Pith. sign in

REVIEW 3 major objections 6 minor 41 references

Reusing G-ICP covariances lets LiDAR Gaussian Splatting SLAM build dense outdoor maps online at over 20 FPS with high mesh fidelity.

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 · grok-4.5

2026-07-11 21:29 UTC pith:SQGY5NQO

load-bearing objection Solid real-time LiDAR-only GS-SLAM with bidirectional G-ICP covariance coupling; headline Newer College F-score/FPS holds on the reported data, with the main soft spot being reliance on those covariances outside clean geometry. the 3 major comments →

arxiv 2607.04127 v1 pith:SQGY5NQO submitted 2026-07-05 cs.CV cs.RO

Real-Time LiDAR Gaussian Splatting SLAM

Seungjun Tak , Yewon Jeon , Jaeik Hwang , SukMin Hwang , Seongbo Ha , Hyeonwoo Yu This is my paper
classification cs.CV cs.RO
keywords LiDAR SLAMGaussian Splattingdense mappingG-ICPreal-time reconstructionspherical rasterizationmap management
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.

This paper claims that real-time dense mapping with Gaussian Splatting is practical for large-scale LiDAR-only outdoor SLAM if tracking and mapping share the same local covariance geometry. Rather than relying on appearance or a separate geometry module, the system reuses the covariances already computed during fast G-ICP registration to initialize surface-oriented Gaussians with range-aware scales and normals, and to form a complexity score that prunes planar regions while densifying edges and corners. Optimized Gaussians and LiDAR-specific confidence then feed back as better registration targets and weights, closing a mutual loop. On Newer College the full online system reaches an 86.78% mesh F-score at more than 20 FPS, with competitive trajectory error and compact maps on Oxford Spires and KITTI as well. A sympathetic reader cares because sparse point or voxel maps localize well but lack continuous surfaces needed for simulation and HD mapping; this coupling aims to deliver those surfaces without losing real-time speed or long-horizon scalability.

Core claim

Tightly coupling G-ICP tracking with spherical 2D Gaussian mapping—by reusing tracking covariances for Gaussian initialization, normal supervision, and a geometry control score, while feeding refined Gaussians back as covariance-aware targets—enables real-time LiDAR-only dense SLAM that is accurate, compact, and scalable on large outdoor sequences.

What carries the argument

Covariance-derived geometry coupling: local G-ICP covariances supply orientation, range-adaptive in-plane scales, normals, and a control score (linearity, curvature, residual) that drives planar cover-and-prune and selective splitting; optimized Gaussians supply O(1) target covariances and confidence for subsequent tracking.

Load-bearing premise

The method assumes local LiDAR neighborhoods produce trustworthy covariances and normals; if the scene is sparse, leafy, or full of moving objects, those geometric priors fail and both the map and the feedback to tracking degrade.

What would settle it

Run the full online system on a dense-vegetation or high-dynamic urban sequence and check whether mesh F-score and ATE fall below competing online dense methods; if control-score pruning and covariance feedback still preserve quality, the claim holds, otherwise the geometric priors are the breaking point.

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

Share X Bluesky LinkedIn Reddit HN

If this is right

  • Dense continuous LiDAR maps can be maintained online above 20 FPS without unbounded primitive growth.
  • Planar compression plus selective densification yields substantially smaller persistent maps (e.g., fewer Gaussians and lower storage on long KITTI sequences) while keeping reconstruction quality competitive.
  • Mapping-refined surfel-like targets improve both trajectory accuracy and tracking speed relative to a frozen map.
  • Geometry-only online LiDAR Gaussian SLAM can approach the mesh quality of offline ground-truth-pose dense mappers on handheld campus scenes.

Where Pith is reading between the lines

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

  • The same covariance-sharing pattern may help multi-sensor systems when photometric cues are weak or intermittent, such as night or adverse weather driving.
  • Control-score map budgeting could transfer to other explicit primitive maps that grow linearly on long trajectories.
  • Keyframe-wise rigid Gaussian correction after loop closure leaves residual local inconsistencies that a later global Gaussian adjustment might remove.
  • Reliability masks from ray-drop and normal inconsistency are a natural stress test for whether bad covariances can be isolated without separate dynamic-object detectors.

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

3 major / 6 minor

Summary. The paper proposes a real-time LiDAR-only Gaussian Splatting SLAM system that couples G-ICP tracking with spherical 2D Gaussian mapping. Tracking covariances initialize range-adaptive Gaussian scales, orientations/normals, opacity via a physics-based confidence u, and a control score c_i (Eq. 9) used for planar cover-and-prune and selective splitting; optimized Gaussians and confidence cues are fed back as covariance-weighted registration targets. Loop closure applies keyframe-wise piecewise-rigid Gaussian updates. On Newer College (online poses) the system reports F-score 86.78% at >20 FPS with compact maps; ATE is competitive on Newer College, Oxford Spires, and KITTI, with ablations isolating bidirectional transfer, losses, and map management (Tables 1–6).

Significance. If the reported online F-score, speed, and map compactness hold under independent reimplementation, this is a solid systems contribution for LiDAR-only dense GS-SLAM: it shows that reusing G-ICP covariances can replace separate geometry estimation and that geometry-aware budget control can keep maps scalable without collapsing reconstruction quality. Strengths include multi-dataset evaluation with online poses (not only GT), structured ablations of tracking↔mapping transfer and map management, explicit comparison to GT-pose dense mappers as upper bounds, and public code/project page. The work is incremental relative to prior GS-ICP and spherical LiDAR GS lines but addresses a practical gap (real-time LiDAR-only dense mapping with bounded growth).

major comments (3)
  1. Limitations §5 and the design of §3.3–3.5: the central claim (online F-score 86.78% at >20 FPS with compact maps) rests on local G-ICP covariances being reliable priors for initialization (range-adaptive scales, normals), normal loss (Eq. 11), control score c_i (Eq. 9), and pruning/splitting. The paper correctly flags failure under sparse returns, vegetation, and dynamics, but the evaluation sets (Newer College, Oxford Spires, KITTI) do not stress these regimes. A load-bearing addition is at least one quantitative stress sequence (or subset) with vegetation/dynamics/sparse returns, reporting ATE, F-score, and map growth with/without the covariance-derived terms; without that, the headline result is scoped more narrowly than the abstract implies.
  2. §3.2 Loop Closure and §5: after pose-graph optimization, Gaussians are updated only by keyframe-wise piecewise-rigid deltas rather than global re-optimization. Table 1 shows competitive ATE where loops help, but there is no measurement of residual map inconsistency (e.g., mesh Acc/Com or local surface error before vs. after large loop corrections, or a long-loop sequence with known large drift). Because dense map quality under online trajectories is a primary claim (Table 2, Fig. 4), this correction model needs either quantitative support or a clearer statement that map metrics are reported only under mild loop corrections.
  3. Tables 1–6 and free parameters listed in §3–4: many weights and thresholds (w_l,w_c,w_r; λ_α,λ_n,λ_s,λ_n,g,λ_n,s; q_plane,q_split; κ,s_max; r0,c0; voxel δ; trackable masks) are fixed without sensitivity analysis. The ablations show components matter, but not that the reported operating point is stable. For a systems paper claiming real-time scalability across datasets, a short sensitivity or leave-one-dataset-tuned check on the control-score weights and prune/split quantiles would strengthen the claim that results are not brittle to these choices.
minor comments (6)
  1. Fig. 1 caption and body: map size (MB) and FPS are central to the efficiency claim, but Table 2 reports FPS and F-score without a uniform map-size column for all methods; align Fig. 1 numbers with Table 2 or add map size/#GS for baselines.
  2. Eq. (2)–(5): curvature κ_i = s_i,0/(s_i,1+ε) is inverted relative to usual curvature language (lower κ̄ means more planar); a one-sentence clarification would avoid confusion with the eigenvalue-based curv_i in Eq. (9).
  3. Table 1: Splat-LOAM fails on several KITTI sequences under the authors’ protocol; briefly state whether failure is divergence, ATE>50 m, or resource limits so the comparison is reproducible.
  4. Implementation details: report the exact spherical image resolution, keyframe policy, and whether multi-pass planar prune runs every keyframe or on a schedule; these affect the claimed >20 FPS.
  5. Related Work: GSO-SLAM and G2S-ICP are cited; a short explicit contrast table (sensor modality, online vs offline, map management) would help position the LiDAR-only claim.
  6. Typos/notation: “amulti-pass” (§3.5); inconsistent use of s_i vs (s_i,x,s_i,y) vs ˜s; ensure σ vs λ notation is defined once before Eq. (9).

Circularity Check

0 steps flagged

No significant circularity: empirical systems paper whose F-score/ATE claims rest on external benchmarks and ablations, not on self-definitional or fitted reductions.

full rationale

The paper is a real-time LiDAR GS-SLAM systems contribution. Its load-bearing claims (F-score 86.78% on Newer College from purely online trajectories at >20 FPS, competitive ATE, compact maps via control-score pruning/densification) are measured against public datasets (Newer College, Oxford Spires, KITTI) and external baselines (KISS-SLAM, PIN-SLAM, SuMa, Splat-LOAM, Voxblox, etc.). The covariance reuse (G-ICP eigendecompositions for range-adaptive scales, normals, control score c_i = clip(w_l linear + w_c curv + w_r fres, 0,1) with fixed weights, trackability confidence) and bidirectional feedback are design choices whose utility is isolated by ablations (Tables 3–6) rather than forced by construction or by a self-citation uniqueness theorem. Self-citations to prior GS-ICP/GSO-SLAM work from the same lab appear in Related Work but are not load-bearing for the reported metrics; those metrics are not algebraic restatements of the loss weights or control-score coefficients. Limitations section candidly flags the covariance reliability premise under sparse/vegetation/dynamic scenes. No self-definitional loop, fitted-input-as-prediction, or ansatz-smuggling reduction is present. Score 0 is therefore the correct, proportionate finding.

Axiom & Free-Parameter Ledger

6 free parameters · 4 axioms · 2 invented entities

The central claim is empirical and rests on standard SLAM/geometry machinery plus many hand-chosen scalars that allocate map budget and weight losses. No new physical entity is postulated; the ‘control score’ and LiDAR confidence are engineered scalars. Load-bearing domain assumptions are that G-ICP local covariances encode usable surface geometry and that spherical 2D Gaussians plus depth/normal losses suffice without photometric cues.

free parameters (6)
  • control score weights (w_l, w_c, w_r)
    Linear combination of linearity, curvature, and residual into c_i; set to 0.55/0.30/0.15 in implementation and drive prune/split decisions that enable the scalability claim.
  • mapping loss weights (λ_α, λ_n, λ_s, λ_n,g, λ_n,s)
    Fixed to 0.1, 1, 10, 0.05, 0.01; directly shape optimized Gaussians that become tracking targets.
  • prune/split quantiles and ratio caps (q_plane, q_split, max prune/split ratios)
    Thresholds that decide which Gaussians are redundant vs split; essential to map-size and F-score trade-off but not derived from first principles.
  • range-adaptive scale factor κ and s_max, opacity bounds α_min/α_max
    Hand-set bounds for Gaussian init and scale regularization; affect far-range coverage and over-smoothing.
  • voxel size δ, k_min, α_min for trackable subset, ω_max=10
    Tracking front-end and planar weighting knobs that affect ATE and FPS.
  • physics confidence parameters (r0, c0) in u
    Range and incidence cutoffs in the LiDAR reliability weight used for opacity init and trackability.
axioms (4)
  • domain assumption G-ICP local neighborhood covariances yield usable principal axes, tangent scales, and normals for surface-oriented 2D Gaussians.
    Invoked throughout §3.2–3.3 for init, normal loss, and control score; Limitations admit failure under weak geometry.
  • domain assumption Spherical range-image rasterization of 2D anisotropic Gaussians is an adequate dense map for LiDAR geometry without appearance.
    Map representation §3.1 and optimization §3.4; inherited from 2DGS / Splat-LOAM-style spherical projection.
  • ad hoc to paper Keyframe-wise piecewise-rigid Gaussian updates after pose-graph loop closure sufficiently correct the map without global re-optimization.
    Loop Closure paragraph in §3.2; Limitations note residual local inconsistencies after large corrections.
  • standard math Standard SE(3) registration, pose-graph optimization, and mesh-to-mesh distance metrics are valid evaluation machinery.
    Used for tracking and mapping evaluation in §4.
invented entities (2)
  • covariance-derived control score c_i no independent evidence
    purpose: Scalar in [0,1] fusing linearity, curvature, and G-ICP residual to drive planar cover-and-prune vs selective splitting.
    Engineered map-budget signal; no independent physical meaning outside this pipeline; validated only via ablations on the same system.
  • LiDAR physics-based confidence u and trackability γ_i=α_i u_i no independent evidence
    purpose: Initialize opacity and filter/weight tracking correspondences from range and incidence.
    Sensor-motivated but parameter-dependent construct; not measured against external reliability labels.

pith-pipeline@v1.1.0-grok45 · 19086 in / 3656 out tokens · 33148 ms · 2026-07-11T21:29:38.395684+00:00 · methodology

0 comments
Cite this review

Pith. "Pith review of Real-Time LiDAR Gaussian Splatting SLAM." pith.science (2026). https://pith.science/paper/SQGY5NQO

@misc{pith2026260704127,
  author       = {Pith},
  title        = {Pith review of: Real-Time LiDAR Gaussian Splatting SLAM},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/SQGY5NQO}},
  note         = {Machine review of arXiv:2607.04127}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

We present a real-time LiDAR-based framework for Gaussian Splatting SLAM that tightly couples fast G-ICP registration with spherical rasterization-based dense mapping for large-scale sequences. Leveraging LiDAR geometry rather than appearance, we reuse tracking-estimated local covariances to initialize Gaussians with range-aware scales and to derive surface normals for geometry-aware map optimization. We further introduce a covariance-derived geometry score that measures local complexity and drives pruning in planar regions and selective densification in structurally rich areas, while optimized Gaussians and LiDAR-specific confidence cues are fed back to improve tracking robustness. On the Newer College dataset, our method achieves an F-score of 86.78\% using purely online trajectories at real-time speed ($>$20 FPS), and additional experiments on other datasets confirm its stability and scalability.

Figures

Figures reproduced from arXiv: 2607.04127 by Hyeonwoo Yu, Jaeik Hwang, Seongbo Ha, Seungjun Tak, SukMin Hwang, Yewon Jeon.

Figure 1
Figure 1. Figure 1: Performance Overview of LiDAR GS. We compare dense mapping SLAM with respect to Map Size (MB) and Frame per Second (FPS), with F-score (%). The two panels illustrate the trade-off between efficiency and accuracy. We address these issues with a LiDAR-based Gaussian Splatting SLAM sys￾tem that enables scalable and reliable dense mapping in large-scale outdoor envi￾ronments. We perform fast tracking using Gen… view at source ↗
Figure 2
Figure 2. Figure 2: Overview of the proposed system. We downsample each LiDAR scan and estimate per-point covariances to form the source set P ′ s t . Tracking registers P ′ s t to a trackable target set from the map via G-ICP to estimate Tt, and the covariances produce a control score ci for pruning/densification. Keyframes are fused into the 2D Gaussian map and optimized, while reusing stored target parameters avoids per-fr… view at source ↗
Figure 3
Figure 3. Figure 3: Covariance-derived control score and decision map. From each tracking covariance, we compute linearity/curvature proxies from eigenvalues and fuse them with a robustly normalized G-ICP residual to obtain a control score ci ∈ [0, 1]. Low ci indicates planar/redundant regions and triggers cover￾and-prune, while high ci indicates geometric complexity (edges/corners) and triggers selective splitting. The visua… view at source ↗
Figure 4
Figure 4. Figure 4: Qualitative mesh reconstruction comparison. The figure shows re￾construction results for the quad-easy sequence using estimated poses from the Newer College dataset. Colors represent surface normals mapped to the RGB color space. Our method recovers geometry with high fidelity. PIN-SLAM shows high levels of noise, and Splat-LOAM shows areas of under-reconstruction and over-smoothing [PITH_FULL_IMAGE:figur… view at source ↗
Figure 5
Figure 5. Figure 5: Geometry-aware map budget control driven by covariance￾derived control scores. The map remains compact in large planar areas while preserving fine structures. We attribute this to the fact that the depth consistency loss alone provides weak local surface orientation constraints. Consequently, the tangent plane ori￾entations of target surfels are inadequately reflected, biasing the optimization towards simp… 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

41 extracted references · 1 canonical work pages

  1. [1]

    In: Robotics: science and systems

    Behley, J., Stachniss, C.: Efficient surfel-based slam using 3d laser range data in urban environments. In: Robotics: science and systems. vol. 2018, p. 59 (2018)

    2018

  2. [2]

    IEEE Robotics and Automation Letters 7(2), 2000–2007 (2022)

    Chen, K., Lopez, B.T., Agha-mohammadi, A.a., Mehta, A.: Direct lidar odometry: Fast localization with dense point clouds. IEEE Robotics and Automation Letters 7(2), 2000–2007 (2022)

    2000

  3. [3]

    In: 2019 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)

    Chen, X., Milioto, A., Palazzolo, E., Giguere, P., Behley, J., Stachniss, C.: Suma++: Efficient lidar-based semantic slam. In: 2019 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). pp. 4530–4537. IEEE (2019)

    2019

  4. [4]

    In: 2022 international conference on robotics and automation (ICRA)

    Dellenbach, P., Deschaud, J.E., Jacquet, B., Goulette, F.: Ct-icp: Real-time elastic lidar odometry with loop closure. In: 2022 international conference on robotics and automation (ICRA). pp. 5580–5586. IEEE (2022)

    2022

  5. [5]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision

    Deng, J., Wu, Q., Chen, X., Xia, S., Sun, Z., Liu, G., Yu, W., Pei, L.: Nerf- loam: Neural implicit representation for large-scale incremental lidar odometry and mapping. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 8218–8227 (2023)

    2023

  6. [6]

    In: Conference on Computer Vision and Pattern Recogni- tion (CVPR) (2012)

    Geiger, A., Lenz, P., Urtasun, R.: Are we ready for autonomous driving? the kitti vision benchmark suite. In: Conference on Computer Vision and Pattern Recogni- tion (CVPR) (2012)

    2012

  7. [7]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision

    Giacomini, E., Di Giammarino, L., De Rebotti, L., Grisetti, G., Oswald, M.R.: Splat-loam: Gaussian splatting lidar odometry and mapping. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 27630–27639 (2025)

    2025

  8. [8]

    In: 2025 IEEE/RSJ International Confer- ence on Intelligent Robots and Systems (IROS)

    Guadagnino, T., Mersch, B., Gupta, S., Vizzo, I., Grisetti, G., Stachniss, C.: KISS-SLAM: A Simple, Robust, and Accurate 3D LiDAR SLAM System With Enhanced Generalization Capabilities. In: 2025 IEEE/RSJ International Confer- ence on Intelligent Robots and Systems (IROS). pp. 5363–5370 (2025).https: //doi.org/10.1109/IROS60139.2025.11246613

    doi:10.1109/iros60139.2025.11246613 2025

  9. [9]

    In: IEEE International Conference on Robotics and Automation (ICRA) (2024)

    Gupta, S., Guadagnino, T., Mersch, B., Vizzo, I., Stachniss, C.: Effectively De- tecting Loop Closures using Point Cloud Density Maps. In: IEEE International Conference on Robotics and Automation (ICRA) (2024)

    2024

  10. [10]

    In: European conference on computer vision

    Ha, S., Yeon, J., Yu, H.: Rgbd gs-icp slam. In: European conference on computer vision. pp. 180–197. Springer (2024)

    2024

  11. [11]

    Advanced Intelligent Sys- tems5(7), 2200459 (2023)

    He, D., Xu, W., Chen, N., Kong, F., Yuan, C., Zhang, F.: Point-lio: robust high- bandwidth light detection and ranging inertial odometry. Advanced Intelligent Sys- tems5(7), 2200459 (2023)

    2023

  12. [12]

    In: 2017 IEEE International Conference on Robotics and Automation (ICRA)

    Hsiao, M., Westman, E., Zhang, G., Kaess, M.: Keyframe-based dense planar slam. In: 2017 IEEE International Conference on Robotics and Automation (ICRA). pp. 5110–5117 (2017).https://doi.org/10.1109/ICRA.2017.7989597

    work page doi:10.1109/icra.2017.7989597 2017

  13. [13]

    In: ACM SIGGRAPH 2024 conference papers

    Huang, B., Yu, Z., Chen, A., Geiger, A., Gao, S.: 2d gaussian splatting for geo- metrically accurate radiance fields. In: ACM SIGGRAPH 2024 conference papers. pp. 1–11 (2024)

    2024

  14. [14]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Huang, H., Li, L., Cheng, H., Yeung, S.K.: Photo-slam: Real-time simultaneous localization and photorealistic mapping for monocular stereo and rgb-d cameras. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 21584–21593 (2024)

    2024

  15. [15]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recog- nition

    Keetha, N., Karhade, J., Jatavallabhula, K.M., Yang, G., Scherer, S., Ramanan, D., Luiten, J.: Splatam: Splat track & map 3d gaussians for dense rgb-d slam. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recog- nition. pp. 21357–21366 (2024) Real-Time LiDAR GS SLAM 17

    2024

  16. [16]

    ACM Transactions on Graphics42(4) (July 2023),https://repo-sam.inria.fr/fungraph/3d-gaussian-splatting/

    Kerbl, B., Kopanas, G., Leimk¨ uhler, T., Drettakis, G.: 3d gaussian splatting for real-time radiance field rendering. ACM Transactions on Graphics42(4) (July 2023),https://repo-sam.inria.fr/fungraph/3d-gaussian-splatting/

    2023

  17. [17]

    arXiv preprint arXiv:2507.04004 (2025)

    Lang, X., Lv, J., Tang, K., Li, L., Huang, J., Liu, L., Liu, Y., Zuo, X.: Gaussian-lic2: Lidar-inertial-camera gaussian splatting slam. arXiv preprint arXiv:2507.04004 (2025)

    Pith/arXiv arXiv 2025

  18. [18]

    In: 2025 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)

    Liu, J., Wan, Y., Wang, B., Zheng, C., Lin, J., Zhang, F.: Gs-sdf: Lidar-augmented gaussian splatting and neural sdf for geometrically consistent rendering and recon- struction. In: 2025 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). pp. 19391–19398. IEEE (2025)

    2025

  19. [19]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recog- nition

    Matsuki, H., Murai, R., Kelly, P.H., Davison, A.J.: Gaussian splatting slam. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recog- nition. pp. 18039–18048 (2024)

    2024

  20. [20]

    In: IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) (2017)

    Oleynikova, H., Taylor, Z., Fehr, M., Siegwart, R., Nieto, J.: Voxblox: Incremen- tal 3d euclidean signed distance fields for on-board mav planning. In: IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) (2017)

    2017

  21. [21]

    arXiv preprint arXiv:2507.18344 (2025)

    Pak, G., Cho, H.M., Kim, E.: G2s-icp slam: Geometry-aware gaussian splatting icp slam. arXiv preprint arXiv:2507.18344 (2025)

    Pith/arXiv arXiv 2025

  22. [22]

    arXiv preprint arXiv:2502.05752 (2025)

    Pan, Y., Zhong, X., Jin, L., Wiesmann, L., Popovi´ c, M., Behley, J., Stachniss, C.: Pings: Gaussian splatting meets distance fields within a point-based implicit neural map. arXiv preprint arXiv:2502.05752 (2025)

    Pith/arXiv arXiv 2025

  23. [23]

    IEEE Transactions on Robotics40, 4045–4064 (2024)

    Pan, Y., Zhong, X., Wiesmann, L., Posewsky, T., Behley, J., Stachniss, C.: Pin- slam: Lidar slam using a point-based implicit neural representation for achieving global map consistency. IEEE Transactions on Robotics40, 4045–4064 (2024)

    2024

  24. [24]

    In: 2020 IEEE/RSJ International Conference on Intelligent Robots and Sys- tems (IROS)

    Ramezani, M., Wang, Y., Camurri, M., Wisth, D., Mattamala, M., Fallon, M.: The newer college dataset: Handheld lidar, inertial and vision with ground truth. In: 2020 IEEE/RSJ International Conference on Intelligent Robots and Sys- tems (IROS). pp. 4353–4360 (2020).https://doi.org/10.1109/IROS45743.2020. 9340849

    doi:10.1109/iros45743.2020 2020

  25. [25]

    In: Robotics: science and systems

    Segal, A., Haehnel, D., Thrun, S., et al.: Generalized-icp. In: Robotics: science and systems. vol. 2, p. 435. Seattle, WA (2009)

    2009

  26. [26]

    In: 2018 IEEE/RSJ international conference on intelligent robots and systems (IROS)

    Shan, T., Englot, B.: Lego-loam: Lightweight and ground-optimized lidar odometry and mapping on variable terrain. In: 2018 IEEE/RSJ international conference on intelligent robots and systems (IROS). pp. 4758–4765. IEEE (2018)

    2018

  27. [27]

    In: 2020 IEEE/RSJ international conference on intelligent robots and systems (IROS)

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

    2020

  28. [28]

    IEEE Robotics and Automation Letters9(6), 5935–5942 (2024)

    Song, S., Zhao, J., Huang, K., Lin, J., Ye, C., Feng, T.:N 3-Mapping: Normal guided neural non-projective signed distance fields for large-scale 3d mapping. IEEE Robotics and Automation Letters9(6), 5935–5942 (2024)

    2024

  29. [29]

    International Journal of Robotics Research (2025)

    Tao, Y., Mu˜ noz-Ba˜ n´ on, M.´A., Zhang, L., Wang, J., Fu, L.F.T., Fallon, M.: The ox- ford spires dataset: Benchmarking large-scale lidar-visual localisation, reconstruc- tion and radiance field methods. International Journal of Robotics Research (2025)

    2025

  30. [30]

    In: Proceedings of the IEEE/CVF con- ference on computer vision and pattern recognition

    Turki, H., Ramanan, D., Satyanarayanan, M.: Mega-nerf: Scalable construction of large-scale nerfs for virtual fly-throughs. In: Proceedings of the IEEE/CVF con- ference on computer vision and pattern recognition. pp. 12922–12931 (2022)

    2022

  31. [31]

    Sensors22(3), 1296 (2022) 18 SeungJun Tak et al

    Vizzo, I., Guadagnino, T., Behley, J., Stachniss, C.: Vdbfusion: Flexible and effi- cient tsdf integration of range sensor data. Sensors22(3), 1296 (2022) 18 SeungJun Tak et al

    2022

  32. [32]

    IEEE Robotics and Automation Letters 10(1), 421–428 (2024)

    Xiao, R., Liu, W., Chen, Y., Hu, L.: Liv-gs: Lidar-vision integration for 3d gaussian splatting slam in outdoor environments. IEEE Robotics and Automation Letters 10(1), 421–428 (2024)

    2024

  33. [33]

    In: Proceedings of the IEEE/CVF Inter- national Conference on Computer Vision

    Xie, Y., Huang, Z., Wu, J., Ma, J.: Gs-livm: Real-time photo-realistic lidar-inertial- visual mapping with gaussian splatting. In: Proceedings of the IEEE/CVF Inter- national Conference on Computer Vision. pp. 26869–26878 (2025)

    2025

  34. [34]

    IEEE Transactions on Robotics38(4), 2053–2073 (2022)

    Xu, W., Cai, Y., He, D., Lin, J., Zhang, F.: Fast-lio2: Fast direct lidar-inertial odometry. IEEE Transactions on Robotics38(4), 2053–2073 (2022)

    2053

  35. [35]

    IEEE Robotics and Automation Letters 6(2), 3317–3324 (2021)

    Xu, W., Zhang, F.: Fast-lio: A fast, robust lidar-inertial odometry package by tightly-coupled iterated kalman filter. IEEE Robotics and Automation Letters 6(2), 3317–3324 (2021)

    2021

  36. [36]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Yan, C., Qu, D., Xu, D., Zhao, B., Wang, Z., Wang, D., Li, X.: Gs-slam: Dense vi- sual slam with 3d gaussian splatting. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 19595–19604 (2024)

    2024

  37. [37]

    In: 2022 IEEE In- ternational Symposium on Mixed and Augmented Reality (ISMAR)

    Yang, X., Li, H., Zhai, H., Ming, Y., Liu, Y., Zhang, G.: Vox-fusion: Dense tracking and mapping with voxel-based neural implicit representation. In: 2022 IEEE In- ternational Symposium on Mixed and Augmented Reality (ISMAR). pp. 499–507. IEEE (2022)

    2022

  38. [38]

    IEEE Robotics and Automation Letters (2026)

    Yeon, J., Ha, S., Yu, H.: Gso-slam: Bidirectionally coupled gaussian splatting and direct visual odometry. IEEE Robotics and Automation Letters (2026)

    2026

  39. [39]

    In: Robotics: Science and systems

    Zhang, J., Singh, S., et al.: Loam: Lidar odometry and mapping in real-time. In: Robotics: Science and systems. vol. 2, pp. 1–9. Berkeley, CA (2014)

    2014

  40. [40]

    In: 2023 IEEE Inter- national Conference on Robotics and Automation (ICRA)

    Zhong, X., Pan, Y., Behley, J., Stachniss, C.: Shine-mapping: Large-scale 3d map- ping using sparse hierarchical implicit neural representations. In: 2023 IEEE Inter- national Conference on Robotics and Automation (ICRA). pp. 8371–8377. IEEE (2023)

    2023

  41. [41]

    In: 2021 IEEE international conference on robotics and automation (ICRA)

    Zhou, L., Wang, S., Kaess, M.:π-lsam: Lidar smoothing and mapping with planes. In: 2021 IEEE international conference on robotics and automation (ICRA). pp. 5751–5757. IEEE (2021)

    2021