pith. sign in

arxiv: 2512.09447 · v2 · pith:TRZKUZYFnew · submitted 2025-12-10 · 💻 cs.RO · cs.CV

Query-Calibrated Segmental Admission for Descriptor-Agnostic LiDAR Loop Closure in Repetitive Environments

Jaehyun Kim , Seungwon Choi , Wonseok Kang , Tae-Wan Kim This is my paper

Pith reviewed 2026-05-21 18:01 UTC · model grok-4.3

classification 💻 cs.RO cs.CV
keywords LiDARloop closurerepetitive environmentsSLAMaliasingpose graphloop admissionG-ICP
0
0 comments X

The pith

QCSA is a descriptor-agnostic policy that admits sparser yet more precise loop factors for LiDAR SLAM in repetitive environments.

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

The paper aims to solve the issue of aliased loop candidates in repetitive environments that can destabilize pose-graph optimization in LiDAR SLAM. It introduces Query-Calibrated Segmental Admission, which scores short descriptor segments against hard negatives and calibrates which hypotheses proceed to G-ICP validation. This leads to fewer inserted factors with higher precision across multiple descriptors. A reader would care because it results in more stable graphs and improved worst-case trajectory accuracy without sacrificing mean performance.

Core claim

The central discovery is that a query-calibrated segmental admission policy can serve as an effective, descriptor-agnostic layer for admitting loop factors in aliasing-heavy environments. By calibrating segment hypotheses at the query level before geometric validation, QCSA reduces inserted loop factors by 3.8 times, increases factor precision from 0.542 to 0.717, and reduces worst-sequence ATE from 1.064 m to 0.778 m on the SNULib dataset across seven LiDAR descriptor families, while generalizing to HeLiPR without retuning.

What carries the argument

Query-Calibrated Segmental Admission (QCSA), which scores short descriptor segments against hard negatives to decide which query-level hypotheses reach G-ICP validation and insertion as loop factors.

If this is right

  • Inserted loop factors decrease by 3.8 times while factor precision increases to 0.717.
  • Worst-sequence absolute trajectory error drops from 1.064 m to 0.778 m compared to dense Top1+G-ICP.
  • Mean absolute trajectory error remains comparable to or better than odometry-only references.
  • Under a matched factor budget, trajectory error is lower than SeqSLAM and sparse Top1+G-ICP.
  • Hard-negative admissions are suppressed on HeLiPR overlap routes with no route-specific tuning.

Where Pith is reading between the lines

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

  • The segmental calibration may extend to other sensor modalities facing similar repetition-induced aliasing.
  • Future work could explore adaptive calibration thresholds based on environment repetition metrics.
  • Combining QCSA with learned descriptors might yield even higher precision gains.
  • This approach highlights the value of query-specific filtering over global descriptor matching in SLAM.

Load-bearing premise

The fixed calibration process is assumed to transfer effectively across different LiDAR descriptors and new environments without any retuning or environment-specific adjustments.

What would settle it

A new test on an eighth LiDAR descriptor or a different repetitive dataset where the precision falls below 0.542 or the worst ATE exceeds that of the dense baseline.

Figures

Figures reproduced from arXiv: 2512.09447 by Jaehyun Kim, Seungwon Choi, Tae-Wan Kim, Wonseok Kang.

Figure 1
Figure 1. Figure 1: The book repository at Seoul National University [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Samples from the dataset. LiDAR scans and camera images for frame 318 (a) and frame 677 in Sequence 04. [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Proposed Seq-SPRT loop-verification pipeline. Left: global-descriptor retrieval forms a candidate set for a query [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Pseudo ground-truth trajectories (top) and single-frame precision-recall (PR) curves (bottom) for all five library [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗
read the original abstract

Structurally repetitive environments produce visually plausible but aliased LiDAR loop candidates that can destabilize pose-graph optimization when admitted as loop factors. We propose Query-Calibrated Segmental Admission (QCSA), a descriptor-agnostic sparse loop-admission policy for graph-stability-oriented insertion. The policy scores short descriptor segments against hard negatives, calibrates which query-level segment hypotheses reach geometry, and inserts representative pairs validated by Generalized Iterative Closest Point (G-ICP). We evaluate it on the SNU Library Dataset (SNULib) and HeLiPR overlap routes. Aggregated over seven LiDAR descriptor families on SNULib, QCSA reduces inserted loop factors by 3.8 times, raises factor precision from 0.542 to 0.717, and sharply lowers false admissions per query group. With this sparser graph, it maintains comparable mean absolute trajectory error (ATE) and substantially reduces worst-sequence ATE versus dense Top1+G-ICP, from 1.064 to 0.778 m. The aggregate mean and worst-sequence ATE remain lower than the odometry-only reference. Under a matched factor budget, QCSA also attains lower trajectory error than SeqSLAM and sparse Top1+G-ICP selections. Fixed-transfer validation on HeLiPR, with no route-specific tuning, likewise suppresses hard-negative admissions. These results support the proposed admission layer for aliasing-heavy simultaneous localization and mapping (SLAM). Our implementation and dataset will be released at: https://github.com/wanderingcar/snu_library_dataset.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The manuscript proposes Query-Calibrated Segmental Admission (QCSA), a descriptor-agnostic sparse loop-admission policy for LiDAR SLAM in repetitive environments. Short descriptor segments are scored against hard negatives to calibrate which query-level hypotheses proceed to G-ICP validation and insertion as loop factors. On the SNULib dataset, aggregated over seven LiDAR descriptor families, QCSA reduces inserted loop factors by 3.8 times, raises factor precision from 0.542 to 0.717, and lowers worst-sequence ATE versus dense Top1+G-ICP from 1.064 m to 0.778 m while remaining below the odometry-only baseline. Fixed-parameter transfer to HeLiPR overlap routes is reported to suppress hard-negative admissions without route-specific retuning.

Significance. If the calibration step generalizes across descriptor families and environments without retuning, the method offers a practical, sparsity-oriented layer that improves pose-graph stability in aliasing-heavy settings. The multi-descriptor evaluation, explicit comparison to SeqSLAM and sparse Top1+G-ICP under matched factor budgets, and planned release of code plus the SNULib dataset are concrete strengths that would support adoption in real-world SLAM pipelines.

major comments (3)
  1. [Abstract / Evaluation] Abstract and evaluation summary: the central descriptor-agnostic claim rests on fixed-transfer validation to HeLiPR with no route-specific tuning, yet no details are provided on how calibration thresholds were chosen, how many hard negatives were sampled, or what segment lengths were used; without this, it is impossible to verify that the reported 3.8× reduction and precision lift from 0.542 to 0.717 are insensitive to descriptor-specific aliasing statistics.
  2. [Results / Tables] Results presentation: aggregate metrics are given without per-descriptor breakdowns, error bars, or ablation on the calibration thresholds; this directly affects the load-bearing assertion that performance holds across seven descriptor families and that the sparser graph still yields lower worst-sequence ATE (1.064 m to 0.778 m).
  3. [Methods / QCSA policy] Methods description of the admission policy: the scoring of short segments against hard negatives and the subsequent calibration step that decides which hypotheses reach G-ICP are presented at a high level; concrete definitions of the scoring function, negative-sampling procedure, and threshold-setting rule are required to substantiate generalization without retuning.
minor comments (2)
  1. [Abstract] The phrase 'sharply lowers false admissions per query group' is used in the abstract but the exact definition of this metric and how it is computed across query groups is not stated.
  2. [Abstract / Conclusion] The manuscript states that implementation and dataset will be released; the GitHub link should be included in the camera-ready version with a clear license and reproduction instructions.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the thorough review and constructive suggestions. We address each of the major comments in detail below, providing clarifications and indicating the revisions we will make to strengthen the manuscript.

read point-by-point responses

  1. Referee: [Abstract / Evaluation] Abstract and evaluation summary: the central descriptor-agnostic claim rests on fixed-transfer validation to HeLiPR with no route-specific tuning, yet no details are provided on how calibration thresholds were chosen, how many hard negatives were sampled, or what segment lengths were used; without this, it is impossible to verify that the reported 3.8× reduction and precision lift from 0.542 to 0.717 are insensitive to descriptor-specific aliasing statistics.

    Authors: We agree that the abstract and evaluation would benefit from more details on the calibration procedure to support the descriptor-agnostic claim. In the revised manuscript, we will include a dedicated paragraph in the methods section describing how the calibration thresholds were selected through validation on SNULib, the number of hard negatives sampled, and the segment lengths used. This will allow verification that the performance gains are robust across descriptors. The fixed transfer to HeLiPR is already reported without retuning, and we will emphasize this with the added details. revision: yes

  2. Referee: [Results / Tables] Results presentation: aggregate metrics are given without per-descriptor breakdowns, error bars, or ablation on the calibration thresholds; this directly affects the load-bearing assertion that performance holds across seven descriptor families and that the sparser graph still yields lower worst-sequence ATE (1.064 m to 0.778 m).

    Authors: We will revise the results section to include per-descriptor breakdowns in an expanded table, add error bars where applicable from repeated experiments, and provide an ablation study on the calibration thresholds to demonstrate the robustness of the chosen parameters. These additions will directly support the claims regarding consistency across descriptor families. revision: yes

  3. Referee: [Methods / QCSA policy] Methods description of the admission policy: the scoring of short segments against hard negatives and the subsequent calibration step that decides which hypotheses reach G-ICP are presented at a high level; concrete definitions of the scoring function, negative-sampling procedure, and threshold-setting rule are required to substantiate generalization without retuning.

    Authors: We concur that a more concrete description is needed. We will expand the methods to define the scoring function explicitly (as the minimum score against hard negatives), detail the negative-sampling procedure (random selection from distant map segments), and specify the threshold-setting rule (percentile-based on calibration queries). Pseudocode will also be added for clarity. This will substantiate the generalization claims. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical policy with external validation

full rationale

The paper introduces QCSA as an algorithmic admission policy that scores segments against hard negatives and applies fixed calibration before G-ICP validation. All central performance claims (3.8× factor reduction, precision lift from 0.542 to 0.717, ATE improvement from 1.064 m to 0.778 m) are obtained via direct empirical comparison against independent baselines (dense Top1+G-ICP, SeqSLAM, odometry-only) on SNULib and fixed-transfer testing on HeLiPR. No equations, fitted parameters, or self-citations are shown to define the evaluation metrics or the reported gains by construction; the calibration thresholds are presented as descriptor-agnostic and route-untuned, with results measured against external benchmarks rather than internal re-use of the same data.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The method relies on standard ICP assumptions and the existence of identifiable hard-negative segments; no new physical entities or ad-hoc constants are introduced in the abstract.

axioms (1)
  • domain assumption Generalized ICP provides reliable geometric validation for selected segment pairs
    The admission policy ultimately inserts only pairs that pass G-ICP; this is invoked in the abstract description of the pipeline.

pith-pipeline@v0.9.0 · 5825 in / 1327 out tokens · 53845 ms · 2026-05-21T18:01:34.299748+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

39 extracted references · 39 canonical work pages

  1. [1]

    Thrun, W

    S. Thrun, W. Burgard, and D. Fox,Probabilistic Robotics. Cambridge, MA, USA: MIT Press, 2005

    work page 2005

  2. [2]

    Past, present, and future of simultaneous localization and mapping: Toward the robust-perception age,

    C. Cadena, L. Carlone, H. Carrillo, Y . Latif, D. Scaramuzza, J. Neira, I. Reid, and J. J. Leonard, “Past, present, and future of simultaneous localization and mapping: Toward the robust-perception age,”IEEE Transactions on Robotics, vol. 32, no. 6, pp. 1309–1332, 2016

    work page 2016

  3. [3]

    Scan Context++: Structural place recognition robust to rotation and lateral variations in urban environ- ments,

    G. Kim, S. Choi, and A. Kim, “Scan Context++: Structural place recognition robust to rotation and lateral variations in urban environ- ments,”IEEE Trans. Robot., vol. 38, no. 3, pp. 1856–1874, 2022

    work page 2022

  4. [4]

    LiDAR Iris for loop-closure detection,

    Y . Wang, Z. Sun, C.-Z. Xu, S. E. Sarma, J. Yang, and H. Kong, “LiDAR Iris for loop-closure detection,” inProc. IEEE/RSJ Int. Conf. on Intelligent Robots and Systems (IROS), 2020, pp. 5769–5775

    work page 2020

  5. [5]

    FreSCo: Frequency-domain Scan Context for robust LiDAR-based place recognition with translation and rotation invariance,

    Y . Fan, X. Du, L. Luo, and J. Shen, “FreSCo: Frequency-domain Scan Context for robust LiDAR-based place recognition with translation and rotation invariance,” inProc. Int. Conf. on Control, Automation, Robotics and Vision (ICARCV), 2022, pp. 606–611

    work page 2022

  6. [6]

    BTC: A binary and triangle combined descriptor for 3d place recognition,

    C. Yuan, J. Lin, Z. Liu, H. Wei, X. Hong, and F. Zhang, “BTC: A binary and triangle combined descriptor for 3d place recognition,” IEEE Trans. Robot., vol. 40, no. 6, pp. 1580–1599, 2024

    work page 2024

  7. [7]

    NDT-Map-Code: A 3d global descriptor for real-time loop closure detection in lidar slam,

    L. Liao, W. Yan, L. Sun, X. Bai, Z. Youet al., “NDT-Map-Code: A 3d global descriptor for real-time loop closure detection in lidar slam,” IEEE Trans. Instrum. Meas., vol. 73, no. 8502115, pp. 1–13, 2024

    work page 2024

  8. [8]

    Effectively detecting loop closures using point cloud density maps,

    S. Gupta, I. Vizzo, J. Behley, and C. Stachniss, “Effectively detecting loop closures using point cloud density maps,”IEEE Robotics and Automation Letters (RA-L), vol. 8, no. 2, pp. 1029–1036, 2023

    work page 2023

  9. [9]

    Are we ready for autonomous driving? the kitti vision benchmark suite,

    A. Geiger, P. Lenz, C. Stiller, and R. Urtasun, “Are we ready for autonomous driving? the kitti vision benchmark suite,” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2012, pp. 3354–3361

    work page 2012

  10. [10]

    4Seasons: A cross-season dataset for multi-weather SLAM in autonomous driving,

    P. Wenzel, L. Stumberg, M. Sch ¨onbein, and D. Cremers, “4Seasons: A cross-season dataset for multi-weather SLAM in autonomous driving,” 2020

    work page 2020

  11. [11]

    MulRan: Multimodal range dataset for urban place recognition,

    G. Kim and A. Kim, “MulRan: Multimodal range dataset for urban place recognition,” inIEEE International Conference on Robotics and Automation (ICRA), 2020, pp. 6246–6253

    work page 2020

  12. [12]

    The newer college dataset: Handheld LiDAR, airborne and ground truth data for research in autonomous navigation,

    M. Ramezani, D. Wisth, M. Gadd, D. De Martini, P. Newman, I. Pos- ner, and J. J. Leonard, “The newer college dataset: Handheld LiDAR, airborne and ground truth data for research in autonomous navigation,” inIEEE International Conference on Robotics and Automation (ICRA), 2020, pp. 5839–5845

    work page 2020

  13. [13]

    1 year, 1000 km: The Oxford RobotCar Dataset,

    W. Maddern, G. Pascoe, C. Linegar, and P. Newman, “1 year, 1000 km: The Oxford RobotCar Dataset,”The International Journal of Robotics Research, vol. 36, no. 1, pp. 3–15, 2017

    work page 2017

  14. [14]

    Detecting loop closure with scene sequences,

    K. L. Ho and P. Newman, “Detecting loop closure with scene sequences,”International Journal of Computer Vision, vol. 74, no. 3, pp. 261–286, 2007

    work page 2007

  15. [15]

    Robust loop closing over time,

    Y . Latif, C. C. Lerma, and J. Neira, “Robust loop closing over time,” inProceedings of Robotics: Science and Systems, Sydney, Australia, July 2012

    work page 2012

  16. [16]

    Switchable constraints vs. max-mixture models vs. RRR – a comparison of three approaches to robust pose graph SLAM,

    N. S ¨underhauf and P. Protzel, “Switchable constraints vs. max-mixture models vs. RRR – a comparison of three approaches to robust pose graph SLAM,” inProceedings of the IEEE International Conference on Robotics and Automation (ICRA), 2013, pp. 5198–5203

    work page 2013

  17. [17]

    A brief survey of loop closure detection: A case for rethinking evaluation of intelligent systems,

    S. B. Nashed, “A brief survey of loop closure detection: A case for rethinking evaluation of intelligent systems,” inNeurIPS 2020 Workshop on Machine Learning Retrospectives, Surveys & Meta- Analyses (ML-RSA), 2020

    work page 2020

  18. [18]

    Seqslam: Visual route-based naviga- tion for sunny summer days and stormy winter nights,

    M. J. Milford and G. F. Wyeth, “Seqslam: Visual route-based naviga- tion for sunny summer days and stormy winter nights,” inProc. IEEE Intl. Conf. on Robotics and Automation (ICRA), 2012, pp. 1643–1649

    work page 2012

  19. [19]

    Scan Context: Egocentric spatial descriptor for place recognition within 3d point cloud map,

    G. Kim and A. Kim, “Scan Context: Egocentric spatial descriptor for place recognition within 3d point cloud map,” inProc. IEEE/RSJ Int. Conf. on Intelligent Robots and Systems (IROS), 2018, pp. 4802–4809

    work page 2018

  20. [20]

    M2dp: A novel 3d point cloud descriptor and its application in loop closure detection,

    L. He, X. Wang, and H. Zhang, “M2dp: A novel 3d point cloud descriptor and its application in loop closure detection,” inProc. IEEE/RSJ Intl. Conf. on Intelligent Robots and Systems (IROS), 2016

    work page 2016

  21. [21]

    Segmatch: Segment based place recognition in 3d point clouds,

    R. Dub ´e, D. Dugas, E. Stumm, J. Nieto, R. Siegwart, and C. Cadena, “Segmatch: Segment based place recognition in 3d point clouds,” in Proc. IEEE Intl. Conf. on Robotics and Automation (ICRA), 2017, pp. 5266–5272

    work page 2017

  22. [22]

    Pointnetvlad: Deep point cloud based retrieval for large-scale place recognition,

    M. A. Uy and G. H. Lee, “Pointnetvlad: Deep point cloud based retrieval for large-scale place recognition,” inProc. IEEE/CVF Conf. on Computer Vision and Pattern Recognition (CVPR), 2018, pp. 4470– 4479

    work page 2018

  23. [23]

    Lpd-net: 3d point cloud learning for large-scale place recognition and environment analysis,

    Z. Liu, S. Zhou, C. Suo, P. Yin, W. Chen, H. Wang, H. Li, and Y .-H. Liu, “Lpd-net: 3d point cloud learning for large-scale place recognition and environment analysis,” inProc. IEEE/CVF Intl. Conf. on Computer Vision (ICCV), 2019

    work page 2019

  24. [24]

    Overlapnet: A siamese network for computing lidar scan similar- ity with applications to loop closing and localization,

    X. Chen, T. L ¨abe, A. Milioto, T. R ¨ohling, J. Behley, and C. Stachniss, “Overlapnet: A siamese network for computing lidar scan similar- ity with applications to loop closing and localization,”Autonomous Robots, vol. 46, no. 1, pp. 61–81, 2022

    work page 2022

  25. [25]

    Overlaptrans- former: An efficient and yaw-angle-invariant transformer network for lidar-based place recognition,

    J. Ma, J. Zhang, J. Xu, R. Ai, W. Gu, and X. Chen, “Overlaptrans- former: An efficient and yaw-angle-invariant transformer network for lidar-based place recognition,”IEEE Robotics and Automation Letters, vol. 7, no. 3, pp. 6958–6965, 2022

    work page 2022

  26. [26]

    Seqot: A spatial-temporal transformer network for place recognition using sequential lidar data,

    J. Ma, X. Chen, J. Xu, and G. Xiong, “Seqot: A spatial-temporal transformer network for place recognition using sequential lidar data,” IEEE Transactions on Industrial Electronics, vol. 70, no. 8, pp. 8225– 8234, 2023

    work page 2023

  27. [27]

    Fab-map: Probabilistic localization and mapping in the space of appearance,

    M. Cummins and P. Newman, “Fab-map: Probabilistic localization and mapping in the space of appearance,”The International Journal of Robotics Research, vol. 27, no. 6, pp. 647–665, 2008

    work page 2008

  28. [28]

    Randomized ransac with sequential prob- ability ratio test,

    J. Matas and O. Chum, “Randomized ransac with sequential prob- ability ratio test,” inProceedings of the 10th IEEE International Conference on Computer Vision (ICCV), 2005

    work page 2005

  29. [29]

    Sequential tests of statistical hypotheses,

    A. Wald, “Sequential tests of statistical hypotheses,”Annals of Math- ematical Statistics, vol. 16, no. 2, pp. 117–186, 1945

    work page 1945

  30. [30]

    New York: John Wiley & Sons, 1947

    ——,Sequential Analysis. New York: John Wiley & Sons, 1947

    work page 1947

  31. [31]

    Optimum character of the sequential probability ratio test,

    A. Wald and J. Wolfowitz, “Optimum character of the sequential probability ratio test,”Annals of Mathematical Statistics, vol. 19, no. 3, pp. 326–339, 1948

    work page 1948

  32. [32]

    Siegmund,Sequential Analysis: Tests and Confidence Intervals, ser

    D. Siegmund,Sequential Analysis: Tests and Confidence Intervals, ser. Springer Series in Statistics. New York: Springer, 1985

    work page 1985

  33. [33]

    A. G. Tartakovsky, I. V . Nikiforov, and M. Basseville,Sequential Analysis: Hypothesis Testing and Changepoint Detection. Boca Raton, FL: CRC Press, 2014

    work page 2014

  34. [34]

    LOAM: Lidar odometry and mapping in real- time,

    J. Zhang and S. Singh, “LOAM: Lidar odometry and mapping in real- time,” inProceedings of Robotics: Science and Systems (RSS), 2014

    work page 2014

  35. [35]

    Large-scale lidar consistent mapping using hierarchical lidar bundle adjustment,

    X. Liu, Z. Liu, F. Kong, and F. Zhang, “Large-scale lidar consistent mapping using hierarchical lidar bundle adjustment,”IEEE Robotics and Automation Letters, vol. 8, no. 3, pp. 1523–1530, 2023

    work page 2023

  36. [36]

    Efficient and consistent bundle adjustment on lidar point clouds,

    Z. Liu, X. Liu, and F. Zhang, “Efficient and consistent bundle adjustment on lidar point clouds,”IEEE Transactions on Robotics, vol. 39, no. 6, pp. 4366–4386, 2023

    work page 2023

  37. [37]

    Flirt: Interest regions for 2d range data with applications to robot navigation,

    G. D. Tipaldi, M. Braun, and K. O. Arras, “Flirt: Interest regions for 2d range data with applications to robot navigation,” inExperi- mental Robotics: The 12th International Symposium on Experimental Robotics. Springer, pp. 695–710

    work page

  38. [38]

    Place recognition using regional point descrip- tors for 3d mapping,

    M. Bosse and R. Zlot, “Place recognition using regional point descrip- tors for 3d mapping,” inField and Service Robotics: Results of the 7th International Conference. Springer, 2010, pp. 195–204

    work page 2010

  39. [39]

    Orb-slam: A versatile and accurate monocular slam system,

    R. Mur-Artal, J. M. M. Montiel, and J. D. Tardos, “Orb-slam: A versatile and accurate monocular slam system,”IEEE transactions on robotics, vol. 31, no. 5, pp. 1147–1163, 2015

    work page 2015