arxiv: 2604.02878 · v1 · submitted 2026-04-03 · 💻 cs.RO · cs.SY· eess.SY

Recognition: no theorem link

An Asynchronous Two-Speed Kalman Filter for Real-Time UUV Cooperative Navigation Under Acoustic Delays

Shuyue Li , Miguel L\'opez-Ben\'itez , Eng Gee Lim , Fei Ma , Qian Dong , Mengze Cao , Limin Yu , Xiaohui Qin

Authors on Pith no claims yet

Pith reviewed 2026-05-13 19:54 UTC · model grok-4.3

classification 💻 cs.RO cs.SYeess.SY

keywords UUV cooperative navigationasynchronous Kalman filteracoustic communication delaysout-of-sequence measurementstwo-speed filteringdead reckoningvariational history distillation

0 comments

The pith

A two-speed Kalman filter processes delayed acoustic measurements in UUV navigation without pausing real-time control.

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

The paper introduces an asynchronous two-speed Kalman filter that splits estimation into a fast thread for immediate dead-reckoning control and a slow thread for handling delayed collaborative data from other vehicles. It stores recent states in a buffer, applies out-of-sequence measurements to the matching past state, and uses a projection step to move the correction forward to the present moment. This structure keeps computation under a millisecond per step even with delays as long as 30 seconds while producing trajectory errors close to those of full batch optimization. A reader would care because standard filters either wait for the delayed data and break the control loop or ignore the data and let position drift grow without bound in GNSS-denied water.

Core claim

The central claim is that decoupling the filter into parallel fast-rate Gaussian-process dead-reckoning and slow-rate delayed-correction threads, linked by a finite state buffer and variational history distillation projection, allows the system to incorporate acoustic measurements arriving up to 30 seconds late without recomputing the entire history or blocking high-frequency control.

What carries the argument

The Variational History Distillation projection mechanism, which takes a measurement correction computed at a past time t-T and advances it to the current time by distilling the intervening state history stored in the buffer.

If this is right

Trajectory RMSE stays comparable to batch optimization even when measurements arrive after 30-second delays.
The filter runs in sub-millisecond time, preserving the high-frequency control loop.
Standard EKF and UKF produce noticeably larger drift when the same delayed measurements are either discarded or cause blocking waits.
The design supports a control-communication-computing co-design that improves resilience of multi-vehicle marine systems.

Where Pith is reading between the lines

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

The same buffer-plus-projection pattern could be tested on other delayed-sensor problems such as multi-drone teams using low-bandwidth radio links.
Replacing the Gaussian-process dead-reckoning model with a more accurate vehicle-specific dynamics model would likely shrink the gap to batch accuracy further.
Field trials with physical vehicles and measured acoustic channel statistics would show whether the simulation delay model matches real propagation and multipath effects.

Load-bearing premise

The projection step that moves a past correction to the present time adds only negligible extra error beyond the error already present in the Gaussian-process dead-reckoning model.

What would settle it

Apply the filter to logged UUV trajectories that include real acoustic round-trip times of 20-30 seconds and compare the resulting position RMSE against an offline batch optimizer run on the same data; a large gap between the two would falsify the claim.

Figures

Figures reproduced from arXiv: 2604.02878 by Eng Gee Lim, Fei Ma, Limin Yu, Mengze Cao, Miguel L\'opez-Ben\'itez, Qian Dong, Shuyue Li, Xiaohui Qin.

**Figure 2.** Figure 2: During long-term acoustic interruption, the predic [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

**Figure 3.** Figure 3: The timing diagram of the OOSM processing mechanism. [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗

**Figure 4.** Figure 4: Dynamic acoustic propagation delay curve simulated [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗

**Figure 5.** Figure 5: Simulated UUV trajectory in the X − Y plane. The delay-ignorant standard EKF diverges significantly (experiencing severe “pull-back” oscillations) when fusing measurements delayed by >10 s. Conversely, the proposed TSKF maintains robust adherence to the ground truth trajectory. Because the UUV has physically moved tens of meters during the severe acoustic delay T , fusing a measurement originating from t … view at source ↗

read the original abstract

In GNSS-denied underwater environments, individual unmanned underwater vehicles (UUVs) suffer from unbounded dead-reckoning drift, making collaborative navigation crucial for accurate state estimation. However, the severe communication delay inherent in underwater acoustic channels poses serious challenges to real-time state estimation. Traditional filters, such as Extended Kalman Filters (EKF) or Unscented Kalman Filters (UKF), usually block the main control loop while waiting for delayed data, or completely discard Out-of-Sequence Measurements (OOSM), resulting in serious drift. To address this, we propose an Asynchronous Two-Speed Kalman Filter (TSKF) enhanced by a novel projection mechanism, which we term Variational History Distillation (VHD). The proposed architecture decouples the estimation process into two parallel threads: a fast-rate thread that utilizes Gaussian Process (GP) compensated dead reckoning to guarantee high-frequency real-time control, and a slow-rate thread dedicated to processing asynchronously delayed collaborative information. By introducing a finite-length State Buffer, the algorithm applies delayed measurements (t-T) to their corresponding historical states, and utilizes a VHD-based projection to fast-forward the correction to the current time without computationally heavy recalculations. Simulation results demonstrate that the proposed TSKF maintains trajectory Root Mean Square Error (RMSE) comparable to computationally intensive batch-optimization methods under severe delays (up to 30 s). Executing in sub-millisecond time, it significantly outperforms standard EKF/UKF. The results demonstrate an effective control, communication, and computing (3C) co-design that significantly enhances the resilience of autonomous marine automation systems.

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note private letter to a colleague

The TSKF gives a workable two-speed split for handling long acoustic delays in UUV teams, but the VHD projection lacks any error bounds or drift analysis for 30 s gaps.

read the letter

The paper's main contribution is a two-speed Kalman filter that runs a fast Gaussian-process dead-reckoning thread for control while a slow thread applies delayed collaborative measurements through a finite state buffer and a variational projection step. This keeps the main loop from stalling on acoustic delays up to 30 s and produces RMSE numbers in simulation that sit close to batch optimization while staying under a millisecond per update. The architecture is a direct engineering response to a constraint that standard EKF and UKF either block on or ignore, and the decoupling plus buffer idea is presented clearly enough to be implemented from the description. That part is useful for anyone who has to keep a vehicle moving while waiting on teammates' data. The evidence is limited to simulation trajectories with no error bars, no detailed noise models, and only basic EKF/UKF baselines. More critically, there is no derivation or bound on how much linearization or approximation error the VHD projection adds when it fast-forwards a correction across 30 s, nor any check on how GP kernel mismatch compounds in the fast thread. Those omissions leave the parity claim tied to the specific test cases rather than shown as general. The work is aimed at marine robotics engineers who need something that runs in real time rather than offline smoothing. A reader who already knows asynchronous filtering literature will see the practical packaging but will still want tighter analysis on the projection step. I would send it to peer review so the authors can add the missing bounds and more varied validation.

Referee Report

3 major / 2 minor

Summary. The manuscript presents an Asynchronous Two-Speed Kalman Filter (TSKF) for real-time cooperative navigation of unmanned underwater vehicles (UUVs) in the presence of significant acoustic communication delays. The approach decouples a fast-rate Gaussian Process (GP) compensated dead-reckoning thread from a slow-rate thread that processes delayed measurements using a finite State Buffer and a Variational History Distillation (VHD) projection mechanism to apply corrections to the current state without full recalculations. Simulation results are reported to show trajectory RMSE comparable to batch optimization methods for delays up to 30 seconds, with sub-millisecond execution time outperforming standard EKF and UKF.

Significance. If the VHD projection and GP model can be shown to introduce bounded error, the two-speed architecture would offer a practical advance for maintaining accurate real-time state estimation in GNSS-denied underwater settings under severe acoustic delays, directly supporting 3C co-design goals in marine autonomy.

major comments (3)

[Abstract] Abstract: the headline claim of RMSE parity with batch optimization under 30 s delays is load-bearing for the contribution, yet the reported simulations provide no error bars, no description of the simulation environment or trajectory generation, and no quantification of projection-induced or GP-approximation error.
[VHD Mechanism] VHD projection description: no derivation or bound is given on the linearization/variational approximation error when mapping a delayed correction at t-T onto the current state; without this, the assertion that the projection avoids substantial additional error relative to batch methods cannot be evaluated.
[Simulation Results] Simulation results section: the comparison is restricted to EKF/UKF; absence of additional baselines, sensitivity to GP kernel mismatch, or analysis of drift accumulation over 30 s intervals leaves open whether the reported performance is an artifact of the chosen noise levels and trajectories.

minor comments (2)

[Method] Clarify the exact length of the finite State Buffer and its relation to the maximum anticipated delay T.
[Figures] Add uncertainty visualization (e.g., covariance ellipses) to the trajectory plots so that the RMSE values can be interpreted in context.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive and detailed comments on our manuscript. We address each major comment point by point below and will incorporate the suggested improvements in the revised version to strengthen the presentation of our results.

read point-by-point responses

Referee: [Abstract] Abstract: the headline claim of RMSE parity with batch optimization under 30 s delays is load-bearing for the contribution, yet the reported simulations provide no error bars, no description of the simulation environment or trajectory generation, and no quantification of projection-induced or GP-approximation error.

Authors: We agree that the abstract claim requires more rigorous supporting details from the simulations. In the revised manuscript we will add error bars to all RMSE plots, include a full description of the simulation environment and trajectory generation procedure, and provide explicit quantification of projection-induced and GP-approximation errors via additional analysis and Monte-Carlo runs. revision: yes
Referee: [VHD Mechanism] VHD projection description: no derivation or bound is given on the linearization/variational approximation error when mapping a delayed correction at t-T onto the current state; without this, the assertion that the projection avoids substantial additional error relative to batch methods cannot be evaluated.

Authors: The VHD mechanism relies on a variational approximation to project corrections forward in time. While the current text describes the projection steps, we acknowledge the absence of an explicit error bound. In the revision we will add a derivation of the linearization and variational approximation together with a theoretical bound on the introduced error, supported by numerical validation showing that the additional error remains small relative to batch optimization. revision: yes
Referee: [Simulation Results] Simulation results section: the comparison is restricted to EKF/UKF; absence of additional baselines, sensitivity to GP kernel mismatch, or analysis of drift accumulation over 30 s intervals leaves open whether the reported performance is an artifact of the chosen noise levels and trajectories.

Authors: We will expand the simulation section to include direct comparisons against batch optimization methods (as referenced in the abstract), perform sensitivity analysis across GP kernel hyperparameters, and add plots and statistics quantifying drift accumulation over 30-second intervals under varied noise and trajectory conditions. These additions will demonstrate that the observed performance is not an artifact of the specific simulation settings. revision: yes

Circularity Check

0 steps flagged

No significant circularity; new architecture with independent simulation validation

full rationale

The manuscript introduces an Asynchronous Two-Speed Kalman Filter (TSKF) with a novel Variational History Distillation (VHD) projection and finite State Buffer to handle delayed acoustic measurements. The derivation chain consists of a standard decoupling into fast GP-compensated dead-reckoning and slow delayed-update threads, followed by a projection step that maps historical corrections forward. No equations reduce a claimed prediction to a fitted parameter by construction, nor does any load-bearing step rely on self-citation of an unverified uniqueness theorem or ansatz. Simulation RMSE comparisons to batch optimization are presented as empirical evidence rather than tautological outputs. The central claim therefore remains externally falsifiable and does not collapse to its own inputs.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 1 invented entities

Review performed on abstract only; no explicit free parameters, axioms, or invented entities are stated. The VHD mechanism is introduced as a novel component without independent evidence provided.

invented entities (1)

Variational History Distillation (VHD) no independent evidence
purpose: Project delayed measurement corrections from historical states to current time
New projection mechanism introduced to avoid recomputation; no independent evidence or falsifiable prediction given in abstract.

pith-pipeline@v0.9.0 · 5625 in / 1160 out tokens · 37967 ms · 2026-05-13T19:54:11.462539+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

16 extracted references · 16 canonical work pages · 1 internal anchor

[1]

AUV navigation an d localiza- tion: A review,

L. Paull, S. Saeedi, M. Seto, and H. Li, “AUV navigation an d localiza- tion: A review,” IEEE Journal of Oceanic Engineering , vol. 39, no. 1, pp. 131–149, 2014

work page 2014
[2]

Enabling cooperative autonomy in UUV clusters: A survey of robust state estimation and information fusion techniques,

S. Li, M. L ´ opez-Ben´ ıtez, E. G. Lim, F. Ma, M. Cao, L. Y u, a nd X. Qin, “Enabling cooperative autonomy in UUV clusters: A survey of robust state estimation and information fusion techniques,” Drones, vol. 9, no. 11, 2025. [Online]. Available: https://www.mdpi.com/ 2504-446X/9/ 11/752

work page 2025
[3]

An asynchro nous multi- cluster network system for AUV swarm communication and posi tioning: Design and trial,

G. Wang, J. Zhu, S. Hu, Z. Peng, and J.-H. Cui, “An asynchro nous multi- cluster network system for AUV swarm communication and posi tioning: Design and trial,” IEEE Internet of Things Journal , vol. 12, no. 13, pp. 25 393–25 406, 2025

work page 2025
[4]

Aqua- Sim fourth generation: Toward general and intelligent simu lation for underwater acoustic networks,

J. Guo, S. Song, H. Chen, B. Huangfu, J. Liu, and J.-H. Cui, “Aqua- Sim fourth generation: Toward general and intelligent simu lation for underwater acoustic networks,” IEEE Internet of Things Journal , vol. 12, no. 15, pp. 30 203–30 214, 2025

work page 2025
[5]

Information-driven Path Planning for Hybrid Aerial Under water V ehi- cles,

Z. Zeng, C. Xiong, X. Y uan, H. Zhou, Y . Bai, Y . Jin, D. Lu, an d L. Lian, “Information-driven Path Planning for Hybrid Aerial Under water V ehi- cles,” IEEE Journal of Oceanic Engineering , vol. 48, no. 3, pp. 689–715, 2023

work page 2023
[6]

Underw ater federated learning: Empowering autonomous underwater veh icle swarm with online learning capabilities,

X. Wang, X. Hou, F. Guan, J. Du, J. Wang, and Y . Ren, “Underw ater federated learning: Empowering autonomous underwater veh icle swarm with online learning capabilities,” GLOBECOM 2024 - 2024 IEEE Global Communications Conference , pp. 379–384, 2024. [Online]. Available: https://api.semanticscholar.org/CorpusID:276936700

work page 2024
[7]

A factor graph with Mer cator projection-based cooperative localization algorithm for multiple AUVs,

Q. Li, H. Huang, Y . Sun, and Y . Ben, “A factor graph with Mer cator projection-based cooperative localization algorithm for multiple AUVs,” IEEE Transactions on Instrumentation and Measurement , vol. 74, pp. 1–10, 2025

work page 2025
[8]

A factor graph-based multiple sensors fusion method for enhanced underwater nav igation of remotely operated vehicles,

S. Cheng, H. Zhu, X. Qu, and Y . Wang, “A factor graph-based multiple sensors fusion method for enhanced underwater nav igation of remotely operated vehicles,” in 2024 IEEE International Conference on Unmanned Systems (ICUS) , 2024, pp. 1980–1984

work page 2024
[9]

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

C. Cadena et al. , “Past, present, and future of simultaneous localization and mapping: Toward the robust-perception age,” IEEE Transactions on Robotics, vol. 32, no. 6, pp. 1309–1332, 2016

work page 2016
[10]

Increment al sparse Gaussian process-based model predictive control for traje ctory tracking of unmanned underwater vehicles,

Y . Dang, Y . Huang, X. Shen, D. Zhu, and Z. Chu, “Increment al sparse Gaussian process-based model predictive control for traje ctory tracking of unmanned underwater vehicles,” IEEE Robotics and Automation Letters, vol. 10, no. 3, pp. 2327–2334, 2025

work page 2025
[11]

Gaussian process regression for improved underwater navigation,

N. Cohen and I. Klein, “Gaussian process regression for improved underwater navigation,” in 2025 IEEE/ION Position, Location and Navigation Symposium (PLANS) , 2025, pp. 1125–1132

work page 2025
[12]

Asynchronous V ariational- Bayes Kalman ﬁltering,

M. Greiff and K. Berntorp, “Asynchronous V ariational- Bayes Kalman ﬁltering,” in 2024 IEEE 63rd Conference on Decision and Control (CDC), 2024, pp. 1987–1992

work page 2024
[13]

One-step solut ion for the multistep out-of-sequence-measurement problem in tra cking,

Y . Bar-Shalom, H. Chen, and M. Mallick, “One-step solut ion for the multistep out-of-sequence-measurement problem in tra cking,” IEEE Transactions on Aerospace and Electronic Systems , vol. 40, no. 1, pp. 27–37, 2004

work page 2004
[14]

OOSM-adaptive sequential To bit Kalman ﬁlter (OS-TKF) for distributed X-Ray pulsar-based navigat ion systems,

J. Li, J. Song, and J. Liu, “OOSM-adaptive sequential To bit Kalman ﬁlter (OS-TKF) for distributed X-Ray pulsar-based navigat ion systems,” IEEE Transactions on Aerospace and Electronic Systems , vol. 61, no. 3, pp. 6564–6575, 2025

work page 2025
[15]

An underwater multisource fusion anomaly detection navigation algorith m based on factor graph and LSTM,

Y . Hou, L. Cheng, Q. Wang, J. Li, and Y . Ke, “An underwater multisource fusion anomaly detection navigation algorith m based on factor graph and LSTM,” IEEE Transactions on Instrumentation and Measurement, vol. 74, pp. 1–11, 2025

work page 2025
[16]

Communication Outage-Resistant UUV State Estimation: A Variational History Distillation Approach

S. Li, M. L ´ opez-Ben´ ıtez, E. G. Lim, F. Ma, Q. Dong, M. Ca o, L. Y u, and X. Qin, “Communication Outage-Resistant UUV State Esti mation: A V ariational History Distillation Approach,” 2026. [Onli ne]. Available: https://arxiv.org/abs/2603.29512

work page internal anchor Pith review Pith/arXiv arXiv 2026