior maps and disjoint tracking segments.

full rationale

Graph-Loc's derivation chain is not circular. The prior map M is fixed offline and tracking segments are disjoint from mapping segments (e.g., Table I), so the reported ATE numbers are not fitted to the tracked trajectories. The pose update in Eq. (2) is computed from the current observation graph S_t and a retrieved subgraph M_t; M_t is selected by LiDAR ray simulation from a constant-velocity prediction \hat{P}_t obtained from previous poses (Eq. (3)), not from the current pose being solved. That makes the visible-subgraph retrieval dependent on an accurate initialization/prediction, but that is an operational robustness assumption, not a definitional circularity. The unbalanced-OT objective (Eq. (4)) and the degeneracy mask (Eqs. (13)-(15)) contain no parameter that is defined in terms of the target trajectory, and the ablation (Table VII) works by removing components rather than by fitting a target quantity. The self-cited PS-LOC [28] appears only as related work and as a baseline in Tables IV-V; no central equation is justified by that citation. The unstated global-initialization procedure and the gating/top-K retrieval radius are legitimate correctness/robustness concerns, but they are not cases where a prediction is equivalent to an input by construction.

Axiom & Free-Parameter Ledger

5 free parameters · 6 axioms · 1 invented entities

The central result rests on several unstated engineering assumptions and undisclosed hyperparameters. No new physical entities are introduced, but one synthetic geometric evidence ('weak orthogonal line') is injected under a Manhattan-world assumption without independent validation.

free parameters (5)
  • UOT weights beta, rho, epsilon
    Eq. (4): weights for context coupling, mass relaxation, and entropic smoothing; no values or sensitivity analysis are reported.
  • Line match cost weights w_theta, w_perp, w_parallel
    Eq. (6): relative penalties for angle, cross-track, and along-track discrepancies between line features; values not given.
  • Degeneracy threshold tau_lambda and damping lambda_r
    Eqs. (13) and (15): eigen-value threshold and damping coefficient controlling when updates are masked and deferred; values not given.
  • Graph construction and retrieval parameters
    kNN neighborhood size, gating radius, sector resolution, and ray-cast maximum range are used in Sec. III-B/C but not specified.
  • UOT shared mass m
    Eq. (11) sets uniform masses mu and nu with a shared total mass m; the value of m is not disclosed and affects how strongly mass conservation is relaxed.
axioms (6)
  • domain assumption Planar 3-DoF motion model
    Eq. (1) parameterizes pose as [x, y, phi]; the method ignores pitch, roll, and vertical translation, which is valid only for ground-vehicle-like motion.
  • domain assumption Constant-velocity motion prediction
    Eq. (3) predicts the next pose by reusing the last motion increment; sharp turns, stops, or fast accelerations would violate this prior.
  • domain assumption Manhattan-world orthogonal wall hypothesis
    Sec. III-B(c) hypothesizes a weak orthogonal line when orthogonal walls are not observed; if the environment is not Manhattan, this injected soft evidence is biased.
  • domain assumption Ray-cast visible-subgraph retrieval assumes an accurate pose prediction
    Sec. III-B 'Pose-conditioned prior retrieval' casts rays from the predicted pose; if the pose error exceeds the gating radius, correct map elements are excluded before matching.
  • domain assumption Fixed prior map with no online updating
    The Introduction and Algorithm 1 treat the map M as fixed; under gradual scene changes, all robustness must come from matching and filtering alone.
  • standard math Standard optimal transport and Sinkhorn scaling background
    Eqs. (4) and (11) rely on entropic unbalanced optimal transport and its iterative solver; these are standard results, though the extension to a quadratic context term is not proved.
invented entities (1)
  • Weak orthogonal line (synthetic structural evidence) no independent evidence
    purpose: Compensate for missing orthogonal walls in corridor-like scenes under a Manhattan-world assumption (Sec. III-B(c)).
    This line is never directly observed; it is hypothesized from the assumed orthogonality of the environment and inserted as soft evidence for stabilization, with no independent validation that the assumption holds.

pith-pipeline@v1.3.0-alltime-deepseek · 17429 in / 11493 out tokens · 133215 ms · 2026-08-03T03:16:59.100409+00:00 · methodology

0 comments
Cite this review

Pith. "Pith review of Graph-Loc: Robust Graph-Based LiDAR Pose Tracking with Compact Structural Map Priors under Low Observability and Occlusion." pith.science (2026). https://pith.science/paper/6TLPYF5O

@misc{pith2026260208417,
  author       = {Pith},
  title        = {Pith review of: Graph-Loc: Robust Graph-Based LiDAR Pose Tracking with Compact Structural Map Priors under Low Observability and Occlusion},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/6TLPYF5O}},
  note         = {Machine review of arXiv:2602.08417}
}
Share X Bluesky LinkedIn Reddit HN
read the original 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

Figures reproduced from arXiv: 2602.08417 by Jingchuan Wang, Rui Li, Tianchen Deng, Wentao Zhao, Yanbo Wang, Yihe Niu, Zikun Chen.

Figure 1
Figure 1. Figure 1: System overview of Graph-Loc. Each LiDAR scan is converted into structural point/line features and organized as an [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Qualitative results of feature extraction and fusion on [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Illustration of structural corner inference under different occlusion scenarios. (a) Corners/intersections are not directly [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Qualitative results of our method on KITTI and ERPoT dataset. [PITH_FULL_IMAGE:figures/full_fig_p009_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Qualitative results of our method on CMU [PITH_FULL_IMAGE:figures/full_fig_p010_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Our data-collection platforms. C. Real-World Experiments We conduct real-world experiments to validate deployable pose tracking with compact priors under two practical factors: (i) weak observability with sustained short-range occlusions, and (ii) long-range tracking under gradual scene changes [PITH_FULL_IMAGE:figures/full_fig_p010_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Trajectory estimated by our method using the CAD [PITH_FULL_IMAGE:figures/full_fig_p011_7.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

34 extracted references · 1 linked inside Pith

  1. [1]

    Real-time performance-focused localization techniques for autonomous vehicle: A review,

    Y . Lu, H. Ma, E. Smart, and H. Yu, “Real-time performance-focused localization techniques for autonomous vehicle: A review,”IEEE Trans- actions on Intelligent Transportation Systems, vol. 23, no. 7, pp. 6082– 6100, 2021

    2021

  2. [2]

    A survey on global lidar localization: Challenges, advances and open problems,

    H. Yin, X. Xu, S. Lu, X. Chen, R. Xiong, S. Shen, C. Stachniss, and Y . Wang, “A survey on global lidar localization: Challenges, advances and open problems,”International Journal of Computer Vision, vol. 132, no. 8, pp. 3139–3171, 2024

    2024

  3. [3]

    Loam: Lidar odometry and mapping in real- time

    J. Zhang, S. Singhet al., “Loam: Lidar odometry and mapping in real- time.” inRobotics: Science and systems, vol. 2, no. 9. Berkeley, CA, 2014, pp. 1–9

    2014

  4. [4]

    Kiss-icp: In defense of point-to-point icp–simple, accu- rate, and robust registration if done the right way,

    I. Vizzo, T. Guadagnino, B. Mersch, L. Wiesmann, J. Behley, and C. Stachniss, “Kiss-icp: In defense of point-to-point icp–simple, accu- rate, and robust registration if done the right way,”IEEE Robotics and Automation Letters, vol. 8, no. 2, pp. 1029–1036, 2023

    2023

  5. [5]

    F-loam: Fast lidar odometry and mapping,

    H. Wang, C. Wang, C.-L. Chen, and L. Xie, “F-loam: Fast lidar odometry and mapping,” in2021 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2021, pp. 4390–4396

    2021

  6. [6]

    Erpot: Effective and reliable pose tracking for mobile robots using lightweight polygon maps,

    H. Gao, Q. Qiu, H. Liu, D. Liang, C. Wang, and X. Zhang, “Erpot: Effective and reliable pose tracking for mobile robots using lightweight polygon maps,”IEEE Transactions on Robotics, 2025

    2025

  7. [7]

    Autonomous indoor exploration via polygon map construction and graph-based slam using directional endpoint features,

    H. Gao, X. Zhang, J. Wen, J. Yuan, and Y . Fang, “Autonomous indoor exploration via polygon map construction and graph-based slam using directional endpoint features,”IEEE Transactions on Automation Science and Engineering, vol. 16, no. 4, pp. 1531–1542, 2018

    2018

  8. [8]

    Topological indoor localization and navigation for autonomous mobile robot,

    H. Cheng, H. Chen, and Y . Liu, “Topological indoor localization and navigation for autonomous mobile robot,”IEEE Transactions on Automation Science and Engineering, vol. 12, no. 2, pp. 729–738, 2014

    2014

  9. [9]

    High definition map mapping and update: A general overview and future directions,

    B. Wijaya, K. Jiang, M. Yang, T. Wen, Y . Wang, X. Tang, Z. Fu, T. Zhou, and D. Yang, “High definition map mapping and update: A general overview and future directions,”arXiv preprint arXiv:2409.09726, 2024

    Pith/arXiv arXiv 2024

  10. [10]

    Dare-slam: Degeneracy-aware and resilient loop closing in perceptually-degraded environments,

    K. Ebadi, M. Palieri, S. Wood, C. Padgett, and A.-a. Agha- mohammadi, “Dare-slam: Degeneracy-aware and resilient loop closing in perceptually-degraded environments,”Journal of Intelligent & Robotic Systems, vol. 102, no. 1, p. 2, 2021

    2021

  11. [11]

    A pose graph- based localization system for long-term navigation in cad floor plans,

    F. Boniardi, T. Caselitz, R. K ¨ummerle, and W. Burgard, “A pose graph- based localization system for long-term navigation in cad floor plans,” Robotics and Autonomous Systems, vol. 112, pp. 84–97, 2019

    2019

  12. [12]

    Vision meets robotics: The kitti dataset,

    A. Geiger, P. Lenz, C. Stiller, and R. Urtasun, “Vision meets robotics: The kitti dataset,”The International Journal of Robotics Research, vol. 32, no. 11, pp. 1231–1237, 2013

    2013

  13. [13]

    Representation gran- ularity enables time-efficient autonomous exploration in large, complex worlds,

    C. Cao, H. Zhu, Z. Ren, H. Choset, and J. Zhang, “Representation gran- ularity enables time-efficient autonomous exploration in large, complex worlds,”Science Robotics, vol. 8, no. 80, p. eadf0970, 2023

    2023

  14. [14]

    A portable three-dimensional lidar-based system for long-term and wide-area people behavior mea- surement,

    K. Koide, J. Miura, and E. Menegatti, “A portable three-dimensional lidar-based system for long-term and wide-area people behavior mea- surement,”International Journal of Advanced Robotic Systems, vol. 16, no. 2, p. 1729881419841532, 2019

    2019

  15. [15]

    Topolog- ical local-metric framework for mobile robots navigation: a long term perspective,

    L. Tang, Y . Wang, X. Ding, H. Yin, R. Xiong, and S. Huang, “Topolog- ical local-metric framework for mobile robots navigation: a long term perspective,”Autonomous Robots, vol. 43, pp. 197–211, 2019

    2019

  16. [16]

    Error analysis-based map compression for efficient 3-d lidar localization,

    Y . Liu, J. Tao, B. He, Y . Zhang, and W. Dai, “Error analysis-based map compression for efficient 3-d lidar localization,”IEEE Transactions on Industrial Electronics, vol. 70, no. 10, pp. 10 323–10 332, 2022

    2022

  17. [17]

    Lio-sam: Tightly-coupled lidar inertial odometry via smoothing and mapping,

    T. Shan, B. Englot, D. Meyers, W. Wang, C. Ratti, and D. Rus, “Lio-sam: Tightly-coupled lidar inertial odometry via smoothing and mapping,” in2020 IEEE/RSJ international conference on intelligent robots and systems (IROS). IEEE, 2020, pp. 5135–5142

    2020

  18. [18]

    Range image-based lidar localization for autonomous vehicles,

    X. Chen, I. Vizzo, T. L ¨abe, J. Behley, and C. Stachniss, “Range image-based lidar localization for autonomous vehicles,” in2021 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2021, pp. 5802–5808

    2021

  19. [19]

    Robust lidar- based localization in architectural floor plans,

    F. Boniardi, T. Caselitz, R. K ¨ummerle, and W. Burgard, “Robust lidar- based localization in architectural floor plans,” in2017 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2017, pp. 3318–3324

    2017

  20. [20]

    Indoor mapping using low-cost mls point clouds and architectural skeleton constraints,

    J. Luo, Q. Ye, S. Zhang, and Z. Yang, “Indoor mapping using low-cost mls point clouds and architectural skeleton constraints,”Automation in Construction, vol. 150, p. 104837, 2023

    2023

  21. [21]

    A heterogeneous 3d map-based place recognition solution using virtual lidar and a polar grid height coding image descriptor,

    D. Xu, J. Liu, J. Hyypp ¨a, Y . Liang, and W. Tao, “A heterogeneous 3d map-based place recognition solution using virtual lidar and a polar grid height coding image descriptor,”ISPRS Journal of Photogrammetry and Remote Sensing, vol. 183, pp. 1–18, 2022. JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2021 13

    2022

  22. [22]

    Precise lidar slam in structured scene using finite plane and prior constraint,

    Y . Xie, W. Zhao, J. Wang, J. Wang, and W. Chen, “Precise lidar slam in structured scene using finite plane and prior constraint,” in2022 IEEE International Conference on Robotics and Biomimetics (ROBIO), 2022, pp. 584–589

    2022

  23. [23]

    Method for registration of 3-d shapes,

    P. J. Besl and N. D. McKay, “Method for registration of 3-d shapes,” inSensor fusion IV: control paradigms and data structures, vol. 1611. Spie, 1992, pp. 586–606

    1992

  24. [24]

    Fast and robust iterative closest point,

    J. Zhang, Y . Yao, and B. Deng, “Fast and robust iterative closest point,”IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 44, no. 7, pp. 3450–3466, 2021

    2021

  25. [25]

    The normal distributions transform: A new approach to laser scan matching,

    P. Biber and W. Straßer, “The normal distributions transform: A new approach to laser scan matching,” inProceedings 2003 IEEE/RSJ Inter- national Conference on Intelligent Robots and Systems (IROS 2003)(Cat. No. 03CH37453), vol. 3. IEEE, 2003, pp. 2743–2748

    2003

  26. [26]

    Lidar-based navigation using normal distributions transform filter,

    A. Shafiezadeh, N. P. Bhatt, and E. Hashemi, “Lidar-based navigation using normal distributions transform filter,” in2024 IEEE 27th Interna- tional Conference on Intelligent Transportation Systems (ITSC). IEEE, 2024, pp. 4046–4051

    2024

  27. [27]

    Sinkhorn distances: Lightspeed computation of optimal transport,

    M. Cuturi, “Sinkhorn distances: Lightspeed computation of optimal transport,”Advances in neural information processing systems, vol. 26, 2013

    2013

  28. [28]

    Ps-loc: Robust lidar localization with prior structural reference,

    R. Li, W. Zhao, T. Deng, Y . Wang, and J. Wang, “Ps-loc: Robust lidar localization with prior structural reference,” in2024 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2024, pp. 12 234–12 239

    2024

  29. [29]

    Manhattanslam: Robust planar tracking and mapping leveraging mixture of manhattan frames,

    R. Yunus, Y . Li, and F. Tombari, “Manhattanslam: Robust planar tracking and mapping leveraging mixture of manhattan frames,” in2021 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2021, pp. 6687–6693

    2021

  30. [30]

    Linear rgb-d slam for atlanta world,

    K. Joo, T.-H. Oh, F. Rameau, J.-C. Bazin, and I. S. Kweon, “Linear rgb-d slam for atlanta world,” in2020 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2020, pp. 1077–1083

    2020

  31. [31]

    Hong kong world: Leveraging structural regularity for line-based slam,

    H. Li, J. Zhao, J.-C. Bazin, P. Kim, K. Joo, Z. Zhao, and Y .-H. Liu, “Hong kong world: Leveraging structural regularity for line-based slam,”IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 45, no. 11, pp. 13 035–13 053, 2023

    2023

  32. [32]

    A two-level framework for place recognition with 3d lidar based on spatial relation graph,

    Y . Gong, F. Sun, J. Yuan, W. Zhu, and Q. Sun, “A two-level framework for place recognition with 3d lidar based on spatial relation graph,” Pattern Recognition, vol. 120, p. 108171, 2021

    2021

  33. [33]

    Graph-based global robot localization informing situational graphs with architectural graphs,

    M. Shaheer, J. A. Millan-Romera, H. Bavle, J. L. Sanchez-Lopez, J. Civera, and H. V oos, “Graph-based global robot localization informing situational graphs with architectural graphs,” in2023 IEEE/RSJ Inter- national Conference on Intelligent Robots and Systems (IROS). IEEE, 2023, pp. 9155–9162

    2023

  34. [34]

    Cornervins: Accurate localization and layout mapping for structural environments leveraging hierarchical geometric representations,

    Y . Zhang, F. Tang, and Y . Wu, “Cornervins: Accurate localization and layout mapping for structural environments leveraging hierarchical geometric representations,”IEEE Transactions on Robotics, 2025

    2025