Pith. sign in

REVIEW 4 major objections 5 minor 46 references

Good Weights: Proactive, Adaptive Dead Reckoning Fusion for Continuous and Robust Visual SLAM

T0 review · 4 major / 5 minor · reviewed 2026-08-04 · deepseek-v4-flash

Pith's one-line read The Good Weights algorithm claims that adaptively scaling dead-reckoning priors by visual tracking health, across every module of visual SLAM, keeps optimization well-conditioned and lets the system complete trajectories through low-texture

desk verdict A credible, well-engineered robustness method for indirect visual SLAM; the headline results are strong but partly calibrated on the same benchmark, and the count-based health score has a plausible blind spot the paper does not test. read the letter →

arxiv 2509.22910 v2 pith:35XTY6N2 submitted 2025-09-26 cs.RO

classification cs.RO
keywords VisualSLAMdeadreckoningadaptivefusionfeaturetrackinghealthbundleadjustmentlow-textureenvironmentsmobilerobotnavigationposeestimation
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

Visual SLAM reconstructs a robot's path from camera images, so plain walls and dim lighting can break tracking. The Good Weights algorithm claims that a lightweight health score, computed from how many features are detected and tracked each frame, can continuously adjust the influence of dead reckoning (wheel and IMU odometry) on the pose estimate. When vision is healthy, dead reckoning stays nearly silent; when tracking degrades, its prior is strengthened to keep the optimization numerically stable and the pose continuous. The paper argues this adaptive gating must be applied not only to the tracking thread but to local and global bundle adjustment as well, and presents experiments where the result completes every test sequence while vision-only and fixed-fusion baselines drop out. If correct, it offers a practical way to make indoor robot navigation robust to visual degradation without retuning fusion weights per environment.

What carries the argument

The central mechanism is the adaptive weight α(Q_t), which interpolates in log-space between a minimum and maximum dead-reckoning information weight based on the tracking health score Q_t. Q_t is the weighted sum of clipped ratios of detected and tracked feature counts to preset reference values (600 and 120 in the experiments). This scalar is computed before optimization, making the gating proactive rather than a post-hoc reweighting; the paper treats it as a cheap proxy for the conditioning of the pose Hessian H(Q_t)=J_v^T Σ_v^{-1} J_v + α(Q_t) J_d^T Σ_{d0}^{-1} J_d. The same rule is reused in tracking, local BA, and global BA, with the local BA version replacing feature counts by a covisi

What would settle it

Run the method on a sequence with plentiful but degenerate features—for example, a long corridor with repetitive texture that yields many detections arranged on a plane or along one direction. If the pose error spikes while the reported quality score remains high, then the raw feature-count proxy fails to track the actual conditioning of the estimation problem, and the paper's central claim collapses.

Watch

Extended reading notes

Core claim

The paper's core discovery is that a single scalar quality score Q_t, built from raw detected and tracked feature counts, can regulate the information weight of a dead-reckoning prior through the log-space rule α(Q_t)=α_min(α_max/α_min)^{1-Q_t}, and that applying this adaptively weighted prior in feature matching, map-to-frame pose estimation, and local and global bundle adjustment keeps the whole SLAM hierarchy well-conditioned under visual degradation. The paper reports that this is the first configuration to achieve full trajectory completeness on eight robot navigation sequences, and that on a public indoor low-texture benchmark full completeness is reached as soon as dead-reckoning ente

Load-bearing premise

The whole method rests on the premise that the number of detected and tracked features, calibrated against one indoor dataset, faithfully measures how ill-conditioned the visual optimization is about to be; if high feature counts can still be geometrically degenerate or mis-associated, the dead-reckoning prior will not switch on when needed.

Editorial extensions

If this is right

  • If the central claim holds, low-texture indoor sequences that previously caused track loss can be completed continuously by augmenting a vision-only SLAM system with adaptively weighted dead reckoning.
  • Full trajectory completeness on the eight navigation sequences is achieved with an average RMSE of 0.87 m, while the vision-only baseline's completeness more than doubles at similar accuracy.
  • On the public indoor benchmark, full completeness is achieved first when dead-reckoning is applied to feature data association, and accuracy improves further when the prior is extended to pose estimation and local bundle adjustment.
  • Under repeat-run map reuse, frame-level pose accuracy converges toward keyframe-level accuracy (RMSE ratio near 1.0), suggesting the visual map can be reused reliably after the first pass.
  • Real-robot closed-loop navigation trials maintain stable maps and an average trajectory RMSE of 0.54 m at 24.6 ms median tracking latency, indicating real-time operation.

Reading between the lines

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

  • A testable extension: run the same health-gated weighting on sequences where features are abundant but geometrically degenerate (e.g., a corridor textured with repetitive coplanar pattern); if the pose error grows while Q_t stays near 1, the raw-count proxy needs to be replaced by a structural degeneracy measure.
  • The same log-space interpolation could be applied to other motion priors, such as inertial pre-integration factors or learned odometry, and to other hierarchical estimators where an upstream module's conditioning determines downstream success.
  • Because the method trusts dead reckoning only in short windows, its promise is continuity, not long-horizon accuracy during prolonged visual outage; a sustained loss of vision would still drift by the DR sensor's accumulated error, so a natural boundary condition is the expected outage length.
  • The health score is computed from front-end statistics, so it could in principle be predicted from image appearance alone, which would let the system decide when to activate the prior before a single feature is matched.
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

4 major / 5 minor

Summary. This paper presents Good Weights (GW), an adaptive fusion framework that inserts dead-reckoning (DR) priors into a feature-based visual SLAM pipeline. The core idea is a scalar tracking-quality score Q_t (Eq. 5), computed from detected and tracked feature counts, which controls the DR information weight alpha(Q_t) in log-space (Eq. 3). DR is applied in data association, map-to-frame pose estimation, local bundle adjustment, and global bundle adjustment. Experiments on the CID benchmark, eight office navigation sequences, and closed-loop robot trials report improved trajectory completeness and RMSE relative to vision-only and fixed-fusion baselines.

Significance. If the reported gains are robust, GW is a practical and low-cost answer to low-texture visual SLAM failure. Its strengths are the transparent formulation, the explicit integration across the whole SLAM hierarchy, and the held-out office and closed-loop evaluations. The main limits are that the count-based health score is validated only against the dataset used to set its thresholds, and that Eq. (5) is not shown to track Hessian conditioning in the high-count/degenerate-feature regime that Eq. (4) nominally protects against. These limit the strength of the cross-domain robustness claim but do not invalidate the empirical results.

major comments (4)
  1. [§III-B.3, Eqs. (4)-(5)] The central claim is that GW keeps the pose Hessian well-conditioned. Eq. (4) defines the relevant quantity as J_v^T Sigma_v^-1 J_v + alpha(Q_t) J_d^T Sigma_d0^-1 J_d, but Eq. (5) replaces the visual information matrix with raw feature counts. The paper explicitly says Jacobian conditioning can be misleading (Sec. III-B.3) and therefore uses counts; however, no evidence is provided that high counts imply a well-conditioned visual Hessian. A frame with many coplanar, clustered, or mismatched features can have Q_t near 1 while J_v^T Sigma_v^-1 J_v is near-degenerate along a motion direction; in that case alpha stays at alpha_min and the DR prior is not up-weighted, so the failure mechanism GW targets persists. The CID and office sequences do not appear to exercise this regime. Please add a synthetic or real test with high-count low-conditioning geometry, or at least report the correlation
  2. [§III-B.4, Fig. 4, §IV-A] The quantitative parameters of the adaptive rule are fitted on CID sequence floor13_1 (alpha_min, alpha_max) and on CID data in Fig. 3 (N_r_det = 600, N_r_trk = 120, omega_1 = omega_2 = 0.5), and the same CID benchmark is then used for the headline head-to-head results in Fig. 6. This is a circular element: the absolute RMSE/completeness numbers on CID are partly self-validating. The office navigation sequences are held out and are a useful check, but they do not repair the CID-specific claim. Please include a sensitivity study (e.g., vary alpha bounds and thresholds within a factor of 3) and/or a leave-one-sequence-out or cross-dataset validation.
  3. [§IV-A, Fig. 6, PGD-VIO comparison] The comparison with PGD-VIO uses numbers reported in [42] rather than from a local run, and the paper excludes two CID multi-floor sequences because the capture device was lifted. It is not stated whether the completeness percentages from [42] are recalculated on the same 20-sequence subset. If they are not, the comparison in Fig. 6 is not on equal footing. Please clarify the exact protocol used for the reported baselines and, if possible, recompute or state the subset used for each method.
  4. [§III-C.3, Eq. (7)] The adaptive DR weight for LBA uses a different quality score Q_ij than Eq. (5), and the 'windowed smoothing strategy' that distributes weights to adjacent keyframes is not specified. Since GW integrated into LBA is one of the paper's headline variants, the reproducibility of this component is not supported. Please provide the explicit weighting formula, the size/shape of the window, and the exact role of C_ref.
minor comments (5)
  1. [§IV-A.1] The notation for quality weights changes from omega_1, omega_2 in Eq. (5) to w1, w2 in the experiments. Please unify.
  2. [Fig. 3] The caption says 'Feature Number' but the text distinguishes detected vs tracked counts; no error bars or correlation coefficients are reported. Please include variance or at least the number of frames used.
  3. [§IV-B] SVO-Pro and DSOL share reference [46]; SVO-Pro needs its own reference. Also, DSOL is described as 'excluded from comparison' because it consistently fails, which should be stated in the comparison-set definition.
  4. [§IV-C] The closed-loop evaluation has no baseline methods, so the claim that the approach is 'reliable for practical navigation' is based on absolute RMSE/latency only. Please frame this as a pilot demonstration rather than a comparative result.
  5. [§IV-A] The text notes the method 'currently assumes available wheel odometry'; this should be stated in the abstract and limitations, since it bounds the applicability.

Circularity Check

2 steps flagged · score 4.0 of 10

Partial benchmark circularity: DR weight and health-score parameters are fitted on CID, then CID results are reported as validation; independent office trials keep the central claim from collapsing entirely.

  1. fitted input called prediction [Sec. III-B.4 (DR bounds); Sec. IV-A (CID evaluation)]
    "We adopt a lightweight, data-driven strategy, using the floor13 1 sequence of the CID dataset with nominal DR covariances Σd0 set to 0.004m/frame (translation) and 0.1 ◦/frame (rotation). Safe α bounds are obtained through empirical sweeps, with each candidate evaluated by trajectory accuracy (frame-level RMSE) as shown in Fig. 4."

    The α bounds in Eq. (3) are selected by sweeping on CID's floor13_1 sequence using the same frame-RMSE metric that the paper later reports on the CID benchmark. Sec. IV-A then evaluates GW on 'the public CID benchmark sequences' and presents GW's RMSE/completeness in Fig. 6 with those bounds fixed. The CID performance is therefore partly a fitted result, not a prediction on unseen data: the calibration metric (frame RMSE on CID) is the headline evaluation metric on the same dataset, and floor13_1 is not among the excluded multi-floor sequences. The claim that the bounds 'preserve generality' does not remove the fact that the test set includes the calibration sequence.

  2. fitted input called prediction [Sec. III-B.3 (Eq. 5 thresholds); Sec. IV-A (evaluation)]
    "In our evaluation of a vision-only SLAM system [3] on the CID dataset [37] which contains challenging low-texture scenes, a clear negative correlation is observed: as the number of detected and tracked features decreases, the frame-level pose RMSE increases (Fig. 3). ... These results demonstrate that the detected feature count Ndet and the tracked feature count Ntrk serve as reliable indicators of visual tracking quality. To map these tracking variables to a quality score, define Qt = ... Each ratio is clipped to [0,1]."

    The target constants N_r_det=600 and N_r_trk=120 in Eq. (5) are read off CID Fig. 3, where the paper observes RMSE growing quadratically below 600 detections and diminishing returns beyond 120 tracked features. The same CID benchmark is then used to measure the method's success (Fig. 6 and Table I). Thus the health score that controls DR engagement is calibrated on the exact dataset used for the claimed 'full completeness' result. This is a fitted proxy evaluated on its fitting set; it does not demonstrate transfer of the score thresholds. The later office navigation experiments provide some external validation but do not undo the CID circularity.

full rationale

The paper's load-bearing claim is an engineering result, not a mathematical derivation: Eq. (3) is a chosen interpolation rule and Eq. (5) is a chosen heuristic scoring rule. The genuinely circular element is benchmark calibration: the DR bounds (α_min=1e-1, α_max=1e3) are selected by a sweep on CID floor13_1 using the frame-RMSE metric later reported on the CID benchmark, and the feature-count targets in Eq. (5) are read off CID Fig. 3. Reporting 'full completeness' and RMSE values on CID after choosing these parameters on CID is partially self-validating: the numbers in Fig. 6 are not a prediction on unseen data with respect to those parameters. This is data leakage of the fitted-input-called-prediction variety, but it is not constructional equivalence, because the parameters do not fully determine the reported outcomes and because independent office navigation and closed-loop trials use the same fixed parameters and provide external evidence. The skeptic's concern that high feature counts can coexist with a degenerate Hessian (Qt≈1 while J_v^T Σ_v^-1 J_v is ill-conditioned) is a correctness/failure-mode risk, not a logical circularity, since the paper nowhere proves that Qt rank-orders conditioning. No load-bearing self-citation chain was found: citation [3] is the baseline system being augmented, [27] supplies a benchmarking protocol, and [39] is only cited for information-theoretic Hessian measures that the method explicitly declines to use. Score 4: partial benchmark circularity, with the central robustness claim retaining independent empirical content.

Assumptions & free parameters 7 free parameters · 6 assumptions · 0 invented entities

The central claim rests on 7 fitted or hand-chosen constants and 6 assumptions. The most consequential fitted constants are the alpha bounds and the feature-count thresholds, calibrated on the CID benchmark and then fixed for the CID evaluation; the nominal DR covariance is not separately identifiable from the alpha bounds. The quality-proxy and calibration assumptions are load-bearing: if feature counts do not track optimization conditioning, or if the DR-to-camera calibration is biased, the adaptive gate either stays closed when it should open, or injects biased priors. Standard bundle-adjustment mathematics is the only standard_math axiom. No invented physical entities are introduced.

free parameters (7)
  • alpha_min (DR weight lower bound) = 10^-1 = 0.1
    Set by empirical sweep on CID sequence floor13 1 (Sec. III-B.4, Fig. 4): log(alpha) = -1.0 is where DR 'begins to take effect'. Fixed across all later experiments.
  • alpha_max (DR weight upper bound) = 10^3 = 1000
    Same sweep (Fig. 4): log(alpha) = 3.0 is where DR 'dominates the optimization'. Caps DR influence during total visual failure.
  • nominal DR covariance Sigma_d0 = 0.004 m/frame translation, 0.1 deg/frame rotation
    Chosen for the floor13 1 sweep; since the effective prior is alpha(Qt) * Sigma_d0^{-1}, the nominal covariance and the alpha bounds are jointly fitted and not separately identifiable.
  • N_r_det (detection target) = 600 features
    Read off the CID RMSE-versus-feature-count trend (Fig. 3): RMSE 'grows quadratically once the count drops below 600'. Used in Eq. (5).
  • N_r_trk (tracking target) = 120 features
    Same figure: tracking count 'shows lower sensitivity once it exceeds 120'. Used in Eq. (5).
  • omega_1, omega_2 (quality weights) = 0.5, 0.5
    Stated in Sec. IV as w1 = 0.5, w2 = 0.5; no justification beyond equal weighting of detection and tracking evidence.
  • C_ref (reference covisibility count) = 20 initially, then online-adjusted
    Sec. III-C.3 and Sec. IV: initialized at 20, replaced online by the median connection count of 'well-tracked' keyframes; the rule for selecting well-tracked keyframes is not specified.
assumptions (6)
  • standard math SE(3) pose representation, reprojection error, and weighted nonlinear least squares (Gauss-Newton/LM) as in Eq. (1)
    The entire formulation assumes standard bundle-adjustment machinery from [34] and the GFGG pipeline [3].
  • domain assumption Dead-reckoning residuals are unbiased Gaussian errors with fixed covariance Sigma_d0 over short horizons
    Eqs. (1)-(2) place DR as a quadratic prior; wheel slip or odometry bias would violate the model, and the adaptive gate cannot detect it.
  • domain assumption Feature detection and tracking counts are sufficient proxies for optimization conditioning
    Eq. (5) defines Qt from counts only; the paper concedes conditioning analysis 'can be misleading in visual SLAM' (Sec. III-B.3) and validates the proxy only on CID (Fig. 3).
  • domain assumption Sensor calibration and time synchronization are correct
    Stated at Sec. III-B: 'assuming that sensor calibration and time synchronization have been properly handled'; a biased DR-to-camera extrinsics would inject systematic pose error.
  • domain assumption Ground-truth trajectories from 2D LiDAR SLAM-Toolbox, spline-smoothed, are accurate enough for RMSE scoring
    Sec. IV-B relies on this for the eight navigation sequences; plausible in planar office environments but not independently verified.
  • domain assumption Reported PGD-VIO baseline numbers from [42] were produced under the same evaluation protocol
    Sec. IV-A: code is not public, so comparison rests on [42]'s reported 5-run protocol and failure thresholds.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Good Weights: Proactive, Adaptive Dead Reckoning Fusion for Continuous and Robust Visual SLAM." pith.science (2026). https://pith.science/paper/35XTY6N2

@misc{pith2026250922910,
  author       = {Pith},
  title        = {Pith review of: Good Weights: Proactive, Adaptive Dead Reckoning Fusion for Continuous and Robust Visual SLAM},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/35XTY6N2}},
  note         = {Machine review of arXiv:2509.22910}
}
read the original abstract

Given that Visual SLAM relies on appearance cues for localization and scene understanding, texture-less or visually degraded environments (e.g., plain walls or low lighting) lead to poor pose estimation and track loss. However, robots are typically equipped with sensors that provide some form of dead reckoning odometry with reasonable short-time performance but unreliable long-time performance. The Good Weights (GW) algorithm described here provides a framework to adaptively integrate dead reckoning (DR) with passive visual SLAM for continuous and accurate frame-level pose estimation. Importantly, it describes how all modules in a comprehensive SLAM system must be modified to incorporate DR into its design. Adaptive weighting increases DR influence when visual tracking is unreliable and reduces when visual feature information is strong, maintaining pose track without overreliance on DR. Good Weights yields a practical solution for mobile navigation that improves visual SLAM performance and robustness. Experiments on collected datasets and in real-world deployment demonstrate the benefits of Good Weights.

Figures

Figures reproduced from arXiv: 2509.22910 by the authors.

Figure 1
Figure 1. Performance comparison between a visual SLAM method, GFGG, [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. System overview following the design of [1], [3], with newly [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Relationship Between Frame Pose RMSE and Tracking Statistics on [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: DR weight sweeping study on sequence floor13 1. Two segments with poor (case 0) and good (case 1) tracking were evaluated independently, each repeated five times. Median RMSE values are connected by the blue dashed line. Dotted green lines show the values of αmin and α…
Figure 5
Figure 5. Figure 5: Adaptive DR weighting in bundle adjustment. Stronger DR con [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]
Figure 7
Figure 7. Figure 7: Trajectories of sequences collected in an office environment, with [PITH_FULL_IMAGE:figures/full_fig_p006_7.png]
Figure 9
Figure 9. Figure 9: Closed-loop navigation evaluation in an office environment. Top [PITH_FULL_IMAGE:figures/full_fig_p007_9.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

46 extracted references · 4 linked inside Pith

  1. [42]

    PGD-VIO: A Plane- Aided RGB-D Inertial Odometry With Graph-Based Drift Suppres- sion,

    Y . Zhang, F. Tang, Z. Xu, Y . Wu, and P. Ma, “PGD-VIO: A Plane- Aided RGB-D Inertial Odometry With Graph-Based Drift Suppres- sion,”IEEE Robotics and Automation Letters, vol. 10, no. 5, pp. 4276– 4283, 2025

  2. [1]

    ORB-SLAM2: an open-source SLAM system for monocular, stereo and RGB-D cameras,

    R. Mur-Artal and J. D. Tard ´os, “ORB-SLAM2: an open-source SLAM system for monocular, stereo and RGB-D cameras,”IEEE T-RO, vol. 33, no. 5, pp. 1255–1262, 2017

  3. [2]

    ORB-SLAM3: An Accurate Open-Source Library for Visual, Visual–Inertial, and Multimap SLAM,

    C. Campos, R. Elvira, J. J. G. Rodr ´ıguez, J. M. M. Montiel, and J. D. Tard ´os, “ORB-SLAM3: An Accurate Open-Source Library for Visual, Visual–Inertial, and Multimap SLAM,”IEEE T-RO, vol. 37, no. 6, pp. 1874–1890, 2021

  4. [3]

    Good Graph to Optimize: Cost- Effective, Budget-Aware Bundle Adjustment in Visual SLAM,

    Y . Zhao, J. S. Smith, and P. A. Vela, “Good Graph to Optimize: Cost- Effective, Budget-Aware Bundle Adjustment in Visual SLAM,”ArXiv, vol. abs/2008.10123, 2020

  5. [4]

    Recent advances in slam for degraded environments a review,

    W. Wang, Q. Zhang, Y . Hu, M. Gallay, W. Zheng, and J. Guo, “Recent advances in slam for degraded environments a review,”IEEE Sensors Journal, 2025

  6. [5]

    Robust SLAM Systems: Are We There Yet?

    M. Bujanca, X. Shi, M. Spear, P. Zhao, B. Lennox, and M. Luj ´an, “Robust SLAM Systems: Are We There Yet?” inIEEE/RSJ IROS, 2021

  7. [6]

    Wheel-SLAM: Simultaneous localization and terrain mapping using one wheel-mounted imu,

    Y . Wu, J. Kuang, X. Niu, J. Behley, L. Klingbeil, and H. Kuhlmann, “Wheel-SLAM: Simultaneous localization and terrain mapping using one wheel-mounted imu,”IEEE Robotics and Automation Letters, 2022

  8. [7]

    VINS-Mono: A Robust and Versatile Monocular Visual-Inertial State Estimator,

    T. Qin, P. Li, and S. Shen, “VINS-Mono: A Robust and Versatile Monocular Visual-Inertial State Estimator,”IEEE Transactions on Robotics, vol. 34, no. 4, pp. 1004–1020, 2018

Show all 46 references
  1. [8]

    A multi-state constraint Kalman filter for vision-aided inertial navigation,

    A. I. Mourikis and S. I. Roumeliotis, “A multi-state constraint Kalman filter for vision-aided inertial navigation,” inIEEE ICRA, 2007, pp. 3565–3572

  2. [9]

    Open- VINS: A Research Platform for Visual-Inertial Estimation,

    P. Geneva, K. Eckenhoff, W. Lee, Y . Yang, and G. P. Huang, “Open- VINS: A Research Platform for Visual-Inertial Estimation,”IEEE International Conference on Robotics and Automation, pp. 4666–4672, 2020

  3. [10]

    Kimera: an open- source library for real-time metric-semantic localization and mapping,

    A. Rosinol, M. Abate, Y . Chang, and L. Carlone, “Kimera: an open- source library for real-time metric-semantic localization and mapping,” inIEEE ICRA, 2020

  4. [11]

    Visual-inertial-wheel slam with high- accuracy localization measurement for wheeled robots on complex terrain,

    J. Zheng, K. Zhou, and J. Li, “Visual-inertial-wheel slam with high- accuracy localization measurement for wheeled robots on complex terrain,”Measurement, vol. 243, p. 116356, 2025

  5. [12]

    VINS on wheels,

    K. Wu, C. X. Guo, G. A. Georgiou, and S. I. Roumeliotis, “VINS on wheels,”IEEE International Conference on Robotics and Automation, pp. 5155–5162, 2017

  6. [13]

    Visual- Inertial-Wheel Odometry with Online Calibration,

    W. Lee, K. Eckenhoff, Y . Yang, P. Geneva, and G. P. Huang, “Visual- Inertial-Wheel Odometry with Online Calibration,”IEEE/RSJ Interna- tional Conference on Intelligent Robots and Systems, pp. 4559–4566, 2020

  7. [14]

    Adaptive estimation of uav altitude in complex indoor environments using degraded and time-delayed measurements with time-varying uncertain- ties,

    V . Pritzl, M. Vrba, C. Tortorici, R. Ashour, and M. Saska, “Adaptive estimation of uav altitude in complex indoor environments using degraded and time-delayed measurements with time-varying uncertain- ties,”Robotics and Autonomous Systems, vol. 160, p. 104315, 2023

  8. [15]

    LiDAR-Visual-Inertial Tightly- coupled Odometry with Adaptive Learnable Fusion Weights,

    V . Hulchuk, J. Bayer, and J. Faigl, “LiDAR-Visual-Inertial Tightly- coupled Odometry with Adaptive Learnable Fusion Weights,” in IEEE/RSJ International Conference on Intelligent Robots and Systems, 2024, pp. 5641–5647

  9. [16]

    Ground-fusion: A low-cost ground slam system robust to corner cases,

    J. Yin, A. Li, W. Xi, W. Yu, and D. Zou, “Ground-fusion: A low-cost ground slam system robust to corner cases,” inIEEE International Conference on Robotics and Automation, 2024, pp. 8603–8609

  10. [17]

    Lvio-Fusion: A Self-adaptive Multi-sensor Fusion SLAM Framework Using Actor-critic Method,

    Y . Jia, H. Luo, F. Zhao, G. Jiang, Y . Li, J. Yan, and Z. Jiang, “Lvio-Fusion: A Self-adaptive Multi-sensor Fusion SLAM Framework Using Actor-critic Method,”IEEE/RSJ International Conference on Intelligent Robots and Systems, pp. 286–293, 2021

  11. [18]

    Rvwo: A robust visual-wheel slam system for mobile robots in dynamic environments,

    J. Mahmoud, A. Penkovskiy, H. T. Long Vuong, A. Burkov, and S. Kolyubin, “Rvwo: A robust visual-wheel slam system for mobile robots in dynamic environments,” inIEEE/RSJ International Confer- ence on Intelligent Robots and Systems, 2023, pp. 3468–3474

  12. [19]

    Switch-SLAM: Switching-Based LiDAR-Inertial- Visual SLAM for Degenerate Environments,

    J. Lee, R. Komatsu, M. Shinozaki, T. Kitajima, H. Asama, Q. An, and A. Yamashita, “Switch-SLAM: Switching-Based LiDAR-Inertial- Visual SLAM for Degenerate Environments,”IEEE Robotics and Automation Letters, vol. 9, pp. 7270–7277, 2024

  13. [20]

    Selective Kalman Filter: When and How to Fuse Multi- Sensor Information to Overcome Degeneracy in SLAM,

    J. Xu, G. Huang, W. Yu, X. Zhang, L. Zhao, R. Li, S. Yuan, and L. Xie, “Selective Kalman Filter: When and How to Fuse Multi- Sensor Information to Overcome Degeneracy in SLAM,”ArXiv, vol. abs/2412.17235, 2024

  14. [21]

    Informed, Constrained, Aligned: A Field Analysis on Degeneracy-Aware Point Cloud Registration in the Wild,

    T. Tuna, J. Nubert, P. Pfreundschuh, C. Cadena, S. Khattak, and M. Hutter, “Informed, Constrained, Aligned: A Field Analysis on Degeneracy-Aware Point Cloud Registration in the Wild,”IEEE Trans- actions on Field Robotics, vol. 2, pp. 485–515, 2024

  15. [22]

    On degeneracy of optimization- based state estimation problems,

    J. Zhang, M. Kaess, and S. Singh, “On degeneracy of optimization- based state estimation problems,”IEEE International Conference on Robotics and Automation, pp. 809–816, 2016

  16. [23]

    SuperLoc: The Key to Robust Lidar-Inertial Localization Lies in Predicting Alignment Risks Superodometry.Com/SuperLoc,

    S. Zhao, H. Zhu, Y . Gao, B. Kim, Y . Qiu, A. M. Johnson, and S. Scherer, “SuperLoc: The Key to Robust Lidar-Inertial Localization Lies in Predicting Alignment Risks Superodometry.Com/SuperLoc,” IEEE International Conference on Robotics and Automation, pp. 14 080–14 086, 2024

  17. [24]

    DAMS- LIO: A Degeneration-Aware and Modular Sensor-Fusion LiDAR- inertial Odometry,

    F. Han, H. Zheng, W. Huang, R. Xiong, Y . Wang, and Y . Jiao, “DAMS- LIO: A Degeneration-Aware and Modular Sensor-Fusion LiDAR- inertial Odometry,”2023 IEEE International Conference on Robotics and Automation (ICRA), pp. 2745–2751, 2023

  18. [25]

    Stereo visual iner- tial pose estimation based on feedforward and feedbacks,

    S. Chen, Y . Feng, C.-Y . Wen, Y . Zou, and W. Chen, “Stereo visual iner- tial pose estimation based on feedforward and feedbacks,”IEEE/ASME Transactions on Mechatronics, vol. 28, no. 6, pp. 3562–3572, 2023

  19. [26]

    Visual-Inertial Monocular SLAM With Map Reuse,

    R. Mur-Artal and J. D. Tard ´os, “Visual-Inertial Monocular SLAM With Map Reuse,”IEEE Robotics and Automation Letters, vol. 2, no. 2, pp. 796–803, 2017

  20. [27]

    Task-driven SLAM Bench- marking For Robot Navigation,

    Y . Du, S. Feng, C. G. Cort, and P. A. Vela, “Task-driven SLAM Bench- marking For Robot Navigation,”arXiv preprint arXiv:2409.16573, 2024

  21. [28]

    A robust and modular multi-sensor fusion approach applied to MA V navigation,

    S. Lynen, M. W. Achtelik, S. Weiss, M. Chli, and R. Siegwart, “A robust and modular multi-sensor fusion approach applied to MA V navigation,” inIEEE/RSJ international conference on intelligent robots and systems. IEEE, 2013, pp. 3923–3929

  22. [29]

    Holistic Fusion: Task- and Setup-Agnostic Robot Localization and State Estimation with Factor Graphs,

    J. Nubert, T. Tuna, J. Frey, C. Cadena, K. J. Kuchenbecker, S. Khattak, and M. Hutter, “Holistic Fusion: Task- and Setup-Agnostic Robot Localization and State Estimation with Factor Graphs,”ArXiv, vol. abs/2504.06479, 2025

  23. [30]

    EKF-LOAM: An Adaptive Fusion of LiDAR SLAM With Wheel Odometry and Inertial Data for Confined Spaces With Few Geometric Features,

    G. P. C. J ´unior, A. M. C. Rezende, V . R. F. Miranda, R. Fernandes, H. Azp ´urua, A. A. Neto, G. Pessin, and G. M. Freitas, “EKF-LOAM: An Adaptive Fusion of LiDAR SLAM With Wheel Odometry and Inertial Data for Confined Spaces With Few Geometric Features,” IEEE Transactions o...

  24. [31]

    Numerical optimization,

    J. Nocedal and S. J. Wright, “Numerical optimization,” inFundamen- tal Statistical Inference, 2018

  25. [32]

    The levenberg-marquardt algorithm,

    A. Ranganathan, “The levenberg-marquardt algorithm,”Tutoral on LM algorithm, vol. 11, no. 1, pp. 101–110, 2004

  26. [33]

    A. R. Conn, N. I. Gould, and P. L. Toint,Trust region methods. SIAM, 2000

  27. [34]

    Bundle adjustment—a modern synthesis,

    B. Triggs, P. F. McLauchlan, R. I. Hartley, and A. W. Fitzgibbon, “Bundle adjustment—a modern synthesis,” inInternational workshop on vision algorithms. Springer, 1999, pp. 298–372

  28. [35]

    MAC-VO: Metrics-Aware Covariance for Learning-Based Stereo Visual Odome- try,

    Y . Qiu, Y . Chen, Z. Zhang, W. Wang, and S. A. Scherer, “MAC-VO: Metrics-Aware Covariance for Learning-Based Stereo Visual Odome- try,”IEEE International Conference on Robotics and Automation, pp. 3803–3814, 2024

  29. [36]

    Present and future of slam in extreme environments: The darpa subt challenge,

    K. Ebadi, L. Bernreiter, H. Biggie, G. Catt, Y . Chang, A. Chatterjee, C. E. Denniston, S.-P. Desch ˆenes, K. Harlow, S. Khattaket al., “Present and future of slam in extreme environments: The darpa subt challenge,”IEEE Transactions on Robotics, vol. 40, pp. 936–959, 2023

  30. [37]

    CID-SIMS: Complex indoor dataset with semantic information and multi-sensor data from a ground wheeled robot viewpoint,

    Y . Zhang, N. An, C. Shi, S. Wang, H. Wei, P. Zhang, X. Meng, Z. Sun, J. Wang, W. Lianget al., “CID-SIMS: Complex indoor dataset with semantic information and multi-sensor data from a ground wheeled robot viewpoint,”The International Journal of Robotics Research, 2023

  31. [38]

    Attention and Anticipation in Fast Visual- Inertial Navigation,

    L. Carlone and S. Karaman, “Attention and Anticipation in Fast Visual- Inertial Navigation,”IEEE Transactions on Robotics, vol. 35, pp. 1–20, 2019

  32. [39]

    Good feature matching: Towards accurate, robust VO/VSLAM with low latency,

    Y . Zhao and P. A. Vela, “Good feature matching: Towards accurate, robust VO/VSLAM with low latency,”IEEE T-RO, vol. 36, no. 3, pp. 657–675, 2020

  33. [40]

    AutoInspect: Toward Long-Term Au- tonomous Inspection and Monitoring,

    M. Staniaszek, T. Flatscher, J. Rowell, H. Niu, W. Liu, Y . You, M. Gadd, M. Mattamala, A. Schutz, D. de Martini, L. Pitt, R. Skilton, M. Fallon, and N. Hawes, “AutoInspect: Toward Long-Term Au- tonomous Inspection and Monitoring,”IEEE Transactions on Field Robotics, vol. 2, p...

  34. [41]

    Review of robotic infrastructure inspec- tion systems,

    D. Lattanzi and G. Miller, “Review of robotic infrastructure inspec- tion systems,”Journal of Infrastructure Systems, vol. 23, no. 3, p. 04017004, 2017

  35. [43]

    SLAM Toolbox: SLAM for the dynamic world,

    S. Macenski and I. Jambrecic, “SLAM Toolbox: SLAM for the dynamic world,”Journal of Open Source Software, vol. 6, no. 61, p. 2783, 2021

  36. [44]

    Continuous-Time Visual-Inertial Odometry for Event Cameras,

    E. Mueggler, G. Gallego, H. Rebecq, and D. Scaramuzza, “Continuous-Time Visual-Inertial Odometry for Event Cameras,” IEEE Transactions on Robotics, vol. 34, pp. 1425–1440, 2017

  37. [45]

    [Online]

    ROS Navigation Tutorial. [Online]. Available: https://wiki.ros.org/ navigation/Tutorials/RobotSetup

  38. [46]

    DSOL: A Fast Direct Sparse Odometry Scheme,

    C. Qu, S. S. Shivakumar, I. D. Miller, and C. J. Taylor, “DSOL: A Fast Direct Sparse Odometry Scheme,” inIEEE/RSJ IROS, 2022, pp. 10 587–10 594

Pith tools

Reviewed August 4, 2026 · model on record in the stance chip above.