Pith. sign in

REVIEW 3 major objections 6 minor 41 references

Swapping only the noise matrices of one continuous underwater filter when motion regimes change cuts translation RMSE from 0.488 m to 0.471 m on held-out pool runs.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · grok-4.5

2026-07-14 10:34 UTC pith:MNBOPONG

load-bearing objection Careful, modest pool result: gated Q/R swaps on one continuous EKF beat a strong global baseline by ~3.5 cm with honest protocol, but the situation-matching story is thinner than the abstract implies. the 3 major comments →

arxiv 2607.10597 v1 pith:MNBOPONG submitted 2026-07-12 cs.RO cs.SYeess.SY

Underwater Dead Reckoning with Deployable Situation-Triggered Covariance Scheduling

Akshay Naik (1) , Ramavarapu S. Sreenivas (2) , Dustin Nottage (3) , Ahmet Soylemezoglu (3) ((1) University of Illinois Urbana-Champaign , Department of Electrical , Computer Engineering , (2) University of Illinois Urbana-Champaign , Department of Industrial
show 4 more authors
Enterprise Systems Engineering (3) U.S. Army Engineer Research Development Center Construction Engineering Research Laboratory)
This is my paper
classification cs.RO cs.SYeess.SY
keywords dead reckoningDoppler velocity logsituation-aware estimationunderwater navigationextended Kalman filtercovariance schedulingBlueROV2
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.

Underwater robots often lose external positioning and must dead-reckon from inertial sensors and Doppler velocity. A single fixed noise model for the filter works poorly when the vehicle turns, transitions, or loses clean bottom lock, because those regimes have different uncertainty. This paper shows that one continuous error-state filter can stay running while a lightweight onboard classifier merely swaps pre-calibrated process- and measurement-noise matrices when it is confident about the current situation. The state, covariance history, dynamics, and measurement models never reset. Calibration uses sparse AprilTag poses offline; at runtime the estimator is strictly vision-free. On four held-out pool runs the scheduled filter beats the identical backbone with one global profile on every run, and a paired bootstrap confirms a small but consistent translation gain while orientation stays essentially unchanged. The practical point is that modest, deployable accuracy can come from situation-aware covariance scheduling rather than from banks of filters or continuous re-tuning.

Core claim

Situation-triggered covariance scheduling on a single continuous calibrated error-state EKF yields a modest but consistent vision-free dead-reckoning improvement: label-weighted per-run translation RMSE falls from 0.488 m to 0.471 m across four held-out BlueROV2 pool runs, every run favors the scheduled method, and a 10-second-segment paired bootstrap gives a candidate-minus-baseline difference of −0.017 m with 95 % CI [−0.024, −0.008] m, while orientation error remains essentially unchanged.

What carries the argument

ST-CAR-EKF: one continuous error-state EKF whose only runtime change is a confidence-gated swap of pre-calibrated fixed Q and R matrices selected by an onboard probabilistic situation trigger; state estimate and covariance history persist across every switch.

Load-bearing premise

The noise profiles and confidence gates fitted on a handful of supervised pool runs will still match the motion regimes of held-out runs from the same pool, and changing only the noise matrices is enough to capture those regime differences.

What would settle it

On a new held-out set of the same BlueROV2 pool protocol, or on open-water runs with currents and acoustic degradation, either the scheduled filter no longer beats the identical global-profile backbone on label-weighted translation RMSE, or the bootstrap confidence interval for the difference includes zero or becomes positive.

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

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. The paper proposes ST-CAR-EKF, a vision-free underwater dead-reckoning estimator for a BlueROV2 that keeps one continuous error-state EKF and swaps only pre-calibrated process- and measurement-noise matrices (Q/R) when an onboard probabilistic situation trigger exceeds validation-selected confidence gates. Offline AprilTag supervision is used solely to fit a one-time DVL yaw correction, situation labels, noise profiles, and the trigger; runtime uses only IMU/DVL/altitude/control streams. On four held-out pool bags the method reduces label-weighted per-run translation RMSE from 0.488 m (global CAR-EKF) to 0.471 m, with a paired 10 s-segment bootstrap delta of −0.017 m (95% CI [−0.024, −0.008] m) and essentially unchanged orientation error. The authors emphasize that state and covariance are never reset and that the scheduling policy is frozen on validation before test.

Significance. If the result holds, the contribution is a carefully scoped, deployable demonstration that discrete, confidence-gated covariance scheduling inside a single continuous filter can yield a small but statistically supported dead-reckoning gain without estimator banks or resets. Strengths include bag-level train/val/test splits, validation-only policy freeze, label-weighted per-run metrics, paired bootstrap uncertainty, alignment-independent relative-pose checks, and transparent reporting of oracle/ungated failures and commercial black-box comparisons. The absolute effect is modest (~3.5% relative T-RMSE reduction on a pool benchmark), so significance is incremental rather than transformative, but the methodology and negative controls make it a useful reference point for situation-aware INS/DVL filtering under sparse supervision.

major comments (3)
  1. [Table III, §VII-B, §V-E] Table III and §VII-B: Oracle GT fixed-matrix routing is worse than the global profile on held-out test (0.494 m vs 0.488 m), while only the validation-gated trigger transfers a gain (0.471 m). This undercuts the central mechanistic claim that pre-calibrated situation profiles better match regime-dependent uncertainty. Please add a load-bearing analysis of where the gated gain actually accrues (e.g., per-situation activation counts on test, contribution of calm-process vs assertive-fusion swaps, and whether trigger errors systematically avoid harmful profile applications). Without that, the result is consistent with a sparse validation-tuned activation pattern rather than situation-matched noise models.
  2. [Table II, §V-E] Table II and §V-E: Turning is mapped to assertive-fusion but gated at τ=0.99, so assertive-fusion on turning is effectively near-certain only (242 near-certain turning rows across cohorts). Straight-line calm-process is always on when predicted (τ=0.00). Please quantify, on the held-out split, the fraction of timesteps and of the 0.017 m gain attributable to each active profile versus residual global operation. If most of the gain is from one profile under a near-always or near-never gate, the multi-situation scheduling narrative should be narrowed accordingly.
  3. [§VI-B, §VII-E, Abstract] §VI–VII and Abstract: The primary improvement is 0.017 m on a ~0.49 m label-weighted T-RMSE with sparse AprilTag labels and an internally estimated tag map. Please bound how this delta compares to plausible label/map uncertainty (e.g., tag pose noise, extrinsics, staleness matching within 0.2 s) so readers can judge whether the gain is larger than residual supervision error. Relative-pose metrics help but do not fully replace that comparison for the absolute headline number.
minor comments (6)
  1. [§V-A] §V-A: State the full state vector, process model f, and measurement model h (or point to a precise supplement section) so the continuous error-state backbone is reproducible without reverse-engineering from narrative description.
  2. [Fig. 4, Fig. 8] Fig. 4 and Fig. 8: Axis units, time base, and which run is shown should be stated in the captions; currently the illustrative runs are hard to map to the train/val/test letters in Table I.
  3. [§IV-A] §IV-A: The kinematic thresholds and priority order used for GT situation labeling are described only qualitatively; a short table of speed/yaw-rate/vertical cutoffs would make the offline labels reproducible.
  4. [Table VII] Table VII: Clarify whether relative-pose pairs are pooled across runs or label-weighted per-run as in the primary metric; the aggregation convention should match the headline protocol or be explicitly different.
  5. [References] References [6] and [26] carry 2026 volume years while the arXiv stamp is also 2026; verify bibliographic completeness and access dates for any not-yet-final items.
  6. [§VIII-E] §VIII-E: The NIS and lag-1 autocorrelation diagnostics are useful; consider moving a one-line summary of mean NIS ratio and max lag-1 into the main results so the white-noise DVL limitation is visible without relying only on discussion.

Circularity Check

0 steps flagged

No circularity: train/val calibration is frozen before held-out test; headline gain is an empirical comparison, not a quantity forced by construction.

full rationale

This is an empirical robotics paper, not a first-principles derivation. Noise profiles, the one-time DVL yaw correction (~1.8° from 64 train velocity pairs), the onboard trigger, and per-situation gates are fitted on train/validation and then frozen before any held-out scoring (§V, Table II). The primary claim is a label-weighted held-out T-RMSE comparison of the same continuous CAR-EKF backbone with vs. without the frozen scheduling policy (0.488 m → 0.471 m; Table V), with paired bootstrap on 10 s segments. That is standard train/val/test methodology: the test metric is not an algebraic rewrite of the fitted parameters. Oracle GT routing is reported as an upper-bound experiment and fails to beat global on test (0.494 m vs 0.488 m; Table III), which is the opposite of label leakage or self-definitional forcing. No uniqueness theorem, load-bearing self-citation chain, or renamed known identity underpins the result. Residual concerns about causal interpretation of the gates (skeptic attack) are correctness/attribution issues, not circularity of the derivation chain.

Axiom & Free-Parameter Ledger

6 free parameters · 6 axioms · 2 invented entities

The central claim rests on standard EKF estimation machinery plus domain assumptions about discrete motion regimes and white-ish sensor noise, plus a large set of free parameters (noise profiles, gates, DVL yaw, classifier) fitted under sparse AprilTag supervision and frozen before test. No new physical entities are postulated; the invented pieces are engineering constructs (situation profiles, ST-CAR-EKF routing). The ledger is dominated by fitted covariances and thresholds rather than novel axioms.

free parameters (6)
  • DVL yaw misalignment correction = ~1.8 degrees
    One-time closed-form yaw-only fit (~1.8°) from 64 sparse velocity pairs on eight training runs; frozen for all subsequent evaluation (§V-A).
  • Global and situational Q/R profile scalars = profile-specific scalars (examples in §V-B)
    Process- and measurement-noise entries for global, calm-process, and assertive-fusion profiles (e.g., velocity random walk 0.24→0.18 m/s²; DVL planar noise 0.045→0.030 m/s) selected by validation translation RMSE (§V-A,B).
  • Per-situation activation thresholds τ_s = 0.00 / 0.35 / 0.99
    Validation grid/coordinate descent yields τ=0.00 (straight), 0.35 (transition), 0.99 (turning); frozen before test (§V-E, Table II).
  • Onboard situation classifier (31 features) = 31 selected onboard features; weights not tabulated
    Probabilistic trigger supervised by GT situation labels offline; feature set and model weights are free parameters of the runtime gate (§III-B, §IV-B).
  • Sensor staleness bound = 0.75 s
    Samples older than 0.75 s rejected in causal as-of merge at 10 Hz (§III-B).
  • Situation labeling kinematics thresholds/priority = fixed but not fully numeric in text
    GT situations assigned from smoothed speed, yaw rate, vertical motion with fixed priority order; defines supervision for the trigger (§IV-A).
axioms (6)
  • domain assumption Error-state EKF with IMU/control propagation and DVL plus altitude/range updates is an adequate causal dead-reckoning backbone for this platform.
    Assumed throughout §V; search includes Huber and innovation-adaptive options but validation selects fixed covariances on pool data.
  • domain assumption DVL velocity errors may be treated as white in the filter, with situation-dependent R capturing residual mismatch.
    Standard EKF assumption; §VIII-E reports lag-1 residual autocorrelation up to ~0.92 and mean |lag-1| ~0.59, acknowledged as unsolved.
  • ad hoc to paper A small discrete situation set (straight, turning, transition/mixed, hover/low speed) is sufficient to retarget fixed noise models.
    Defined in §IV-A from GT kinematics; not derived from a uniqueness theorem.
  • domain assumption Sparse unsurveyed pool-floor AprilTag poses provide usable offline six-DoF labels after co-visibility map recovery.
    §III-C; relative-pose metrics used to reduce but not eliminate map-alignment dependence (§VI-B, §VIII-G).
  • ad hoc to paper Cross-run transfer within the same pool benchmark is the right evaluation of calibration generalization for this claim.
    Framed explicitly as small transfer learning in §V-D; open-water transfer is out of scope.
  • standard math Standard Kalman covariance update equations remain valid when Q_k and R_k are piecewise-constant switched signals.
    Eqs. (1)–(2) in §V-C; state and P continuous across switches.
invented entities (2)
  • ST-CAR-EKF scheduling policy no independent evidence
    purpose: Name the confidence-gated mapping from onboard situation probabilities to fixed Q/R profiles on a continuous CAR-EKF backbone.
    Engineering construct, not a physical entity; independent evidence is the held-out RMSE comparison itself, not an external measurement channel.
  • Calm-process and assertive-fusion fixed-matrix profiles no independent evidence
    purpose: Two discrete noise retargetings: lower process noise for clean straight motion; tighter DVL R for turns/transitions.
    Defined by calibration on situation slices (§V-B, Table IV); no existence claim beyond filter tuning knobs.

pith-pipeline@v1.1.0-grok45 · 21666 in / 4250 out tokens · 46910 ms · 2026-07-14T10:34:25.736786+00:00 · methodology

0 comments
read the original abstract

Underwater dead reckoning estimates vehicle position when vision is unavailable and external positioning cannot be assumed. A single set of filter parameters can work well in many situations, but fixed tuning may be poorly matched during turns, motion transitions, or periods when sensor measurements are less reliable. This paper presents the Situation-Triggered Calibrated Adaptive Robust Extended Kalman Filter for a BlueROV2. An onboard probabilistic trigger identifies the current motion situation while one error-state filter runs continuously. When the trigger is confident, the filter changes only to the corresponding pre-calibrated process- and measurement-noise matrices; the state estimate, covariance history, dynamics, and measurement models are not reset or replaced. The trigger, noise profiles, and a one-time Doppler velocity log yaw-alignment correction are calibrated offline using sparse AprilTag-supervised pool runs. A separate validation set selects the scheduling policy, which is then fixed before held-out testing. Across four held-out pool runs, the method reduces label-weighted mean per-run translation root-mean-square error from 0.488 m to 0.471 m relative to the same filter backbone with one global noise profile, and every held-out run favors the scheduled method. A paired bootstrap over 10-second segments gives a candidate-minus-baseline difference of -0.017 m with a 95% confidence interval of [-0.024, -0.008] m, while orientation error remains essentially unchanged. These results indicate that situation-aware covariance scheduling provides a modest but consistent vision-free dead-reckoning improvement without switching estimators or resetting the filter.

Figures

Figures reproduced from arXiv: 2607.10597 by (2) University of Illinois Urbana-Champaign, (3) U.S. Army Engineer Research, Ahmet Soylemezoglu (3) ((1) University of Illinois Urbana-Champaign, Akshay Naik (1), Computer Engineering, Construction Engineering Research Laboratory), Department of Electrical, Department of Industrial, Development Center, Dustin Nottage (3), Enterprise Systems Engineering, Ramavarapu S. Sreenivas (2).

Figure 1
Figure 1. Figure 1: Data processing and model training pipeline (offline only). [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Trusted AprilTag map used to generate sparse camera and vehicle pose labels. Tag 0 defines the origin and anchors the reference frame; each square [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Curated combined dataset. (a) AprilTag-derived pose labels are a sparse fraction of each camera stream; run letters and bar color encode the [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Situation triggering on a clean illustrative run. (a) The path is segmented into motion situations from GT kinematics. (b) GT labels (top) supervise an [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Situation-gated Q/R profile update. Onboard features produce situation probabilities; a validation-selected confidence gate either selects a fixed situation profile or keeps the global CAR-EKF profile. The selected Qk and Rk enter the ordinary EKF covariance terms, while the state, covariance history, transition model, and measurement model remain continuous. D. Cross-Run Calibration Transfer Noise calibra… view at source ↗
Figure 6
Figure 6. Figure 6: Deployable ST-CAR-EKF runtime layout. Live onboard streams (IMU, DVL, relative altitude, control) feed both an onboard situation trigger classifier [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Global baseline selection on the test run where the commercial stream is logged. (a) Estimated versus ground-truth trajectory and (b) per-frame [PITH_FULL_IMAGE:figures/full_fig_p010_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Confidence-gated switching. (a) The onboard transition/mixed trigger probability over time on an example run; a swap fires when it crosses the [PITH_FULL_IMAGE:figures/full_fig_p011_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Equal-footing comparison on test run N, where the commercial stream is logged. (a) ST-CAR-EKF (green) tracks the ground-truth path while Commercial [PITH_FULL_IMAGE:figures/full_fig_p012_9.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

41 extracted references · 14 canonical work pages

  1. [1]

    A survey of underwater vehicle navigation: Recent advances and new challenges,

    J. C. Kinsey, R. M. Eustice, and L. L. Whitcomb, “A survey of underwater vehicle navigation: Recent advances and new challenges,” inProceedings of the 7th IFAC Conference on Manoeuvring and Control of Marine Craft, 2006, pp. 1–12. [Online]. Available: https://robots.engin.umich.edu/publications/jkinsey-2006a.pdf

    2006

  2. [2]

    AUV navigation and localization: A review,

    L. Paull, S. Saeedi, M. Seto, and H. Li, “AUV navigation and localization: A review,”IEEE Journal of Oceanic Engineering, vol. 39, no. 1, pp. 131–149, 2014. [Online]. Available: https://doi.org/10.1109/JOE.2013.2278891

    doi:10.1109/joe.2013.2278891 2014

  3. [3]

    The interacting multiple model algorithm for systems with markovian switching coefficients,

    H. A. P. Blom and Y . Bar-Shalom, “The interacting multiple model algorithm for systems with markovian switching coefficients,”IEEE Transactions on Automatic Control, vol. 33, no. 8, pp. 780–783, 1988. [Online]. Available: https://doi.org/10.1109/9.1299

    doi:10.1109/9.1299 1988

  4. [4]

    A multi-model EKF integrated navigation algorithm for deep water AUV,

    D. Li, D. Ji, J. Liu, and Y . Lin, “A multi-model EKF integrated navigation algorithm for deep water AUV,”International Journal of Advanced Robotic Systems, 2016. [Online]. Available: https://doi.org/10.5772/62076

    work page doi:10.5772/62076 2016

  5. [5]

    Multiple model AUV navigation methodology with adaptivity and robustness,

    X. Zhang, B. He, S. Gao, P. Mu, J. Xu, and N. Zhai, “Multiple model AUV navigation methodology with adaptivity and robustness,”Ocean Engineering, vol. 254, p. 111258, 2022. [Online]. Available: https://doi.org/10.1016/j.oceaneng.2022.111258

    doi:10.1016/j.oceaneng.2022.111258 2022

  6. [6]

    A hybrid-kernel-based adaptive robust kalman filter for INS/DVL integrated underwater navigation,

    L. Kang, K. He, J. Zhao, X. Wang, and P. Tan, “A hybrid-kernel-based adaptive robust kalman filter for INS/DVL integrated underwater navigation,” Ocean Engineering, vol. 350, p. 124269, 2026. [Online]. Available: https://doi.org/10.1016/j.oceaneng.2026.124269

    work page doi:10.1016/j.oceaneng.2026.124269 2026

  7. [7]

    A survey on transfer learning,

    S. J. Pan and Q. Yang, “A survey on transfer learning,”IEEE Transactions on Knowledge and Data Engineering, vol. 22, no. 10, pp. 1345–1359,

  8. [8]

    Available: https://doi.org/10.1109/TKDE.2009.191

    [Online]. Available: https://doi.org/10.1109/TKDE.2009.191

    doi:10.1109/tkde.2009.191 2009

  9. [9]

    AprilTag: A robust and flexible visual fiducial system,

    E. Olson, “AprilTag: A robust and flexible visual fiducial system,” inProceedings of the IEEE International Conference on Robotics and Automation, 2011, pp. 3400–3407. [Online]. Available: https://doi.org/10.1109/ICRA.2011.5979561

    doi:10.1109/icra.2011.5979561 2011

  10. [10]

    Flexible layouts for fiducial tags,

    M. Krogius, A. Haggenmiller, and E. Olson, “Flexible layouts for fiducial tags,” inProceedings of the IEEE/RSJ International Conference on Intelligent Robots and Systems, 2019, pp. 1898–1903. [Online]. Available: https://doi.org/10.1109/IROS40897.2019.8967787

    doi:10.1109/iros40897.2019.8967787 2019

  11. [11]

    Evaluation of underwater AprilTag localization for highly agile micro underwater robots,

    N. Bauschmann, D. A. Duecker, T. L. Alff, and R. Seifried, “Evaluation of underwater AprilTag localization for highly agile micro underwater robots,” inProceedings of the IEEE/RSJ International Conference on Intelligent Robots and Systems, 2023, pp. 9926–9932. [Online]. Available: https://doi.org/10.1109/IROS55552.2023.10341764

    doi:10.1109/iros55552.2023.10341764 2023

  12. [12]

    Evaluation of UKF-based fusion strategies for autonomous underwater vehicles multisensor navigation,

    A. Bucci, M. Franchi, A. Ridolfi, N. Secciani, and B. Allotta, “Evaluation of UKF-based fusion strategies for autonomous underwater vehicles multisensor navigation,”IEEE Journal of Oceanic Engineering, vol. 48, no. 1, pp. 1–26, 2023. [Online]. Available: https://doi.org/10.1109/JOE.2022.3168934

    work page doi:10.1109/joe.2022.3168934 2023

  13. [13]

    Terrain-aided navigation with coarse maps—toward an arctic crossing with an AUV,

    G. Salavasidis, A. Munafo, S. McPhail, C. A. Harris, D. Fenucci, M. Pebody, E. Rogers, and A. B. Phillips, “Terrain-aided navigation with coarse maps—toward an arctic crossing with an AUV,”IEEE Journal of Oceanic Engineering, vol. 46, no. 4, pp. 1192–1212, 2021. [Online]. Available: https://doi.org/10.1109/JOE.2021.3085941

    work page doi:10.1109/joe.2021.3085941 2021

  14. [14]

    A quantitative evaluation of bathymetry-based bayesian localization methods for autonomous underwater robots,

    J. Hong, M. Fulton, K. Orpen, K. Barthelemy, K. Berlin, and J. Sattar, “A quantitative evaluation of bathymetry-based bayesian localization methods for autonomous underwater robots,”IEEE Journal of Oceanic Engineering, vol. 50, no. 2, pp. 985–1000, 2025. [Online]. Available: https://doi.org/10.1109/JOE.2025.3535598

    doi:10.1109/joe.2025.3535598 2025

  15. [15]

    One-way-travel-time hybrid baseline navigation for micro autonomous underwater vehicles,

    Y . Wang, R. Hu, P. Du, W. Yang, Y . Chen, and S. H. Huang, “One-way-travel-time hybrid baseline navigation for micro autonomous underwater vehicles,”IEEE Journal of Oceanic Engineering, vol. 50, no. 2, pp. 968–984, 2025. [Online]. Available: https://doi.org/10.1109/JOE.2024.3447739

    work page doi:10.1109/joe.2024.3447739 2025

  16. [16]

    Seamless underwater navigation with limited doppler velocity log measurements,

    N. Cohen and I. Klein, “Seamless underwater navigation with limited doppler velocity log measurements,” arXiv:2404.13742, 2024. [Online]. Available: https://arxiv.org/abs/2404.13742

    Pith/arXiv arXiv 2024

  17. [17]

    Robust estimation of a location parameter,

    P. J. Huber, “Robust estimation of a location parameter,”The Annals of Mathematical Statistics, vol. 35, no. 1, pp. 73–101, 1964. [Online]. Available: https://doi.org/10.1214/aoms/1177703732

    doi:10.1214/aoms/1177703732 1964

  18. [18]

    Variational bayesian adaptation of noise covariances in non-linear kalman filtering,

    S. S ¨arkk¨a and J. Hartikainen, “Variational bayesian adaptation of noise covariances in non-linear kalman filtering,” arXiv:1302.0681, 2013. [Online]. Available: https://arxiv.org/abs/1302.0681

    Pith/arXiv arXiv 2013

  19. [19]

    Adaptive adjustment of noise covariance in kalman filter for dynamic state estimation,

    S. Akhlaghi, N. Zhou, and Z. Huang, “Adaptive adjustment of noise covariance in kalman filter for dynamic state estimation,” arXiv:1702.00884, 2017. [Online]. Available: https://arxiv.org/abs/1702.00884

    Pith/arXiv arXiv 2017

  20. [20]

    ProNet: Adaptive process noise estimation for INS/DVL fusion,

    B. Or and I. Klein, “ProNet: Adaptive process noise estimation for INS/DVL fusion,” arXiv:2212.08882, 2022. [Online]. Available: https://arxiv.org/abs/2212.08882

    Pith/arXiv arXiv 2022

  21. [21]

    iSAM2: Incremental smoothing and mapping using the bayes tree,

    M. Kaess, H. Johannsson, R. Roberts, V . Ila, J. J. Leonard, and F. Dellaert, “iSAM2: Incremental smoothing and mapping using the bayes tree,”The International Journal of Robotics Research, vol. 31, no. 2, pp. 216–235, 2012. [Online]. Available: https://doi.org/10.1177/0278364911430419

    doi:10.1177/0278364911430419 2012

  22. [22]

    IMU preintegration on manifold for efficient visual-inertial maximum-a-posteriori estimation,

    C. Forster, L. Carlone, F. Dellaert, and D. Scaramuzza, “IMU preintegration on manifold for efficient visual-inertial maximum-a-posteriori estimation,” inProceedings of Robotics: Science and Systems, 2015. [Online]. Available: https://doi.org/10.15607/RSS.2015.XI.006

    work page doi:10.15607/rss.2015.xi.006 2015

  23. [23]

    A robust INS/USBL/DVL integrated navigation algorithm using graph optimization,

    P. Li, Y . Liu, T. Yan, S. Yang, and R. Li, “A robust INS/USBL/DVL integrated navigation algorithm using graph optimization,”Sensors, vol. 23, no. 2, p. 916, 2023. [Online]. Available: https://doi.org/10.3390/s23020916

    work page doi:10.3390/s23020916 2023

  24. [24]

    A novel INS/USBL/DVL integrated navigation scheme against complex underwater environment,

    H. Qin, X. Wang, G. Wang, M. Hu, Y . Bian, X. Qin, and R. Ding, “A novel INS/USBL/DVL integrated navigation scheme against complex underwater environment,”Ocean Engineering, vol. 286, p. 115485, 2023. [Online]. Available: https://doi.org/10.1016/j.oceaneng.2023.115485

    work page doi:10.1016/j.oceaneng.2023.115485 2023

  25. [25]

    FGO-ILNS: Tightly coupled multi-sensor integrated navigation system based on factor graph optimization for autonomous underwater vehicle,

    J. Song, W. Li, R. Liu, and X. Zhu, “FGO-ILNS: Tightly coupled multi-sensor integrated navigation system based on factor graph optimization for autonomous underwater vehicle,” arXiv:2310.14163, 2023. [Online]. Available: https://arxiv.org/abs/2310.14163

    Pith/arXiv arXiv 2023

  26. [26]

    Combining DVL-INS and laser-based loop closures in a batch estimation framework for underwater positioning,

    A. Al-Baali, T. Hitchcox, and J. R. Forbes, “Combining DVL-INS and laser-based loop closures in a batch estimation framework for underwater positioning,”IEEE Journal of Oceanic Engineering, vol. 48, no. 4, pp. 1096–1111, 2023. [Online]. Available: https://doi.org/10.1109/JOE.2023.3286854

    work page doi:10.1109/joe.2023.3286854 2023

  27. [27]

    A robust INS/USBL/DVL integrated navigation method based on adaptive correlation entropy factor graph optimization,

    S. Cheng, Y . Wang, Q. Zhao, H. Zhu, and X. Qu, “A robust INS/USBL/DVL integrated navigation method based on adaptive correlation entropy factor graph optimization,”Ocean Engineering, vol. 356, no. Part 1, p. 125234, 2026. [Online]. Available: https://doi.org/10.1016/j.oceaneng.2026.125234

    work page doi:10.1016/j.oceaneng.2026.125234 2026

  28. [28]

    RoNIN: Robust neural inertial navigation in the wild: Benchmark, evaluations, and new methods,

    S. Herath, H. Yan, and Y . Furukawa, “RoNIN: Robust neural inertial navigation in the wild: Benchmark, evaluations, and new methods,” inProceedings of the IEEE International Conference on Robotics and Automation, 2020, pp. 3146–3152. [Online]. Available: https://doi.org/10.1109/ICRA40945.2020.9196860

    doi:10.1109/icra40945.2020.9196860 2020

  29. [29]

    TLIO: Tight learned inertial odometry,

    W. Liu, D. Caruso, E. Ilg, J. Dong, A. I. Mourikis, K. Daniilidis, V . Kumar, and J. Engel, “TLIO: Tight learned inertial odometry,”IEEE Robotics and Automation Letters, vol. 5, no. 4, pp. 5653–5660, 2020. [Online]. Available: https://doi.org/10.1109/LRA.2020.3007421

    work page doi:10.1109/lra.2020.3007421 2020

  30. [31]

    Available: https://arxiv.org/abs/2310.04874 16

    [Online]. Available: https://arxiv.org/abs/2310.04874 16

    Pith/arXiv arXiv

  31. [32]

    BeamsNet: A data-driven approach enhancing doppler velocity log measurements for autonomous underwater vehicle navigation,

    N. Cohen and I. Klein, “BeamsNet: A data-driven approach enhancing doppler velocity log measurements for autonomous underwater vehicle navigation,” Engineering Applications of Artificial Intelligence, vol. 114, p. 105216, 2022. [Online]. Available: https://doi.org/10.1016/j.engappai.2022.105216

    work page doi:10.1016/j.engappai.2022.105216 2022

  32. [33]

    MissBeamNet: Learning missing doppler velocity log beam measurements,

    M. Yona and I. Klein, “MissBeamNet: Learning missing doppler velocity log beam measurements,”Neural Computing and Applications, vol. 36, no. 9, pp. 4947–4958, 2024. [Online]. Available: https://doi.org/10.1007/s00521-023-09303-4

    work page doi:10.1007/s00521-023-09303-4 2024

  33. [34]

    Enhancing underwater navigation through cross-correlation-aware deep INS/DVL fusion,

    N. Cohen and I. Klein, “Enhancing underwater navigation through cross-correlation-aware deep INS/DVL fusion,” arXiv:2503.21727, 2025. [Online]. Available: https://arxiv.org/abs/2503.21727

    arXiv 2025

  34. [35]

    TagSLAM: Robust SLAM with fiducial markers,

    B. Pfrommer and K. Daniilidis, “TagSLAM: Robust SLAM with fiducial markers,” arXiv:1910.00679, 2019. [Online]. Available: https: //arxiv.org/abs/1910.00679

    Pith/arXiv arXiv 1910

  35. [36]

    Tank dataset: An underwater multi-sensor dataset for SLAM evaluation,

    S. Xu, J. S. Willners, J. Roe, S. Katagiri, T. Luczynski, Y . Petillot, and S. Wang, “Tank dataset: An underwater multi-sensor dataset for SLAM evaluation,”The International Journal of Robotics Research, vol. 45, no. 4, 2026. [Online]. Available: https://doi.org/10.1177/02783649251364904

    work page doi:10.1177/02783649251364904 2026

  36. [37]

    BlueROV2 Technical Specifications,

    Blue Robotics, “BlueROV2 Technical Specifications,” Blue Robotics, Tech. Rep., 2025, product specifications, accessed 2026-06-24. [Online]. Available: https://bluerobotics.com/store/rov/bluerov2/

    2025

  37. [38]

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

    A. Geiger, P. Lenz, and R. Urtasun, “Are we ready for autonomous driving? the KITTI vision benchmark suite,” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2012, pp. 3354–3361. [Online]. Available: https://doi.org/10.1109/CVPR.2012.6248074

    doi:10.1109/cvpr.2012.6248074 2012

  38. [39]

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

    J. Sturm, N. Engelhard, F. Endres, W. Burgard, and D. Cremers, “A benchmark for the evaluation of RGB-D SLAM systems,” in Proceedings of the IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2012, pp. 573–580. [Online]. Available: https://doi.org/10.1109/IROS.2012.6385773

    doi:10.1109/iros.2012.6385773 2012

  39. [40]

    Dead reckoning,

    Water Linked, “Dead reckoning,” Product documentation, 2026, accessed 2026-06-03. [Online]. Available: https://docs.waterlinked.com/dvl/ dead-reckoning/

    2026

  40. [41]

    Bootstrap methods: Another look at the jackknife,

    B. Efron, “Bootstrap methods: Another look at the jackknife,”The Annals of Statistics, vol. 7, no. 1, pp. 1–26, 1979. [Online]. Available: https://doi.org/10.1214/aos/1176344552

    doi:10.1214/aos/1176344552 1979

  41. [42]

    Robust model-aided inertial localization for autonomous underwater vehicles,

    S. Arnold and L. Medagoda, “Robust model-aided inertial localization for autonomous underwater vehicles,” arXiv:1805.08011, 2018. [Online]. Available: https://arxiv.org/abs/1805.08011

    Pith/arXiv arXiv 2018