Pith. sign in

REVIEW 3 major objections 5 minor 58 references

The paper argues that RGB-only SLAM stays reliable in low light only when an IMU and global optimization are both present, based on a six-system benchmark across five LaMARia sequences.

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 ↗

On five LaMARia sequences, only Kimera-VIO tracked all to completion; DPVO/DPV-SLAM never lost tracking but had roughly 100 m absolute error.

T0 review reviewed 2026-08-01 challenge →

load-bearing objection Useful, honestly-reported low-light SLAM benchmark whose abstract claim outruns its own tables. the 3 major comments →

arxiv 2607.17699 v1 pith:IY33I2F7 submitted 2026-07-20 cs.RO cs.CV

SLAM in Low-Light Environments: Project Report

Oleh Basystyi , Anna Stasyshyn , Oleksandr Kosovan , Yaroslav Prytula This is my paper
classification cs.RO cs.CV
keywords visual SLAMlow-light navigationvisual-inertial odometrybenchmarkingabsolute trajectory errorlearning-based odometryUGV autonomyLaMARia dataset
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 reading

This project report tries to establish how far standard RGB-camera SLAM can be pushed in low-light conditions. By benchmarking six systems across feature-based, direct, filter-based, and learning-based paradigms on five LaMARia sequences, the authors find that only Kimera-VIO—an inertial, optimization-based visual-inertial odometry—tracks all sequences to completion with low local error. The learning-based monocular systems DPVO and DPV-SLAM never lose tracking but accumulate absolute errors near 100 m under low light, while ORB-SLAM3, DSO, and OpenVINS fail or diverge on most hard and dark sequences. The authors conclude that stable low-light localization with RGB cameras requires both inertial fusion and global optimization, and that closing the remaining gap will need low-light-specific learned front-ends or complementary sensors.

Core claim

Across the five tested LaMARia sequences, only Kimera-VIO—a visual-inertial odometry with a tightly coupled optimization backend—completed every track. It held the lowest relative pose error, but its absolute error grew steadily because the evaluated configuration has no loop closure. The learning-based monocular systems DPVO and DPV-SLAM never lost tracking, yet under low light their absolute trajectory error reached roughly 100 m. ORB-SLAM3, DSO, and OpenVINS either failed outright or diverged on most hard and dark sequences. The authors read this as evidence that stable RGB-only low-light localization requires both inertial fusion and global optimization; absent either, the system either

What carries the argument

The comparison itself is the apparatus: six representative systems spanning feature-based, direct, filter-based, and learning-based paradigms, run on five LaMARia sequences of escalating difficulty and illumination. The load-bearing comparison is between Kimera-VIO (IMU plus optimization backend) and DPVO/DPV-SLAM (monocular, inertial-free, learned patch tracking), which isolates the contributions of inertial fusion and global optimization. ATE and RPE metrics, supplemented by control-point recall, separate local motion accuracy from global drift.

Load-bearing premise

The two low-light sequences are treated as representative of real UGV darkness, yet part of their degradation is synthetic and does not model real camera exposure adaptation; if that assumption fails, the observed ranking may not transfer to deployed platforms.

What would settle it

Capture the same LaMARia routes at night with a real camera whose exposure and gain are allowed to adapt, then run the six systems. If a monocular method such as DPVO or DPV-SLAM achieves absolute trajectory error comparable to Kimera-VIO, or if Kimera-VIO loses tracking, the claim that both IMU and global optimization are necessary would be contradicted. A simpler check: if adding a single known metric scale anchor removes most of the ~100 m error from DPVO, then the failure is scale observability rather than a general low-light limitation.

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

If this is right

  • A UGV relying on RGB-only monocular SLAM at night should expect either tracking loss (feature-based, direct, filter-based) or roughly 100 m scale/drift error (learning-based) unless an IMU and an optimization backend are added.
  • Visual-inertial odometry alone is not enough: Kimera-VIO's absolute error grows with trajectory length because loop closure is absent, so full SLAM with loop closure is the natural next step.
  • Because DPV-SLAM's loop-closure backend barely improved over DPVO's odometry on dark sequences, appearance-based loop detection under low light is a bottleneck that must be solved for global consistency.
  • The gap will not close by algorithmic tuning alone; the authors argue that learned low-light front-ends or complementary sensors such as LiDAR, thermal, or event cameras are needed.

Where Pith is reading between the lines

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

  • The paper leaves implicit that the DPVO/DPV-SLAM ~100 m ATE may be largely a scale-observability problem: if a metric anchor such as a ground plane or a single known dimension were provided, the monocular drift might drop sharply without requiring an IMU—this is testable.
  • Because the low-light degradation is partly synthetic and excludes camera exposure adaptation, real night footage could shift the rankings; a natural next experiment is the same benchmark on physically captured low-light sequences with auto-gain and auto-exposure.
  • The RPE/ATE split suggests a modular recipe: a learned patch-based front-end for tracking robustness plus an IMU for scale plus pose-graph loop closure for global consistency would likely beat any single paradigm.
  • For mapping, the surveyed low-light 3DGS/NeRF work implies the bottleneck may shift from tracking to dense reconstruction; a low-light SLAM benchmark should therefore also report map quality, not just trajectory error.
Share X Bluesky LinkedIn Reddit HN

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 / 5 minor

Summary. The paper benchmarks six visual(-inertial) SLAM systems (ORB-SLAM3, DSO, Kimera-VIO, OpenVINS, DPVO, DPV-SLAM) on five LaMARia sequences of varying difficulty and illumination, reporting ATE, RPE, and control-point recall (CP@1m, R@5m). The results show that Kimera-VIO is the only system to track all five sequences to completion with comparatively low ATE and RPE, while DPVO and DPV-SLAM never lose tracking but incur roughly 100 m ATE on the low-light sequences, and the classical feature-based, direct, and filter-based systems fail or diverge on most harder sequences. The abstract and conclusions state that RGB-only SLAM maintains stable low-light tracking only when both inertial fusion and global optimization are present, and the paper discusses paradigm-specific failure modes and suggests learned front-ends or complementary sensing as future directions.

Significance. If the reported measurements are reproducible, the paper provides a useful empirical comparison of six SLAM paradigms under degraded illumination on the LaMARia benchmark. Strengths include the head-to-head DPVO vs DPV-SLAM comparison as an ablation of the global backend, the use of control-point metrics in addition to ATE/RPE, and the candid, detailed discussion of failure modes per paradigm. The paper also explicitly discloses several limitations. However, the central conclusion as stated is not supported by the paper's own data, which substantially reduces the contribution's scientific value unless the claim is revised. With a corrected interpretation, the work would be a modest but useful datapoint for the low-light SLAM community.

major comments (3)
  1. [Abstract; §5.5, Table 1] The abstract's claim that RGB-only SLAM maintains stable low-light tracking 'only when both inertial fusion and global optimization are present' is contradicted by the reported data. Table 1 shows DPVO (no IMU, no global optimization) and DPV-SLAM (no IMU) track sequence_4_10 and sequence_4_11 to completion, while Kimera-VIO, which has no loop closure / global optimization, achieves the lowest ATE there. The data support a weaker conclusion: learned patch-based front-ends preserve tracking continuity, while metric accuracy is achieved only by the IMU-equipped optimization-based system. The 'only when' necessity claim should be removed or rephrased as a correlation, and the text should distinguish tracking continuity from metric accuracy.
  2. [RQ1 (Section 2); Abstract; Section 7] The key term 'stable' is used with at least two meanings. RQ1 asks for 'stable and sufficiently accurate localization', but the abstract/conclusion use 'stable low-light tracking' without definition. If 'stable' means tracking continuity, the data contradict the abstract. If 'stable' means accurate enough for UGV navigation, an explicit threshold (e.g., ATE below some value) is required, and the conclusion must state that it concerns metric quality rather than tracking continuity. As written, the central conclusion is unfalsifiable due to this ambiguity. Please define 'stable' operationally and rephrase the conclusion accordingly.
  3. [Section 6 (Limitations)] Section 6 discloses that 'the low-light degradation in part of the data is synthetic and does not fully model camera exposure adaptation.' This is load-bearing for the external validity of the central claim: the observed 100 m ATE of DPVO/DPV-SLAM and the failures of ORB-SLAM3/DSO/OpenVINS may be partly artifacts of the synthetic darkening rather than representative of a real camera with auto-exposure and hardware noise. The abstract and conclusion should either state the result as conditional on synthetic degradation or restrict the generalization to the tested sequences. Please add this qualifier to the conclusion and, if possible, validate at least one sequence with naturally low-light capture.
minor comments (5)
  1. [Section 5.2 / Appendix B.2] In the paragraph after the RPE definition, 'We discuss and report them in the appendix' is vague. Consider explicitly naming CP@1m and R@5m in the main text so the reader knows what is being deferred.
  2. [Appendix B.2, Table 4] The text 'The strict 1 msfull column' appears to be missing a space ('1 ms full'). Also, the terms 'assoc' and 'cov' are used without definitions in the caption; please define them and explain the relationship 'cov≈assoc' in the text.
  3. [Tables 1–3] The 'fail' entries are used with two different meanings: in Tables 1–2, 'fail' means no usable track, while in Table 3 it means 'fewer than 3 control points triangulated'. This should be clarified in the table captions or footnotes.
  4. [General] The paper evaluates each system with a single run per sequence and reports no repeated-run statistics. Since the conclusion involves comparative rankings, reporting at least mean/variance over 2–3 runs (or a statement that single runs are standard for these systems) would strengthen the empirical claims.
  5. [References] Several references are dated 2025/2026 preprints (e.g., Deng et al. 2026, Liu et al. 2026, Siméoni et al. 2025). Please mark them clearly as preprints or add publication status if available.

Circularity Check

0 steps flagged

No circularity: empirical benchmark with standard external metrics; conclusions are inductive generalizations from the data.

full rationale

This is an empirical benchmark paper. The derivation chain is: select six published SLAM systems, run them on fixed LaMARia sequences, compute standard ATE/RPE and LaMARia control-point metrics, then generalize from the resulting tables. No model parameter is fitted to the target conclusion; no equation in the paper defines the claimed result in terms of the metrics; and no load-bearing claim rests on the authors' own prior work (the reference list contains no papers by Basystyi, Stasyshyn, Kosovan, or Prytula). The only self-referential elements are disclosed limitations: the low-light degradation is partly synthetic and 'does not fully model camera exposure adaptation,' and monocular systems are compared after Sim(3) alignment. These are validity caveats, not circular reductions. The abstract's strong claim that stable low-light tracking requires both inertial fusion and global optimization is contestable on the paper's own tables (DPVO and DPV-SLAM track all low-light sequences without an IMU), but an over-broad inductive generalization is a correctness/validity concern, not circular reasoning. Score 0.

Axiom & Free-Parameter Ledger

0 free parameters · 3 axioms · 0 invented entities

The paper introduces no fitted free parameters and no new theoretical entities. It relies on the domain assumptions listed above: the representativeness of the selected sequences, the comparability of metrics across paradigms after Sim(3) alignment, and the adequacy of the undocumented system configurations.

axioms (3)
  • domain assumption The five selected LaMARia sequences are representative of low-light UGV operating conditions.
    The paper selects a subset for computational reasons and notes that some low-light degradation is synthetic (Section 6), so this representativeness is assumed but not established.
  • domain assumption ATE and RPE computed after Sim(3) alignment are sufficient to compare monocular, stereo, and visual-inertial systems fairly.
    The paper acknowledges 'monocular systems are only comparable to the metric ones after a Sim(3) alignment' (Section 6), which assumes that the scale ambiguity is the only difference affecting these metrics.
  • domain assumption The off-the-shelf configurations of the six systems reflect standard best practice.
    No parameter settings or tuning details are provided; the results depend on the particular configurations chosen, which are not documented enough to assess.

reviewed 2026-08-01 · how reviews work

0 comments
Cite this review

Pith. "Pith review of SLAM in Low-Light Environments: Project Report." pith.science (2026). https://pith.science/paper/IY33I2F7

@misc{pith2026260717699,
  author       = {Pith},
  title        = {Pith review of: SLAM in Low-Light Environments: Project Report},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/IY33I2F7}},
  note         = {Machine review of arXiv:2607.17699}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Simultaneous localization and mapping (SLAM) is one of the fundamental problems in robotics, as it enables autonomous operations in real-world scenarios. Under low illumination, reduced contrast, sensor noise, and motion blur degrade both feature extraction and feature matching, while compensating with LiDAR, depth, or thermal sensors raises cost, power draw, and integration complexity. Existing benchmarks remain dominated by well-lit indoor or daylight sequences, leaving open how far SLAM with standard RGB cameras can be pushed in the dark. We benchmark six systems spanning the feature-based, direct, filter-based, and learning-based paradigms - ORB-SLAM3, DSO, Kimera-VIO, OpenVINS, DPVO, and DPV-SLAM - on five LaMARia sequences of varying difficulty and illumination, reporting absolute and relative pose error alongside control-point recall. Kimera-VIO is the only system to track all five sequences to completion, combining the lowest relative pose error with steadily growing absolute error due to the absence of loop closure; DPVO and DPV-SLAM never lose tracking but incur absolute errors of roughly 100 m under low light; and the classical monocular pipelines (ORB-SLAM3, DSO) together with the filter-based OpenVINS fail outright or diverge on most of the harder and low-light sequences. The results suggest that RGB-only SLAM maintains stable low-light tracking only when both inertial fusion and global optimization are present. Closing the remaining gap will likely require low-light-specific learned front-ends or a return to complementary sensing.

Figures

Figures reproduced from arXiv: 2607.17699 by Anna Stasyshyn, Oleh Basystyi, Oleksandr Kosovan, Yaroslav Prytula.

Figure 1
Figure 1. Figure 1: Standard modules for SLAM pipeline implementations [PITH_FULL_IMAGE:figures/full_fig_p013_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Estimated versus ground-truth trajectories on [PITH_FULL_IMAGE:figures/full_fig_p014_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Estimated versus ground-truth trajectories on [PITH_FULL_IMAGE:figures/full_fig_p015_3.png] view at source ↗
Figure 5
Figure 5. Figure 5: Estimated versus ground-truth trajectories on the low [PITH_FULL_IMAGE:figures/full_fig_p016_5.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

58 extracted references · 2 canonical work pages

  1. [1]

    Monocular Visual Simultaneous Localization and Mapping: (R)Evolution From Geometry to Deep Learning-Based Pipelines , volume=

    Álvarez-Tuñón, Olaya and Brodskiy, Yury and Kayacan, Erdal , year=. Monocular Visual Simultaneous Localization and Mapping: (R)Evolution From Geometry to Deep Learning-Based Pipelines , volume=. IEEE Transactions on Artificial Intelligence , publisher=. doi:10.1109/tai.2023.3321032 , number=

    arXiv 2023

  2. [2]

    ORB: An efficient alternative to SIFT or SURF , year=

    Rublee, Ethan and Rabaud, Vincent and Konolige, Kurt and Bradski, Gary , booktitle=. ORB: An efficient alternative to SIFT or SURF , year=

  3. [3]

    , title =

    Lowe, David G. , title =. Int. J. Comput. Vision , month = nov, pages =. 2004 , issue_date =. doi:10.1023/B:VISI.0000029664.99615.94 , abstract =

    arXiv 2004

  4. [4]

    CoRR , volume =

    Yifan Jiang and Xinyu Gong and Ding Liu and Yu Cheng and Chen Fang and Xiaohui Shen and Jianchao Yang and Pan Zhou and Zhangyang Wang , title =. CoRR , volume =. 2019 , url =. 1906.06972 , timestamp =

    Pith/arXiv arXiv 2019

  5. [5]

    Zhao, Xiaoming and Wu, Xingming and Chen, Weihai and Chen, Peter C. Y. and Xu, Qingsong and Li, Zhengguo , year =. ALIKED: A Lighter Keypoint and Descriptor Extraction Network via Deformable Transformation , url =. doi:10.1109/TIM.2023.3271000 , journal =

    arXiv 2023

  6. [6]

    Zhao, Xiaoming and Wu, Xingming and Miao, Jinyu and Chen, Weihai and Chen, Peter C. Y. and Li, Zhengguo , month =. ALIKE: Accurate and Lightweight Keypoint Detection and Descriptor Extraction , url =. doi:10.1109/TMM.2022.3155927 , journal =

    arXiv 2022

  7. [7]

    Proceedings of the 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition , year =

    Dusmanu, Mihai and Rocco, Ignacio and Pajdla, Tomas and Pollefeys, Marc and Sivic, Josef and Torii, Akihiko and Sattler, Torsten , title =. Proceedings of the 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition , year =

    2019

  8. [8]

    CoRR , volume =

    Chunle Guo and Chongyi Li and Jichang Guo and Chen Change Loy and Junhui Hou and Sam Kwong and Runmin Cong , title =. CoRR , volume =. 2020 , url =. 2001.06826 , timestamp =

    Pith/arXiv arXiv 2020

  9. [9]

    Guo, Xiaojie and Hu, Qiming , title =. Int. J. Comput. Vision , month = oct, pages =. 2022 , issue_date =. doi:10.1007/s11263-022-01667-9 , abstract =

    work page doi:10.1007/s11263-022-01667-9 2022

  10. [10]

    Dual Illumination Estimation for Robust Exposure Correction , journal =

    Qing Zhang and Yongwei Nie and Wei. Dual Illumination Estimation for Robust Exposure Correction , journal =. 2019 , url =. 1910.13688 , timestamp =

    Pith/arXiv arXiv 2019

  11. [11]

    Sun, Jiaming and Shen, Zehong and Wang, Yuang and Bao, Hujun and Zhou, Xiaowei , journal=

  12. [12]

    CVPR , year =

    Paul-Edouard Sarlin and Daniel DeTone and Tomasz Malisiewicz and Andrew Rabinovich , title =. CVPR , year =

  13. [13]

    arXiv preprint arXiv:2304.11310 , year =

    Twilight SLAM: A Comparative Study of Low-Light Visual SLAM Pipelines , author =. arXiv preprint arXiv:2304.11310 , year =

    Pith/arXiv arXiv

  14. [14]

    Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , year =

    LightGlue: Local Feature Matching at Light Speed , author =. Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , year =

  15. [15]

    Fast Approximate Nearest Neighbors with Automatic Algorithm Configuration

    Muja, Marius and Lowe, David , year =. Fast Approximate Nearest Neighbors with Automatic Algorithm Configuration. , booktitle=

  16. [16]

    and Netanyahu, Nathan S

    Arya, Sunil and Mount, David M. and Netanyahu, Nathan S. and Silverman, Ruth and Wu, Angela Y. , title =. J. ACM , month = nov, pages =. 1998 , issue_date =. doi:10.1145/293347.293348 , abstract =

    arXiv 1998

  17. [17]

    SuperPoint: Self-Supervised Interest Point Detection and Description , year=

    DeTone, Daniel and Malisiewicz, Tomasz and Rabinovich, Andrew , booktitle=. SuperPoint: Self-Supervised Interest Point Detection and Description , year=

  18. [18]

    SURF: Speeded Up Robust Features

    Bay, Herbert and Tuytelaars, Tinne and Van Gool, Luc. SURF: Speeded Up Robust Features. Computer Vision -- ECCV 2006. 2006

    2006

  19. [19]

    BRIEF: Binary Robust Independent Elementary Features

    Calonder, Michael and Lepetit, Vincent and Strecha, Christoph and Fua, Pascal. BRIEF: Binary Robust Independent Elementary Features. Computer Vision -- ECCV 2010. 2010

    2010

  20. [20]

    Machine Learning for High-Speed Corner Detection

    Rosten, Edward and Drummond, Tom. Machine Learning for High-Speed Corner Detection. Computer Vision -- ECCV 2006. 2006

    2006

  21. [21]

    ArXiv , year=

    VAR-SLAM: Visual Adaptive and Robust SLAM for Dynamic Environments , author=. ArXiv , year=. 2510.16205 , archivePrefix=

    arXiv

  22. [22]

    Proceedings of the 33rd ACM International Conference on Multimedia , pages =

    Sun, Hao and Yu, Fenggen and Xu, Huiyao and Zhang, Tao and Zou, Changqing , title =. Proceedings of the 33rd ACM International Conference on Multimedia , pages =. 2025 , isbn =. doi:10.1145/3746027.3755375 , abstract =

    arXiv 2025

  23. [23]

    Machines , VOLUME =

    Liu, Cong and Wang, You and Luo, Weichao and Peng, Yanhong , TITLE =. Machines , VOLUME =. 2026 , NUMBER =

    2026

  24. [24]

    LFPD-SLAM: Visual SLAM with Learning-Based Features and Pseudo-Depth , year=

    Lu, Xu and Yu, Fujian and Liu, Jun and Chen, Jurong and Huang, Hanyuan , booktitle=. LFPD-SLAM: Visual SLAM with Learning-Based Features and Pseudo-Depth , year=

  25. [25]

    and Pollefeys, Marc , title =

    Pautrat, R\'emi and Barath, Daniel and Larsson, Viktor and Oswald, Martin R. and Pollefeys, Marc , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =. 2023 , pages =

    2023

  26. [26]

    2026 , eprint=

    What Is The Best 3D Scene Representation for Robotics? From Geometric to Foundation Models , author=. 2026 , eprint=

    2026

  27. [27]

    and Hartley, Richard I

    Triggs, Bill and McLauchlan, Philip F. and Hartley, Richard I. and Fitzgibbon, Andrew W. Bundle Adjustment --- A Modern Synthesis. Vision Algorithms: Theory and Practice. 2000

    2000

  28. [28]

    Montiel and Andrew J

    Hauke Strasdat and J.M.M. Montiel and Andrew J. Davison , keywords =. Visual SLAM: Why filter? , journal =. 2012 , issn =. doi:https://doi.org/10.1016/j.imavis.2012.02.009 , url =

    work page doi:10.1016/j.imavis.2012.02.009 2012

  29. [29]

    Mur-Artal, Raúl and Montiel, J. M. M. and Tardós, Juan D. , journal=. ORB-SLAM: A Versatile and Accurate Monocular SLAM System , year=

  30. [30]

    , journal=

    Mur-Artal, Raúl and Tardós, Juan D. , journal=. ORB-SLAM2: An Open-Source SLAM System for Monocular, Stereo, and RGB-D Cameras , year=

  31. [31]

    Gómez and M

    Campos, Carlos and Elvira, Richard and Rodríguez, Juan J. Gómez and M. Montiel, José M. and D. Tardós, Juan , journal=. ORB-SLAM3: An Accurate Open-Source Library for Visual, Visual–Inertial, and Multimap SLAM , year=

  32. [32]

    and Johnston, R.E

    Pizer, S.M. and Johnston, R.E. and Ericksen, J.P. and Yankaskas, B.C. and Muller, K.E. , booktitle=. Contrast-limited adaptive histogram equalization: speed and effectiveness , year=

  33. [33]

    and Bennewitz, Maren and Stachniss, Cyrill and Burgard, Wolfram , title =

    Hornung, Armin and Wurm, Kai M. and Bennewitz, Maren and Stachniss, Cyrill and Burgard, Wolfram , title =. Auton. Robots , month = apr, pages =. 2013 , issue_date =. doi:10.1007/s10514-012-9321-0 , abstract =

    doi:10.1007/s10514-012-9321-0 2013

  34. [34]

    CoRR , volume =

    Antoni Rosinol and Arjun Gupta and Marcus Abate and Jingnan Shi and Luca Carlone , title =. CoRR , volume =. 2020 , url =. 2002.06289 , timestamp =

    Pith/arXiv arXiv 2020

  35. [35]

    and Tancik, Matthew and Barron, Jonathan T

    Mildenhall, Ben and Srinivasan, Pratul P. and Tancik, Matthew and Barron, Jonathan T. and Ramamoorthi, Ravi and Ng, Ren , title =. Commun. ACM , month = dec, pages =. 2021 , issue_date =. doi:10.1145/3503250 , abstract =

    doi:10.1145/3503250 2021

  36. [36]

    ACM Trans

    Kerbl, Bernhard and Kopanas, Georgios and Leimkuehler, Thomas and Drettakis, George , title =. ACM Trans. Graph. , month = jul, articleno =. 2023 , issue_date =. doi:10.1145/3592433 , abstract =

    doi:10.1145/3592433 2023

  37. [37]

    2024 , eprint=

    Gaussian Splatting SLAM , author=. 2024 , eprint=

    2024

  38. [38]

    Kalman, R. E. , title =. Journal of Basic Engineering , volume =. 1960 , month =. doi:10.1115/1.3662552 , url =

    doi:10.1115/1.3662552 1960

  39. [39]

    2024 , eprint=

    DUSt3R: Geometric 3D Vision Made Easy , author=. 2024 , eprint=

    2024

  40. [40]

    2025 , eprint=

    VGGT: Visual Geometry Grounded Transformer , author=. 2025 , eprint=

    2025

  41. [41]

    2025 , eprint=

    DINOv3 , author=. 2025 , eprint=

    2025

  42. [42]

    2018 , volume=

    Engel, Jakob and Koltun, Vladlen and Cremers, Daniel , journal=. 2018 , volume=. doi:10.1109/TPAMI.2017.2658577 , url =

    arXiv 2018

  43. [43]

    , journal=

    Cadena, Cesar and Carlone, Luca and Carrillo, Henry and Latif, Yasir and Scaramuzza, Davide and Neira, José and Reid, Ian and Leonard, John J. , journal=. Past, Present, and Future of Simultaneous Localization and Mapping: Toward the Robust-Perception Age , year=

  44. [44]

    and Siegwart, Roland , title =

    Burri, Michael and Nikolic, Janosch and Gohl, Pascal and Schneider, Thomas and Rehder, Joern and Omari, Sammy and Achtelik, Markus W. and Siegwart, Roland , title =. The International Journal of Robotics Research , volume =. 2016 , doi =

    2016

  45. [45]

    Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , year =

    Geiger, Andreas and Lenz, Philip and Urtasun, Raquel , title =. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , year =

  46. [46]

    The TUM VI Benchmark for Evaluating Visual-Inertial Odometry , url=

    Schubert, David and Goll, Thore and Demmel, Nikolaus and Usenko, Vladyslav and Stuckler, Jorg and Cremers, Daniel , year=. The TUM VI Benchmark for Evaluating Visual-Inertial Odometry , url=. doi:10.1109/iros.2018.8593419 , booktitle=

    arXiv 2018

  47. [47]

    2025 , eprint=

    Benchmarking Egocentric Visual-Inertial SLAM at City Scale , author=. 2025 , eprint=

    2025

  48. [48]

    A benchmark for the evaluation of RGB-D SLAM systems , year=

    Sturm, Jürgen and Engelhard, Nikolas and Endres, Felix and Burgard, Wolfram and Cremers, Daniel , booktitle=. A benchmark for the evaluation of RGB-D SLAM systems , year=

  49. [49]

    2020 , booktitle =

    Kimera: an Open-Source Library for Real-Time Metric-Semantic Localization and Mapping , author =. 2020 , booktitle =

    2020

  50. [50]

    2005 , publisher=

    Probabilistic Robotics , author=. 2005 , publisher=

    2005

  51. [51]

    OpenVINS: A Research Platform for Visual-Inertial Estimation , year=

    Geneva, Patrick and Eckenhoff, Kevin and Lee, Woosik and Yang, Yulin and Huang, Guoquan , booktitle=. OpenVINS: A Research Platform for Visual-Inertial Estimation , year=

  52. [52]

    Advances in Neural Information Processing Systems (NeurIPS) , volume=

    Deep Patch Visual Odometry , author=. Advances in Neural Information Processing Systems (NeurIPS) , volume=. 2023 , eprint=

    2023

  53. [53]

    Deep Patch Visual

    Lipson, Lahav and Teed, Zachary and Deng, Jia , booktitle=. Deep Patch Visual. 2024 , eprint=

    2024

  54. [54]

    and Nilsson, Nils J

    Hart, Peter E. and Nilsson, Nils J. and Raphael, Bertram , title =. IEEE Transactions on Systems Science and Cybernetics , volume =. 1968 , doi =

    1968

  55. [55]

    , title =

    LaValle, Steven M. , title =

  56. [56]

    2020 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) , pages =

    Macenski, Steve and Martin, Francisco and White, Ruffin and Ginés Clavero, Jonatan , title =. 2020 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) , pages =. 2020 , doi =

    2020

  57. [57]

    Autonomous Robot Vehicles , editor =

    Smith, Randall and Self, Matthew and Cheeseman, Peter , title =. Autonomous Robot Vehicles , editor =. 1990 , doi =

    1990

  58. [58]

    Gamini and Newman, Paul and Clark, Steve and Durrant-Whyte, Hugh F

    Dissanayake, M.W.M. Gamini and Newman, Paul and Clark, Steve and Durrant-Whyte, Hugh F. and Csorba, Michael , title =. IEEE Transactions on Robotics and Automation , volume =. 2001 , doi =

    2001

This paper was first reviewed by deepseek-v4-flash on August 1, 2026.