Pith. sign in

REVIEW 2 major objections 5 minor 63 references

Self-Healing Visual Recovery for Autonomous Ground Vehicles Using Camera-Only Visual Odometry

T0 review · 2 major / 5 minor · reviewed 2026-07-14 · grok-4.5

Pith's one-line read A two-stage camera-only recovery restores floor-line tracking on cheap robots in 86.6% of losses, median 3.26 seconds, with no LiDAR, GPS, or GPU.

desk verdict Solid, carefully instrumented systems paper that shows two-stage camera-only recovery works in Webots; the only load-bearing limit is the sim-to-real gap the authors already flag. read the letter →

arxiv 2607.11686 v1 pith:3IXNFLMR submitted 2026-07-13 cs.RO cs.LGcs.SE

classification cs.ROcs.LGcs.SE
keywords autonomousgroundvehicleself-adaptivesystemslinefollowingvisualodometryobstacleavoidancefaulttolerancelow-costroboticsMAPE-K
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

Cheap warehouse and farm robots often follow a painted floor line with a single camera, but the line can vanish behind obstacles or leave the frame after a sharp turn. Sensor-heavy platforms recover with extra hardware; camera-only ones usually stop. This paper shows that a lightweight two-stage procedure can reacquire the line without LiDAR, GPS, or a GPU. Stage 1 spins in place while relaxing color checks and requiring multi-frame confirmation. If that fails, monocular visual odometry drives the robot back to stored breadcrumb poses and retries. The whole MAPE-K sense–decide–act loop fits inside a 50 ms tick and runs at 20 Hz on CPU-only hardware. Across 119 fault-injected episodes on three simulated courses the system succeeded 86.6% of the time with a median recovery of 3.26 s, arguing that reliable self-healing is already practical under strict cost and compute limits.

What carries the argument

Two-stage visual recovery: Stage 1 rotates in place under relaxed HSV and floor-plane thresholds with five-frame confirmation; Stage 2 navigates monocular VO breadcrumbs to the nearest prior line pose and re-invokes Stage 1 (up to three cycles).

What would settle it

Run the identical controller and fault protocol on a physical JetBot with an Orbbec Astra camera under real warehouse or office lighting and flooring; if the success rate falls substantially below the 0.79 lower confidence bound or median TTR rises well above a few seconds, the claim does not hold.

Watch

Extended reading notes

Core claim

Reliable guideline recovery after complete visual loss is achievable on camera-only, CPU-only UGVs by a two-stage policy—spin-and-search with relaxed thresholds followed by visual-odometry return to stored breadcrumb poses—that embeds a full MAPE-K loop in each 50 ms control tick and recovers 86.6% of 119 simulated line-loss episodes at a median 3.26 s.

Load-bearing premise

The measured success rate and recovery times will transfer from idealized Webots RGB-D and physics models to real JetBot-class hardware under motion blur, glare, depth noise, and floor-texture variation.

Editorial extensions

If this is right

  • Low-cost line-following fleets can keep operating after line loss without adding LiDAR, GPS, or GPU hardware.
  • Recovery policy can be chosen by course geometry: partial-visibility settings favor sub-second spin recovery; complete-occlusion settings require VO breadcrumb navigation of a few seconds.
  • A complete MAPE-K self-adaptation loop can be embedded inside a single 50 ms control tick on commodity CPUs, removing the need for an external adaptation manager.
  • Breadcrumb density and monocular VO drift become the practical limits that future hardware-light systems must improve.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If sim-to-real transfer holds, the same staged pattern could be reused for other camera-only references such as crop rows or wall edges without new infrastructure.
  • Because Stage-2 targets are local breadcrumbs rather than global place recognition, adding a lightweight appearance descriptor would be a natural next step to handle larger displacements.
  • The geometry-driven split between Stage 1 and Stage 2 suggests that real deployments could pre-select recovery budgets from a simple map of corridor shape rather than learning them online.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 5 minor

Summary. The paper presents a camera-only, CPU-only self-healing controller for low-cost UGVs that recover from complete guideline loss. A depth-gated HSV line tracker (online EMA hue + per-row floor model + periodic SVD plane fit), a depth-fused YOLOv8n obstacle detector, and monocular VO breadcrumbs feed an inlined MAPE-K loop that runs at 20 Hz inside a single 50 ms tick. Recovery is two-stage: Stage 1 is an in-place spin with relaxed thresholds and multi-frame confirmation; Stage 2 navigates to the nearest stored breadcrumb and retries. Evaluation consists of 119 programmatically injected visual-loss episodes on three Webots courses, yielding 86.6 % success (95 % Wilson CI [0.79, 0.92]) and median TTR 3.26 s. The authors claim this demonstrates that reliable visual recovery is feasible under the three simultaneous constraints of camera-only sensing, CPU-only compute, and no external adaptation manager.

Significance. If the measured recovery rates transfer, the work fills a genuine gap for warehouse, inspection, and small-scale agricultural platforms that cannot afford LiDAR/GPS/GPU suites. The evaluation is carefully instrumented (Wilson CIs, per-course mechanism counts, TTR histograms, explicit failure-mode taxonomy) and all free parameters are listed in the appendix, which is a strength relative to many systems papers. The inlined MAPE-K design and the staged spin-then-breadcrumb policy are concrete, reproducible engineering contributions. The principal limitation is that every reported number is produced inside Webots under idealized RGB-D and rigid-body models; the abstract and conclusion nevertheless frame the result as feasibility “on camera-only UGVs.” Physical validation is therefore the decisive next step, but the simulation evidence itself is solid enough to justify publication after modest revision.

major comments (2)
  1. Abstract, §VI and §IX: the central claim is framed as feasibility “on camera-only UGVs,” yet every quantitative result (Table 2, Figs. 10–11) is obtained exclusively in Webots. Section VII correctly lists motion blur, glare, flying-pixel depth noise and floor-texture variation as primary threats that can break the depth-gated HSV mask (Eqs. 1–3), the MAD residual obstacle filter, and monocular VO breadcrumb accuracy. The manuscript should either (a) add a physical JetBot/Astra validation set, or (b) systematically qualify every success-rate and TTR statement as “in simulation” and move the stronger real-world claim to future work. Without one of these changes the abstract overstates the evidence.
  2. §V-D and Table 2: the 119-episode corpus contains only “visual-loss” (type-1) faults. Dynamic obstacles and pose-displacement faults are mentioned but excluded from the reported statistics. Because two of the three failure modes identified in §VI-A5 (persistent occlusion and sparse breadcrumb coverage after avoidance) arise precisely from obstacle-induced secondary line loss, the evaluation under-samples the conditions that most stress Stage 2. At minimum the authors should report the number of secondary line-loss episodes that occurred during the dynamic-obstacle trials, even if those trials remain qualitative.
minor comments (5)
  1. Table 3 comparison with Lee et al. [14] is useful but the footnote “qualitative success rate; not per-episode” should be expanded: state the exact protocol difference so readers can judge whether the 0.87 vs 0.87 match is meaningful.
  2. Eq. (1)–(3) and the recovery-stage threshold relaxations (δh += 8, floor residual +0.02 m, etc.) are listed in the appendix but never justified by ablation. A short sensitivity paragraph or a single ablation table would strengthen confidence that the chosen values are not brittle.
  3. Figure 2 is dense; the three perception modules and the P1/P2/P3 priority arrows become hard to parse at print size. Consider splitting the MAPE-K phases from the recovery sub-flow (already shown in Fig. 5).
  4. §IV-C3 notes that monocular VO drift is “the main limitation of Stage-2 recovery” yet no quantitative drift measurement (e.g., endpoint error after 5 m pure rotation) is supplied. Even a short offline calibration on the same Webots trajectories would make the failure-mode discussion more concrete.
  5. Typographical: “Anal yze” and “La yer” appear with spurious spaces in several section headings; “reaquires” in Fig. 5; “V alidating” in §VIII.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: empirical systems evaluation whose success rate and TTR are measured outcomes of fault-injected episodes, not quantities forced by construction or self-citation.

full rationale

The paper's central claim is an empirical performance result (86.6 % recovery success, median TTR 3.26 s across 119 Webots episodes) obtained by running a fixed two-stage controller against programmatically injected line-loss faults. The controller parameters (hue tolerances, confirmation frames N_confirm=5, spin budget T_spin=2 s, breadcrumb distance, etc.) are design choices listed in the appendix; they are not fitted to the reported success rate or TTR and then re-used to 'predict' those same quantities. No equation equates a claimed prediction to an input by definition, no uniqueness theorem is imported from the authors' prior work, and the load-bearing numbers in Table 2 and Figures 10-11 are direct episode outcomes rather than restatements of a fit. Self-citations are absent from the derivation chain. The evaluation is therefore self-contained against its own external fault-injection protocol; any remaining concerns (sim-to-real gap) are validity threats, not circularity.

Assumptions & free parameters 6 free parameters · 4 assumptions · 2 invented entities

The central empirical claim rests on a large set of hand-chosen controller thresholds, the assumption that Webots physics and sensors adequately proxy real JetBot hardware, and the modeling choice that short-range monocular VO breadcrumbs remain accurate enough for Stage-2 return. No new physical entities are postulated; the “invented” constructs are engineering modules whose independent evidence is the simulation evaluation itself.

free parameters (6)
  • N_confirm (confirmation frames) = 5
    Fixed at 5 consecutive frames; chosen by simulation tuning, directly gates acceptance of Stage-1 reacquisition.
  • T_spin (spin-search budget) = 2.0 s
    Fixed at 2.0 s; determines when Stage 1 is declared failed and Stage 2 is escalated.
  • Grace window before declaring line loss = 0.6 s
    0.6 s continuous absence required; hand-tuned to suppress brief dropouts.
  • N_max (max spin–navigate cycles) = 3
    Budget of 3 cycles before failure is declared; directly affects the reported success rate.
  • δh recovery expansion / floor residual / gradient ceilings = see Table 4
    Multiple perception thresholds (+8 hue units, 0.02 m residual, 0.04 gradient, etc.) set by hand and listed in Table 4; they control both false-positive rate and recovery success.
  • PD gains Kp, Kd and obstacle distances d_warn, d_near = Kp=0.9, Kd=0.1, d_warn=1.40 m, d_near=0.75 m
    Controller gains and avoidance distances (0.9, 0.1, 1.40 m, 0.75 m) are free design choices that shape both normal tracking and secondary line-loss frequency.
assumptions (4)
  • domain assumption Webots RGB-D and rigid-body models are sufficiently faithful that measured recovery statistics transfer to physical JetBot-class platforms.
    Stated as the primary threat in §VII; the entire evaluation (119 episodes) is conducted exclusively inside Webots.
  • domain assumption Short-range monocular VO (GFTT+ORB+5-pt RANSAC) produces breadcrumb poses accurate enough for Stage-2 return under the tested geometries.
    Explicitly used in §IV-C3 and §IV-E3; VO drift is later listed as a failure mode, confirming the assumption is load-bearing.
  • domain assumption A painted floor line on a mostly planar surface remains the primary navigational reference and can be segmented by depth-gated HSV.
    Environment assumptions in §VII; the whole pipeline is built around this cue.
  • standard math Standard computer-vision primitives (HSV inRange, SVD plane fit, five-point essential matrix, YOLOv8n ONNX) behave as documented in OpenCV/Ultralytics.
    Used throughout Phase 2 without re-derivation.
invented entities (2)
  • Two-stage visual recovery policy (spin-and-search then VO-breadcrumb navigation)
    purpose: Provides the autonomous recovery behavior that is the paper’s central contribution.
    The staged escalation with explicit budgets, multi-frame confirmation and breadcrumb lookup is the novel control construct; its only evidence is the simulation evaluation itself.
  • Inlined MAPE-K loop inside a single 50 ms control tick
    purpose: Eliminates the need for an external adaptation manager on CPU-only hardware.
    Architectural choice described in §II-C and §IV; not previously demonstrated for this recovery task under the stated constraints.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Self-Healing Visual Recovery for Autonomous Ground Vehicles Using Camera-Only Visual Odometry." pith.science (2026). https://pith.science/paper/3IXNFLMR

@misc{pith2026260711686,
  author       = {Pith},
  title        = {Pith review of: Self-Healing Visual Recovery for Autonomous Ground Vehicles Using Camera-Only Visual Odometry},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3IXNFLMR}},
  note         = {Machine review of arXiv:2607.11686}
}
read the original abstract

Low-cost unmanned ground vehicles are often used in indoor places like warehouses, inspection corridors, and farm rows, where painted floor lines guide the robot. Line following is useful because it only needs one camera and little computing power, but it can fail when the line is blocked or turns sharply and goes out of view. Sensor-rich platforms tolerate this through hardware redundancy (LiDAR, GPS, multiple cameras), but camera-only systems must recover at runtime with no additional infrastructure. This paper presents a lightweight, two-stage recovery approach that restores guideline tracking without LiDAR, GPS, or a GPU. When the line is lost, the robot first turns in place while slowly relaxing its color checks and waiting for confirmation across multiple frames (Stage 1). If the line is still not found, monocular visual odometry moves the robot back to saved breadcrumb positions before it tries again (Stage 2). The system uses a depth-gated HSV line tracker, a YOLOv8n obstacle detector, and a visual odometry breadcrumb mapper, and it runs at 20 Hz on CPU-only hardware. The controller embeds a complete MAPE-K loop within a single 50 ms control tick, with no external adaptation manager required. The approach is evaluated across 119 fault-injected episodes on three Webots simulation courses. The method was successful in 86.6% of cases, with a median recovery time of 3.26 seconds. These results demonstrate that reliable visual recovery is feasible on camera-only UGVs within practical cost and computational limits.

Figures

Figures reproduced from arXiv: 2607.11686 by the authors.

Figure 1
Figure 1. FIGURE 1 [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Each 50 ms control tick proceeds through four [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 2
Figure 2. FIGURE 2 [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figures from the paper (9 more)
Figure 3
Figure 3. Figure 3: FIGURE 3 [PITH_FULL_IMAGE:figures/full_fig_p007_3.png]
Figure 4
Figure 4. Figure 4: FIGURE 4 [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: FIGURE 5 [PITH_FULL_IMAGE:figures/full_fig_p009_5.png]
Figure 6
Figure 6. Figure 6: FIGURE 6 [PITH_FULL_IMAGE:figures/full_fig_p010_6.png]
Figure 7
Figure 7. Figure 7: FIGURE 7 [PITH_FULL_IMAGE:figures/full_fig_p011_7.png]
Figure 8
Figure 8. Figure 8: FIGURE 8 [PITH_FULL_IMAGE:figures/full_fig_p011_8.png]
Figure 9
Figure 9. Figure 9: FIGURE 9 [PITH_FULL_IMAGE:figures/full_fig_p011_9.png]
Figure 10
Figure 10. Figure 10: FIGURE 10 [PITH_FULL_IMAGE:figures/full_fig_p013_10.png]
Figure 11
Figure 11. Figure 11: FIGURE 11 [PITH_FULL_IMAGE:figures/full_fig_p013_11.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

63 extracted references · 5 linked inside Pith

  1. [1]

    W. Wei, M. Xiao, W. Duan, H. Wang, Y . Zhu, C. Zhai, and G. Geng. ‘‘Research Progress on Autonomous Operation Technology for Agricultural Equipment in Large Fields.’’ In:Agriculture14.9 (2024), p. 1473. DOI: 10.3390/agriculture14091473

  2. [2]

    Agelli, N

    M. Agelli, N. Corona, F. Maggio, and P . V . Moi. ‘‘Unmanned Ground V ehicles for Continuous Crop Monitoring in Agriculture: Assessing the Readiness of Current ICT Technology.’’ In:Ma- chines12.11 (2024), p. 750. DOI: 10.3390/machines12110750

  3. [3]

    Maimone, Y

    M. Maimone, Y . Cheng, and L. Matthies. ‘‘Two Y ears of Visual Odometry on the Mars Exploration Rovers.’’ In:J. Field Robotics 24.3 (2007), pp. 169–186. DOI: 10.1002/rob.20184

  4. [4]

    Cadena, L

    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.’’ In:IEEE Trans. Robotics32.6 (2016), pp. 1309–

  5. [5]

    DOI: 10.1109/TRO.2016.2624754

  6. [6]

    Misaros, O

    M. Misaros, O. -P . Stan, I.-C. Donca, and L. -C. Miclea. ‘‘Au- tonomous Robots for Services—State of the Art, Challenges, and Research Areas.’’ In:Sensors23.10 (2023), p. 4962. DOI: 10.3390/s23104962

  7. [7]

    Spagnuolo, G

    M. Spagnuolo, G. Todde, M. Caria, N. Furnitto, G. Schillaci, and S. Failla. ‘‘Agricultural Robotics: A Technical Review Addressing Challenges in Sustainable Crop Production.’’ In:Robotics14.2 (2025), p. 9. DOI: 10.3390/robotics14020009

  8. [8]

    Bonin-Font, A

    F. Bonin-Font, A. Ortiz, and G. Oliver. ‘‘Visual Navigation for Mobile Robots: A Survey.’’ In:J. Intelligent and Robotic Systems 53.3 (2008), pp. 263–296. DOI: 10.1007/s10846-008-9235-4

Show all 63 references
  1. [9]

    C.-Y . Wang, A. Bochkovskiy, and H. -Y . M. Liao. ‘‘YOLOv7: Trainable Bag-of-Freebies Sets New State-of-the-Art for Real- Time Object Detectors.’’ In:IEEE/CVF Conf. Computer Vision and Pattern Recognition. 2023, pp. 7464–7475

  2. [10]

    Jocher, A

    G. Jocher, A. Chaurasia, and J. Qiu.Ultralytics YOLOv8. Url: https://github.com/ultralytics/ultralytics. 2023

  3. [11]

    Godard, O

    C. Godard, O. Mac Aodha, M. Firman, and G. J. Brostow. ‘‘Digging Into Self-Supervised Monocular Depth Estimation.’’ In:IEEE/CVF Int’l Conf. Computer Vision. 2019, pp. 3828–3838

  4. [12]

    Ranftl, K

    R. Ranftl, K. Lasinger, D. Hafner, K. Schindler, and V . Koltun. ‘‘Towards Robust Monocular Depth Estimation: Mixing Datasets 16 J. S. Berntzen, S. Fatima, and L. Moonen: Self-Healing Visual Recovery for Autonomous Ground Vehicles . . . for Zero-Shot Cross-Dataset Transfer.’’ ...

  5. [13]

    Scaramuzza and F

    D. Scaramuzza and F. Fraundorfer. ‘‘Visual Odometry [Tutorial].’’ In:IEEE Robotics & Automation Magazine18.4 (2011), pp. 80–

  6. [14]

    doi: 10.1109/MRA.2011.943233

  7. [15]

    Engel, V

    J. Engel, V . Koltun, and D. Cremers. ‘‘Direct Sparse Odometry.’’ In:IEEE Trans. Pattern Analysis and Machine Intelligence40.3 (2018), pp. 611–625. doi: 10.1109/TPAMI.2017.2658577

  8. [16]

    S. Lee, S. Lee, and S. Baek. ‘‘Vision-Based Kidnap Recovery with SLAM for Home Cleaning Robots.’’ In:J. Intelligent & Robotic Systems67.1 (2012), pp. 7–24. doi: 10.1007/s10846-011-9647-4

  9. [17]

    R. C. Luo, K. C. Y eh, and K. H. Huang. ‘‘Resume Navigation and Re-Localization of an Autonomous Mobile Robot after Being Kidnapped.’’ In:IEEE Int’l Symp. Robotic and Sensors Environments (ROSE). 2013, pp. 7–12. doi: 10.1109/ROSE.2013. 6698410

  10. [18]

    White, J

    S. White, J. Hanson, I. Whalley, D. Chess, and J. Kephart. ‘‘An Architectural Approach to Autonomic Computing.’’ In:Int’l Conf. Autonomic Computing, 2004. Proceedings.2004, pp. 2–9. doi: 10.1109/ICAC.2004.1301340

  11. [19]

    Cheng et al.Using Models at Runtime to Address Assurance for Self-Adaptive Systems

    B. Cheng et al.Using Models at Runtime to Address Assurance for Self-Adaptive Systems. 2015. arXiv: 1505.00903

  12. [20]

    de Lemos et al

    R. de Lemos et al. ‘‘Software Engineering for Self-Adaptive Systems: A Second Research Roadmap.’’ In:Software Engineer- ing for Self-Adaptive Systems II: Int’l Seminar , Dagstuhl Castle, Germany, October 24-29, 2010 Revised Selected and Invited Papers. Springer, 2013, pp. 1–32...

  13. [21]

    Romero-Garcés, A

    A. Romero-Garcés, A. Hidalgo-Paniagua, M. González-García, and A. Bandera. ‘‘On Managing Knowledge for MAPE-K Loops in Self-Adaptive Robotics Using a Graph-Based Runtime Model.’’ In:Applied Sciences12.17 (2022), p. 8583. doi: 10 . 3390 / app12178583

  14. [22]

    Thrun, D

    S. Thrun, D. Fox, W. Burgard, and F. Dellaert. ‘‘Robust Monte Carlo Localization for Mobile Robots.’’ In:Artificial Intelligence 128.1 (2001), pp. 99–141. doi: 10.1016/S0004-3702(01)00069-8

  15. [23]

    S. Y u, F. Y an, Y . Zhuang, and D. Gu. ‘‘A Deep-Learning-based Strategy for Kidnapped Robot Problem in Similar Indoor En- vironment.’’ In:J. Intelligent & Robotic Systems100.3 (2020), pp. 765–775. doi: 10.1007/s10846-020-01216-x

  16. [24]

    Teed and J

    Z. Teed and J. Deng. ‘‘DROID-SLAM: Deep Visual SLAM for Monocular, Stereo, and RGB-D Cameras.’’ In:Advances in Neural Information Processing Systems. V ol. 34. Curran Associates, Inc., 2021, pp. 16558–16569

  17. [25]

    S. Thrun. ‘‘Probabilistic Robotics.’’ In:Commun. ACM45.3 (2002), pp. 52–57. doi: 10.1145/504729.504754

  18. [26]

    Nister, O

    D. Nister, O. Naroditsky, and J. Bergen. ‘‘Visual Odometry.’’ In:IEEE Computer Society Conf. Computer Vision and Pattern Recognition, 2004. CVPR 2004.V ol. 1. 2004, pp. I–I. doi: 10. 1109/CVPR.2004.1315094

  19. [27]

    Url: https://ifr.org/ifr-press-releases/ news/global-robot-demand-in-factories-doubles-over-10-years

    IFR International Federation of Robotics.World Robotics 2025 Report – Industrial Robots. Url: https://ifr.org/ifr-press-releases/ news/global-robot-demand-in-factories-doubles-over-10-years. 2025

  20. [28]

    Url: https://ifr.org/ifr-press-releases/ news/service-robots-see-global-growth-boom

    IFR International Federation of Robotics.World Robotics 2025 Report – Service Robots. Url: https://ifr.org/ifr-press-releases/ news/service-robots-see-global-growth-boom. 2025

  21. [29]

    Url: https : / / ifr

    IFR International Federation of Robotics.Robot Density Surges in Europe, Asia, and Americas. Url: https : / / ifr. org / ifr- press - releases / news / robot - density - surges - in - europe - asia - and - americas. 2025

  22. [30]

    O. Liu, S. Y uan, and Z. Li. ‘‘A Survey on Sensor Technologies for Unmanned Ground V ehicles.’’ In:3rd Int’l Conf. Unmanned Systems (ICUS). 2020, pp. 638–645. doi: 10.1109/ICUS50048. 2020.9274845

  23. [31]

    P . Mittal. ‘‘A Comprehensive Survey of Deep Learning-Based Lightweight Object Detection Models for Edge Devices.’’ In: Artificial Intelligence Review57.9 (2024), p. 1. doi: 10.1007/ s10462-024-10877-1

  24. [32]

    Klein and D

    G. Klein and D. Murray. ‘‘Parallel Tracking and Mapping for Small AR Workspaces.’’ In:IEEE and ACM Int’l Symp. Mixed and Augmented Reality. 2007, pp. 225–234. doi: 10.1109/ISMAR. 2007.4538852

  25. [33]

    Mur-Artal, J

    R. Mur-Artal, J. M. M. Montiel, and J. D. Tardós. ‘‘ORB-SLAM: A V ersatile and Accurate Monocular SLAM System.’’ In:IEEE Trans. Robotics31.5 (2015), pp. 1147–1163. doi: 10.1109/TRO. 2015.2463671

  26. [34]

    Campos, R

    C. Campos, R. Elvira, J. J. G. Rodríguez, J. M. M. Montiel, and J. D. Tardós. ‘‘ORB-SLAM3: An Accurate Open-Source Library for Visual, Visual–Inertial, and Multimap SLAM.’’ In:IEEE Trans. Robotics37.6 (2021), pp. 1874–1890. doi: 10.1109/TRO.2021. 3075644

  27. [35]

    Galvez-López and J

    D. Galvez-López and J. D. Tardos. ‘‘Bags of Binary Words for Fast Place Recognition in Image Sequences.’’ In:IEEE Trans. Robotics 28.5 (2012), pp. 1188–1197. doi: 10.1109/TRO.2012.2197158

  28. [36]

    Kaess, H

    M. Kaess, H. Johannsson, R. Roberts, V . Ila, J. J. Leonard, and F. Dellaert. ‘‘iSAM2: Incremental Smoothing and Mapping Using the Bayes Tree.’’ In:The International J. Robotics Research31.2 (2012), pp. 216–235. doi: 10.1177/0278364911430419

  29. [37]

    Engel, T

    J. Engel, T. Schöps, and D. Cremers. ‘‘LSD-SLAM: Large-Scale Direct Monocular SLAM.’’ In:Computer Vision – ECCV 2014. Springer International Publishing, 2014, pp. 834–849. doi: 10. 1007/978-3-319-10605-2_54

  30. [38]

    Forster, L

    C. Forster, L. Carlone, F. Dellaert, and D. Scaramuzza. ‘‘On- Manifold Preintegration for Real-Time Visual–Inertial Odome- try.’’ In:IEEE Trans. Robotics33.1 (2017), pp. 1–21. doi: 10 . 1109/TRO.2016.2597321

  31. [39]

    Leutenegger, S

    S. Leutenegger, S. Lynen, M. Bosse, R. Siegwart, and P . Furgale. ‘‘Keyframe-Based Visual–Inertial Odometry Using Nonlinear Optimization.’’ In:The International J. Robotics Research34.3 (2015), pp. 314–334. doi: 10.1177/0278364914554813

  32. [40]

    T. Qin, P . Li, and S. Shen. ‘‘VINS-Mono: A Robust and V ersatile Monocular Visual-Inertial State Estimator.’’ In:IEEE Trans. Robotics34.4 (2018), pp. 1004–1020. doi: 10.1109/TRO.2018. 2853729

  33. [41]

    Z. Teed, L. Lipson, and J. Deng. ‘‘Deep Patch Visual Odome- try.’’ In:Adv. Neural Information Processing Systems36 (2023), pp. 39033–39051

  34. [42]

    M. N. Favorskaya. ‘‘Deep Learning for Visual SLAM: The State-of-the-Art and Future Trends.’’ In:Electronics12.9 (2023), p. 2006. doi: 10.3390/electronics12092006

  35. [43]

    D. Fox. ‘‘KLD-Sampling: Adaptive Particle Filters.’’ In:Advances in Neural Information Processing Systems. V ol. 14. MIT Press, 2001

  36. [44]

    Cummins and P

    M. Cummins and P . Newman. ‘‘FAB-MAP: Probabilistic Lo- calization and Mapping in the Space of Appearance.’’ In:The International J. Robotics Research27.6 (2008), pp. 647–665. doi: 10.1177/0278364908090961

  37. [45]

    Kim and A

    G. Kim and A. Kim. ‘‘Scan Context: Egocentric Spatial Descriptor for Place Recognition Within 3D Point Cloud Map.’’ In:IEEE/RSJ Int’l Conf. Intelligent Robots and Systems (IROS). 2018, pp. 4802–

  38. [46]

    doi: 10.1109/IROS.2018.8593953

  39. [47]

    R. Dubé, D. Dugas, E. Stumm, J. Nieto, R. Siegwart, and C. Cadena. ‘‘SegMatch: Segment Based Place Recognition in 3D Point Clouds.’’ In:IEEE Int’l Conf. Robotics and Automation (ICRA). 2017, pp. 5266–5272. doi: 10.1109/ICRA.2017.7989618

  40. [48]

    Kendall, M

    A. Kendall, M. Grimes, and R. Cipolla. ‘‘PoseNet: A Convolu- tional Network for Real-Time 6-DOF Camera Relocalization.’’ In:IEEE Int’l Conf. Computer Vision. 2015, pp. 2938–2946

  41. [49]

    Arandjelovic, P

    R. Arandjelovic, P . Gronat, A. Torii, T. Pajdla, and J. Sivic. ‘‘NetVLAD: CNN Architecture for Weakly Supervised Place Recognition.’’ In:IEEE Conf. Computer Vision and Pattern Recognition. 2016, pp. 5297–5307

  42. [50]

    Calinescu, L

    R. Calinescu, L. Grunske, M. Kwiatkowska, R. Mirandola, and G. Tamburrelli. ‘‘Dynamic QoS Management and Optimization in Service-Based Systems.’’ In:IEEE Trans. Software Engineering (TSE)37.3 (2011), pp. 387–409. doi: 10.1109/tse.2010.92

  43. [51]

    Kwiatkowska, G

    M. Kwiatkowska, G. Norman, and D. Parker. ‘‘PRISM 4.0: V erification of Probabilistic Real-Time Systems.’’ In:Computer Aided V erification. Springer, 2011, pp. 585–591. doi: 10.1007/ 978-3-642-22110-1_47

  44. [52]

    T.-Y . Lin, M. Maire, S. Belongie, J. Hays, P . Perona, D. Ramanan, P . Dollár, and C. L. Zitnick. ‘‘Microsoft COCO: Common Ob- jects in Context.’’ In:Computer Vision – ECCV 2014. Springer International Publishing, 2014, pp. 740–755. doi: 10.1007/978-3- 319-10602-1_48. 17 J. S...

  45. [53]

    Mallik, M

    A. Mallik, M. L. Gaopande, G. Singh, A. Ravindran, Z. Iqbal, S. Chao, H. Revalla, and V . Nagasamy. ‘‘Real-Time Detection and Avoidance of Obstacles in the Path of Autonomous V ehicles Using Monocular RGB Camera.’’ In:SAE International J. Advances and Current Practices in Mobi...

  46. [54]

    doi: 10.4271/2022-01-0074

  47. [55]

    Paszke, A

    A. Paszke, A. Chaurasia, S. Kim, and E. Culurciello.ENet: A Deep Neural Network Architecture for Real-Time Semantic Segmentation. 2016. arXiv: 1606.02147

  48. [56]

    C. Y u, J. Wang, C. Peng, C. Gao, G. Y u, and N. Sang. ‘‘BiSeNet: Bilateral Segmentation Network for Real-time Semantic Seg- mentation.’’ In:European Conf. Computer Vision (ECCV). 2018, pp. 325–341

  49. [57]

    R. P . K. Poudel, S. Liwicki, and R. Cipolla.Fast-SCNN: Fast Semantic Segmentation Network. 2019. arXiv: 1902.04502

  50. [58]

    O’Reilly Media, Inc

    B. Beyer, C. Jones, J. Petoff, and N. R. Murphy.Site Reliability Engineering: How Google Runs Production Systems. "O’Reilly Media, Inc.", 2016. isbn: 978-1-4919-5118-7

  51. [59]

    Url: https://github.com/NVIDIA- AI- IOT/jetbot

    NVIDIA-AI-IOT/Jetbot. Url: https://github.com/NVIDIA- AI- IOT/jetbot. 2026

  52. [60]

    Url: https://www

    AI Robot Kits from NVIDIA JetBot Partners. Url: https://www. nvidia.com/en- us/autonomous- machines/embedded- systems/ jetbot-ai-robot-kit/

  53. [61]

    Url: https : / / www

    Astra Series: Structured Light Cameras | Orbbec. Url: https : / / www. orbbec . com / products / structured - light - camera / astra - series/

  54. [62]

    Url: https : / / cyberbotics.com/

    Cyberbotics: Robotics Simulation with Webots. Url: https : / / cyberbotics.com/

  55. [63]

    Ayala, F

    A. Ayala, F. Cruz, D. Campos, R. Rubio, B. Fernandes, and R. Dazeley. ‘‘A Comparison of Humanoid Robot Simulators: A Quantitative Approach.’’ In:Joint IEEE 10th Int’l Conf. Devel- opment and Learning and Epigenetic Robotics (ICDL-EpiRob). 2020, pp. 1–6. doi: 10.1109/ICDL-EpiRo...

Pith tools

Reviewed July 14, 2026 · model on record in the stance chip above.