Neural Aided Kalman Filtering for UAV State Estimation in Degraded Sensing Environments

Akhil Gupta; Erhan Guven

arxiv: 2604.28107 · v1 · submitted 2026-04-30 · 💻 cs.LG

Neural Aided Kalman Filtering for UAV State Estimation in Degraded Sensing Environments

Akhil Gupta , Erhan Guven This is my paper

Pith reviewed 2026-05-07 06:47 UTC · model grok-4.3

classification 💻 cs.LG

keywords Bayesian neural networksKalman filteringUAV state estimationdegraded sensingnonlinear dynamicshybrid filtersuncertainty quantification

0 comments

The pith

A Bayesian neural network Kalman filter improves UAV state estimation accuracy under noisy and sparse sensor conditions compared to standard filters.

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

The paper proposes the Bayesian Neural Kalman Filter as a hybrid method that trains a Bayesian neural network on synthetic nonlinear UAV trajectories to generate state predictions and uncertainty estimates, then feeds those into a Kalman filter correction step. This is tested via five-fold cross validation on data simulating agile UAV motion with varying radar noise levels and sampling rates. The approach aims to overcome the breakdown of classical Extended and Unscented Kalman Filters when measurements are degraded, unknown inputs are present, and dynamics are highly nonlinear. A sympathetic reader would care because reliable real-time tracking of fast-moving UAVs is needed for applications ranging from air traffic safety to countering adversarial drones, and the method also supplies uncertainty bounds that can guide downstream decisions. An ensemble version of the filter is shown to further boost precision in the toughest noise cases.

Core claim

The Bayesian Neural Kalman Filter (BNKF) replaces the traditional process model in a Kalman filter with predictions from a Bayesian neural network that outputs full state vectors along with uncertainties obtained through Monte Carlo sampling over the posterior distribution of network weights. These Bayesian uncertainties are incorporated directly into the covariance propagation during the measurement update step. On synthetic nonlinear UAV flight trajectories under different radar noise levels and sampling rates, five-fold cross validation shows BNKF outperforming both the Extended Kalman Filter and Unscented Kalman Filter in estimation accuracy, precision, and the frequency with which true

What carries the argument

Bayesian Neural Kalman Filter (BNKF): a hybrid that uses a Bayesian neural network to supply state predictions and Bayesian uncertainties which are then used in the Kalman filter's correction and covariance update to handle nonlinear dynamics and degraded sensing.

If this is right

BNKF achieves higher accuracy and better truth containment than EKF or UKF under high-noise and low-sampling-rate conditions.
The ensemble variant BNKFe delivers improved precision in high-noise edge cases with only a minor accuracy tradeoff.
The method incurs only minimal inference overhead, supporting real-time deployment on UAV platforms.
It handles agile nonlinear motion and unknown control inputs more robustly than classical Kalman variants.

Where Pith is reading between the lines

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

If the synthetic-trained model generalizes, this hybrid could enable safer autonomous navigation or interception in real environments with sensor interference or failure.
The built-in uncertainty estimates could support adaptive sensor scheduling or risk-aware control decisions in tracking systems.
Applying the same BNN-Kalman coupling to other nonlinear platforms such as ground robots or spacecraft would test whether the gains are UAV-specific or more general.

Load-bearing premise

A Bayesian neural network trained on synthetic nonlinear UAV trajectories will produce reliable state predictions and well-calibrated uncertainties when deployed on real-world flights with actual sensor degradation and unmodeled disturbances.

What would settle it

Collect real UAV flight data under degraded radar conditions, run the trained BNKF on it, and compare root-mean-square error and uncertainty coverage rates against EKF and UKF; if BNKF does not outperform or its uncertainties are miscalibrated, the performance claim is falsified.

Figures

Figures reproduced from arXiv: 2604.28107 by Akhil Gupta, Erhan Guven.

**Figure 1.** Figure 1: Example Simulated Trajectories. II. RELATED WORK Several lines of prior research intersect with the topic of this paper. For instance, deep learning methods have been explored as replacements for traditional linear solvers, particularly in applications such as solving the Poisson equation [6]. Physics-Informed Neural Networks (PINNs) have shown promise in modeling nonlinear dynamics, including recent work… view at source ↗

**Figure 3.** Figure 3: Example Downsampled Trajectories each trajectory with sampling rates of 0.75 and 0.50 (Figure 3), alongside the original full-measurement sequences. These downsampled trajectories were concatenated into the train/test datasets for both the baseline and proposed models. Our hypothesis is that, under lower sampling conditions, the benchmark Kalman Filter’s position estimates will diverge more frequently and… view at source ↗

**Figure 4.** Figure 4: NN versus BNN [23] the system state to the measurement space. The analytical Kullback–Leibler (KL) divergence loss is LKLD = − 1 2 X d i=1 view at source ↗

**Figure 5.** Figure 5: BNKF Process Flow view at source ↗

**Figure 6.** Figure 6: BNKFe Process Flow outputs are then passed to the standard KF based corrector as we did with the BNKF (Equation 6), (Equation 7). The goal of this ensemble design is to reduce individual model complexity while maintaining full-state predictive capability in a stable and robust manner. Ensemble machine learning architectures are generally intended as a technique to mitigate general model uncertainty in mach… view at source ↗

**Figure 7.** Figure 7: Single Trajectory Inference Time Comparison view at source ↗

**Figure 9.** Figure 9: Noise Level Averaged Method Performance VII. DISCUSSION Overall, the BNKF outperforms its EKF and UKF counterpart when metrics are averaged across all scenarios (Figure 9). We additionally note that its advantage becomes pronounced in mid to high noise level datasets as sensor noise increases ( view at source ↗

**Figure 10.** Figure 10: Method Comparison across Metrics percent of observations were removed. Here, the BNKF demonstrates a clear advantage, with an average Euclidean Distance (ED) error of 9 meters compared to 30+ meters for the UKF/EKF—a substantial improvement in state estimation accuracy. Here we also note that the projected uncertainty bounds the error consistently among all three methods, implying that the BNKF maintains… view at source ↗

read the original abstract

Accurate state estimation of nonlinear dynamical systems is fundamental to modern aerospace operations across air, sea, and space domains. Online tracking of adversarial unmanned aerial vehicles (UAVs) is especially challenging due to agile nonlinear motion, noisy and sparse sensor measurements, and unknown control inputs; conditions that violate key assumptions of classical Kalman filter variants and degrade estimation performance. Neural networks (NNs) can learn complex nonlinear relationships from data, but lack principled uncertainty quantification, which is critical for state estimation tasks where confidence bounds drive downstream decisions. We address this with Bayesian Neural Networks (BNNs), which model uncertainty through distributions over network weights and produce predictive means and uncertainties via Monte Carlo sampling. Building on this, we propose the Bayesian Neural Kalman Filter (BNKF): a hybrid framework coupling a trained BNN with a Kalman correction step for robust online UAV state estimation. Unlike related neural Kalman approaches, BNKF produces full state predictions and incorporates Bayesian uncertainty directly into covariance propagation, improving robustness under high noise conditions. We evaluate BNKF under varying radar noise levels and sampling rates using synthetic nonlinear UAV flight data. Five fold cross validation demonstrates that BNKF outperforms Extended and Unscented Kalman Filters in accuracy, precision, and truth containment under degraded sensing. An ensemble variant (BNKFe) further improves precision in high-noise edge cases at a slight accuracy tradeoff. Runtime analysis confirms minimal inference overhead, supporting real-time deployment feasibility.

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

BNKF is a reasonable hybrid that feeds BNN uncertainty straight into Kalman covariance updates, but the synthetic-only results with no numbers shown leave the real gains unproven.

read the letter

The main takeaway is that this paper proposes a Bayesian Neural Kalman Filter that trains a BNN on simulated UAV trajectories and then plugs both its mean predictions and its sampled uncertainties directly into the Kalman prediction and update steps. That specific insertion of Bayesian covariance is presented as the distinction from earlier neural Kalman work, and it is a clean, mechanically plausible move for handling sparse noisy radar measurements on agile platforms. The runtime numbers they give suggest the extra sampling cost stays low enough for real-time use, which is a practical plus if the accuracy holds up. Five-fold cross-validation on the synthetic set is at least a controlled way to compare against EKF and UKF, and the ensemble variant is a sensible extra knob for high-noise cases. Those pieces are done competently on their own terms. The soft spots are not small. The abstract supplies no actual RMSE values, no confidence intervals, no ablation on the covariance insertion mechanism, and no description of the BNN architecture or training details beyond the free parameters. More importantly, every result stays inside the same synthetic distribution used for training, with no real flight logs, no domain-randomization tests, and no out-of-distribution sensor or dynamics shifts. The stress-test note is correct on this point: without evidence that the BNN uncertainties remain calibrated when wind, actuator lag, or actual radar characteristics appear, the claim of better accuracy and truth containment under degraded sensing is still an in-distribution simulation result. For readers working on UAV sensor fusion who already know both Kalman filters and Bayesian nets, the paper gives a clear recipe worth trying in their own simulators. It does not yet supply a tool one would drop onto a real platform or cite as settled evidence. A serious editor should send it to review, but only with the expectation that the authors add the missing quantitative tables, implementation specifics, and at least one stronger validation step before the central claim can be treated as demonstrated.

Referee Report

2 major / 2 minor

Summary. The paper proposes the Bayesian Neural Kalman Filter (BNKF), a hybrid method that couples a Bayesian neural network (trained to predict UAV states and uncertainties via Monte Carlo sampling) with a Kalman filter correction step. On synthetic nonlinear UAV trajectories with injected radar noise and variable sampling rates, five-fold cross-validation is used to claim that BNKF outperforms Extended and Unscented Kalman Filters in accuracy, precision, and truth containment under degraded sensing; an ensemble variant (BNKFe) further improves precision in high-noise cases at a small accuracy cost. Runtime analysis supports real-time feasibility.

Significance. If the hybrid integration is sound and the comparative gains hold under the stated synthetic conditions, the work provides a concrete example of incorporating BNN-derived uncertainties into classical filtering for improved robustness in noisy UAV tracking. This could be useful for aerospace applications where principled uncertainty matters for downstream decisions. The controlled synthetic evaluation allows isolation of noise and sampling effects, but the absence of real-flight validation or domain-shift tests limits broader claims about degraded sensing environments.

major comments (2)

[§3] §3 (Method description): The manuscript does not specify how the BNN's Monte Carlo-sampled predictive means and covariances are inserted into the Kalman prediction or update equations. This mechanism is load-bearing for the central claim that BNKF improves robustness under high noise, as the hybrid benefit cannot be assessed or reproduced without it.
[§4] §4 (Evaluation): The five-fold cross-validation results are described only qualitatively (outperformance in accuracy/precision/truth containment) with no reported numerical metrics such as RMSE, MAE, or confidence intervals, nor any ablation on BNN components or covariance usage. This prevents quantitative evaluation of the claimed gains over EKF/UKF.

minor comments (2)

[Abstract] Abstract and §4: Define 'truth containment' explicitly (e.g., percentage of true states within predicted bounds) and report the exact BNN architecture, training hyperparameters, and synthetic trajectory generator details for reproducibility.
[Introduction] Introduction: The title references 'Degraded Sensing Environments' while the evaluation is confined to synthetic data; a short paragraph acknowledging sim-to-real gaps would clarify scope without altering the central claim.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their thoughtful review and constructive feedback on our manuscript. We appreciate the recognition of the potential utility of the BNKF approach for UAV state estimation in degraded sensing conditions. Below, we address each major comment point by point, indicating the revisions we plan to make.

read point-by-point responses

Referee: [§3] §3 (Method description): The manuscript does not specify how the BNN's Monte Carlo-sampled predictive means and covariances are inserted into the Kalman prediction or update equations. This mechanism is load-bearing for the central claim that BNKF improves robustness under high noise, as the hybrid benefit cannot be assessed or reproduced without it.

Authors: We agree that a clear specification of the integration mechanism is essential for reproducibility and understanding the hybrid benefit. In the revised manuscript, we will expand §3 to include the explicit equations showing how the BNN's Monte Carlo-sampled predictive mean and covariance are used to initialize or augment the Kalman filter's prediction step, and how the uncertainty is propagated into the update equations. Specifically, the BNN provides the state prediction and covariance from the MC samples, which are then used in the standard Kalman update with the measurement. This will be detailed with pseudocode and mathematical derivations to clarify the covariance propagation. revision: yes
Referee: [§4] §4 (Evaluation): The five-fold cross-validation results are described only qualitatively (outperformance in accuracy/precision/truth containment) with no reported numerical metrics such as RMSE, MAE, or confidence intervals, nor any ablation on BNN components or covariance usage. This prevents quantitative evaluation of the claimed gains over EKF/UKF.

Authors: We acknowledge that the current presentation relies on qualitative descriptions of the five-fold cross-validation outcomes. To address this, we will include quantitative metrics in the revised §4, such as tables reporting RMSE, MAE, and average containment rates with confidence intervals across the folds for BNKF, BNKFe, EKF, and UKF under different noise levels and sampling rates. Additionally, we will add an ablation study examining the impact of using BNN-derived covariances versus fixed covariances, and the contribution of the ensemble variant. This will allow for a more rigorous quantitative comparison. revision: yes

Circularity Check

0 steps flagged

No significant circularity in derivation or claims

full rationale

The paper's central contribution is an empirical hybrid estimator (BNKF) that trains a BNN on external synthetic nonlinear UAV trajectories, then couples its predictive mean and covariance into a standard Kalman correction step. All performance numbers (accuracy, precision, truth containment) are obtained via five-fold cross-validation on held-out synthetic runs with injected noise; these are comparative results against EKF/UKF on the same data distribution and do not reduce, by the paper's own equations, to quantities defined solely in terms of fitted parameters or self-referential normalizations. No self-definitional steps, fitted-input-called-prediction patterns, or load-bearing self-citations appear in the derivation chain. The method remains falsifiable through the reported simulation experiments and does not smuggle ansatzes or rename known results as novel derivations.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 1 invented entities

The central claim rests on the BNN learning useful dynamics and calibrated uncertainties from synthetic data and on the validity of feeding those uncertainties directly into the Kalman covariance update; all evaluation occurs inside a closed simulation loop.

free parameters (1)

BNN architecture and training hyperparameters
Network depth, width, prior distributions, learning rate, and number of Monte Carlo samples are chosen to fit the synthetic UAV trajectories.

axioms (1)

domain assumption Synthetic nonlinear UAV dynamics and sensor models are representative of real degraded-sensing conditions
All training and testing data are generated from assumed flight equations; no real sensor traces are used.

invented entities (1)

Bayesian Neural Kalman Filter (BNKF) no independent evidence
purpose: Hybrid estimator that fuses BNN predictions with Kalman correction while propagating Bayesian uncertainty
Newly defined framework whose performance is demonstrated only on the synthetic experiments described in the abstract.

pith-pipeline@v0.9.0 · 5552 in / 1565 out tokens · 120502 ms · 2026-05-07T06:47:35.383249+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

28 extracted references · 21 canonical work pages · 1 internal anchor

[1]

Cusack, R.,

J. de Curt `o and I. de Zarz `a, “Hybrid State Estimation: Integrat- ing Physics-Informed Neural Networks with Adaptive UKF for Dy- namic Systems,” *Electronics*, vol. 13, no. 11, p. 2208, 2024, doi: 10.3390/electronics13112208

work page doi:10.3390/electronics13112208 2024
[2]

The smooth variable structure filter: A comprehensive re- view,

M. Avzayesh, M. Abdel-Hafez, M. AlShabi, and S. A. Gads- den, “The smooth variable structure filter: A comprehensive re- view,” *Digital Signal Processing*, vol. 110, p. 102912, 2021, doi: 10.1016/j.dsp.2020.102912

work page doi:10.1016/j.dsp.2020.102912 2021
[3]

Unmanned-Aerial-Vehicle (UA V) Drones Market Size Expected to Reach USD 169.7 Bn by 2033,

Precedence Research, “Unmanned-Aerial-Vehicle (UA V) Drones Market Size Expected to Reach USD 169.7 Bn by 2033,” *GlobeNewswire*, Aug. 2024. [On- line]. Available: [https://www.globenewswire.com/news- release/2024/08/09/2927741/0/en/Unmanned-Aerial-Vehicle- UA V-Drones-Market-Size-Expected-to-Reach-USD-169-7- Bn-by-2033.html](https://www.globenewswire.com...

2033
[4]

Neural network-aided adaptive unscented Kalman filter for nonlinear state estimation,

R. Zhan and J. Wan, “Neural network-aided adaptive unscented Kalman filter for nonlinear state estimation,” *IEEE Signal Pro- cessing Letters*, vol. 13, no. 7, pp. 445–448, 2006, doi: 10.1109/LSP.2006.871854

work page doi:10.1109/lsp.2006.871854 2006
[5]

Neural Networks: A General Framework for Non- Linear Function Approximation,

M. Fischer, “Neural Networks: A General Framework for Non- Linear Function Approximation,” *Transactions in GIS*, vol. 10, pp. 521–533, Jul. 2006, doi: 10.1111/j.1467-9671.2006.01010.x

work page doi:10.1111/j.1467-9671.2006.01010.x 2006
[6]

The Old and the New: Can Physics- Informed Deep-Learning Replace Traditional Linear Solvers?,

S. Markidis, “The Old and the New: Can Physics- Informed Deep-Learning Replace Traditional Linear Solvers?,” *Frontiers in Big Data*, vol. 4, 2021. [Online]. Available: [https://api.semanticscholar.org/CorpusID:235743161]

2021
[7]

Tracking an untracked space debris after an inelastic collision using physics in- formed neural network,

H. M., G. Singh, V . Kumar, A. B. Buduru, and S. K. Biswas, “Tracking an untracked space debris after an inelastic collision using physics in- formed neural network,” *Scientific Reports (Nature)*, 2024. [Online]. Available: [https://www.nature.com/articles/s41598-024-51897-9]

2024
[8]

Synthetic-UA V-Flight-Trajectories,

riotu-lab, “Synthetic-UA V-Flight-Trajectories,” *HuggingFace*, 2024. [Online]. Available: [https://huggingface.co/datasets/riotu-lab/Synthetic-UA V-Flight- Trajectories](https://huggingface.co/datasets/riotu-lab/Synthetic-UA V- Flight-Trajectories)

2024
[9]

VECTOR: Velocity-Enhanced GRU Neural Net- work for Real-Time 3D UA V Trajectory Prediction,

O. Nacar *et al.*, “VECTOR: Velocity-Enhanced GRU Neural Net- work for Real-Time 3D UA V Trajectory Prediction,” *Drones*, vol. 9, no. 1, p. 8, 2025, doi: 10.3390/drones9010008

work page doi:10.3390/drones9010008 2025
[10]

Stone Soup: A framework for multi-target tracking and state estimation,

Defence Science and Technology Laboratory (Dstl), “Stone Soup: A framework for multi-target tracking and state estimation,” *GitHub*, 2020. [Online]. Available: [https://github.com/dstl/Stone- Soup](https://github.com/dstl/Stone-Soup). [Accessed: May 21, 2025]

2020
[11]

Invariant Kalman Filtering,

A. Barrau and S. Bonnabel, “Invariant Kalman Filtering,” *Annual Review of Control, Robotics, and Autonomous Systems*, vol. 1, May 2018, doi: 10.1146/annurev-control-060117-105010

work page doi:10.1146/annurev-control-060117-105010 2018
[12]

Sigma-Point Filters: An Overview with Applications to Integrated Navigation and Vision Assisted Control,

E. Wan, “Sigma-Point Filters: An Overview with Applications to Integrated Navigation and Vision Assisted Control,” in *Proc. 2006 IEEE Nonlinear Statistical Signal Processing Workshop*, 2006, pp. 201–202, doi: 10.1109/NSSPW ˙2006.4378854

work page doi:10.1109/nsspw 2006
[13]

Nonlinear signal processing using neural networks: Prediction and system modelling,

A. Lapedes and R. Farber, “Nonlinear signal processing using neural networks: Prediction and system modelling,” *Los Alamos National Laboratory / OSTI*, Jun. 1987. [Online]. Available: [https://www.osti.gov/biblio/5470451]

work page arXiv 1987
[14]

Predictive non-linear modeling of complex data by artificial neural networks,

J. S. Almeida, “Predictive non-linear modeling of complex data by artificial neural networks,” *Current Opinion in Biotechnology*, vol. 13, no. 1, pp. 72–76, 2002, doi: 10.1016/S0958-1669(02)00288-4

work page doi:10.1016/s0958-1669(02)00288-4 2002
[15]

Iden- tification of process and measurement noise covariance for state and parameter estimation using extended Kalman filter,

V . A. Bavdekar, A. P. Deshpande, and S. C. Patwardhan, “Iden- tification of process and measurement noise covariance for state and parameter estimation using extended Kalman filter,” *Jour- nal of Process Control*, vol. 21, no. 4, pp. 585–601, 2011, doi: 10.1016/j.jprocont.2011.01.001

work page doi:10.1016/j.jprocont.2011.01.001 2011
[16]

Adam: A Method for Stochastic Optimization

D. P. Kingma and J. Ba, “Adam: A Method for Stochastic Optimiza- tion,” *arXiv preprint*, arXiv:1412.6980, 2017. [Online]. Available: [https://arxiv.org/abs/1412.6980] (https://arxiv.org/abs/1412.6980)

work page internal anchor Pith review arXiv 2017
[17]

Julier, J

S. Julier, J. Uhlmann and H. F. Durrant-Whyte, ”A new method for the nonlinear transformation of means and covariances in filters and estimators,” in IEEE Transactions on Automatic Control, vol. 45, no. 3, pp. 477-482, March 2000, doi: 10.1109/9.847726. keywords: Filtering;Additive noise;Nonlinear systems;Nonlinear filters;State es- timation;Probability d...

work page doi:10.1109/9.847726 2000
[18]

Culotta, Fabrizio. (2023). Given N forecasting models, what to do?. 317-322. 10.36253/979-12-215-0106-3.55

work page doi:10.36253/979-12-215-0106-3.55 2023
[19]

Revach, N

G. Revach, N. Shlezinger, R. J. G. van Sloun and Y . C. El- dar, ”Kalmannet: Data-Driven Kalman Filtering,” ICASSP 2021 - 2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), Toronto, ON, Canada, 2021, pp. 3905-3909, doi: 10.1109/ICASSP39728.2021.9413750. keywords: Training;Computational modeling;Signal processing algori...

work page doi:10.1109/icassp39728.2021.9413750 2021
[20]

Coskun, Huseyin & Achilles, Felix & DiPietro, Robert & Navab, Nassir & Tombari, Federico. (2017). Long Short-Term Memory Kalman Filters:Recurrent Neural Estimators for Pose Regularization. 10.48550/arXiv.1708.01885

work page doi:10.48550/arxiv.1708.01885 2017
[21]

Ding, Di & He, Kai & Qian, Wei. (2021). A Bayesian Adaptive Unscented Kalman Filter for Aircraft Parameter and Noise Estimation. Journal of Sensors. 2021. 10.1155/2021/9002643

work page doi:10.1155/2021/9002643 2021
[22]

Lee, Sungyoon & Kim, Hoki & Lee, Jaewook. (2022). Graddiv: Adversarial robustness of randomized neural networks via gradient diversity regularization. IEEE Transactions on Pattern Analysis and Machine Intelligence. 2022

2022
[23]

A first insight into Bayesian Neural Networks (BNN),

A. Costa Leirbag, “A first insight into Bayesian Neural Networks (BNN),” Medium, 2019. [Online]. Available: https://medium.com/@costaleirbag/a-first-insight-into-bayesian- neural-networks-bnn-c767551e9526

2019
[24]

Ge, B., Zhang, H., Jiang, L., Li, Z., & Butt, M. M. (2019). Adaptive Unscented Kalman Filter for Target Tracking with Un- known Time-Varying Noise Covariance. Sensors, 19(6), 1371. https://doi.org/10.3390/s19061371

work page doi:10.3390/s19061371 2019
[25]

Perrusqu ´ıa, A., Guo, W., Fraser, B. et al. Uncovering drone intentions using control physics informed machine learning. Commun Eng 3, 36 (2024). https://doi.org/10.1038/s44172-024-00179-3

work page doi:10.1038/s44172-024-00179-3 2024
[26]

Khazraj, F

H. Khazraj, F. Faria da Silva and C. L. Bak, ”A performance comparison between extended Kalman Filter and unscented Kalman Filter in power system dynamic state estimation,” 2016 51st Interna- tional Universities Power Engineering Conference (UPEC), Coimbra, Portugal, 2016, pp. 1-6, doi: 10.1109/UPEC.2016.8114125

work page doi:10.1109/upec.2016.8114125 2016
[27]

V ., Buntine, W

Jospin, L. V ., Buntine, W. L., Boussa ¨ıd, F., Laga, H., & Bennamoun, M. (2020). Hands-on Bayesian neural networks: A tutorial for deep learning users. arXiv. https://arxiv.org/abs/2007.06823

work page arXiv 2020
[28]

Akhlaghi, Shahrokh & Zhou, Ning & Huang, Zhenyu. (2017). Adap- tive Adjustment of Noise Covariance in Kalman Filter for Dynamic State Estimation. 10.48550/arXiv.1702.00884

work page doi:10.48550/arxiv.1702.00884 2017

[1] [1]

Cusack, R.,

J. de Curt `o and I. de Zarz `a, “Hybrid State Estimation: Integrat- ing Physics-Informed Neural Networks with Adaptive UKF for Dy- namic Systems,” *Electronics*, vol. 13, no. 11, p. 2208, 2024, doi: 10.3390/electronics13112208

work page doi:10.3390/electronics13112208 2024

[2] [2]

The smooth variable structure filter: A comprehensive re- view,

M. Avzayesh, M. Abdel-Hafez, M. AlShabi, and S. A. Gads- den, “The smooth variable structure filter: A comprehensive re- view,” *Digital Signal Processing*, vol. 110, p. 102912, 2021, doi: 10.1016/j.dsp.2020.102912

work page doi:10.1016/j.dsp.2020.102912 2021

[3] [3]

Unmanned-Aerial-Vehicle (UA V) Drones Market Size Expected to Reach USD 169.7 Bn by 2033,

Precedence Research, “Unmanned-Aerial-Vehicle (UA V) Drones Market Size Expected to Reach USD 169.7 Bn by 2033,” *GlobeNewswire*, Aug. 2024. [On- line]. Available: [https://www.globenewswire.com/news- release/2024/08/09/2927741/0/en/Unmanned-Aerial-Vehicle- UA V-Drones-Market-Size-Expected-to-Reach-USD-169-7- Bn-by-2033.html](https://www.globenewswire.com...

2033

[4] [4]

Neural network-aided adaptive unscented Kalman filter for nonlinear state estimation,

R. Zhan and J. Wan, “Neural network-aided adaptive unscented Kalman filter for nonlinear state estimation,” *IEEE Signal Pro- cessing Letters*, vol. 13, no. 7, pp. 445–448, 2006, doi: 10.1109/LSP.2006.871854

work page doi:10.1109/lsp.2006.871854 2006

[5] [5]

Neural Networks: A General Framework for Non- Linear Function Approximation,

M. Fischer, “Neural Networks: A General Framework for Non- Linear Function Approximation,” *Transactions in GIS*, vol. 10, pp. 521–533, Jul. 2006, doi: 10.1111/j.1467-9671.2006.01010.x

work page doi:10.1111/j.1467-9671.2006.01010.x 2006

[6] [6]

The Old and the New: Can Physics- Informed Deep-Learning Replace Traditional Linear Solvers?,

S. Markidis, “The Old and the New: Can Physics- Informed Deep-Learning Replace Traditional Linear Solvers?,” *Frontiers in Big Data*, vol. 4, 2021. [Online]. Available: [https://api.semanticscholar.org/CorpusID:235743161]

2021

[7] [7]

Tracking an untracked space debris after an inelastic collision using physics in- formed neural network,

H. M., G. Singh, V . Kumar, A. B. Buduru, and S. K. Biswas, “Tracking an untracked space debris after an inelastic collision using physics in- formed neural network,” *Scientific Reports (Nature)*, 2024. [Online]. Available: [https://www.nature.com/articles/s41598-024-51897-9]

2024

[8] [8]

Synthetic-UA V-Flight-Trajectories,

riotu-lab, “Synthetic-UA V-Flight-Trajectories,” *HuggingFace*, 2024. [Online]. Available: [https://huggingface.co/datasets/riotu-lab/Synthetic-UA V-Flight- Trajectories](https://huggingface.co/datasets/riotu-lab/Synthetic-UA V- Flight-Trajectories)

2024

[9] [9]

VECTOR: Velocity-Enhanced GRU Neural Net- work for Real-Time 3D UA V Trajectory Prediction,

O. Nacar *et al.*, “VECTOR: Velocity-Enhanced GRU Neural Net- work for Real-Time 3D UA V Trajectory Prediction,” *Drones*, vol. 9, no. 1, p. 8, 2025, doi: 10.3390/drones9010008

work page doi:10.3390/drones9010008 2025

[10] [10]

Stone Soup: A framework for multi-target tracking and state estimation,

Defence Science and Technology Laboratory (Dstl), “Stone Soup: A framework for multi-target tracking and state estimation,” *GitHub*, 2020. [Online]. Available: [https://github.com/dstl/Stone- Soup](https://github.com/dstl/Stone-Soup). [Accessed: May 21, 2025]

2020

[11] [11]

Invariant Kalman Filtering,

A. Barrau and S. Bonnabel, “Invariant Kalman Filtering,” *Annual Review of Control, Robotics, and Autonomous Systems*, vol. 1, May 2018, doi: 10.1146/annurev-control-060117-105010

work page doi:10.1146/annurev-control-060117-105010 2018

[12] [12]

Sigma-Point Filters: An Overview with Applications to Integrated Navigation and Vision Assisted Control,

E. Wan, “Sigma-Point Filters: An Overview with Applications to Integrated Navigation and Vision Assisted Control,” in *Proc. 2006 IEEE Nonlinear Statistical Signal Processing Workshop*, 2006, pp. 201–202, doi: 10.1109/NSSPW ˙2006.4378854

work page doi:10.1109/nsspw 2006

[13] [13]

Nonlinear signal processing using neural networks: Prediction and system modelling,

A. Lapedes and R. Farber, “Nonlinear signal processing using neural networks: Prediction and system modelling,” *Los Alamos National Laboratory / OSTI*, Jun. 1987. [Online]. Available: [https://www.osti.gov/biblio/5470451]

work page arXiv 1987

[14] [14]

Predictive non-linear modeling of complex data by artificial neural networks,

J. S. Almeida, “Predictive non-linear modeling of complex data by artificial neural networks,” *Current Opinion in Biotechnology*, vol. 13, no. 1, pp. 72–76, 2002, doi: 10.1016/S0958-1669(02)00288-4

work page doi:10.1016/s0958-1669(02)00288-4 2002

[15] [15]

Iden- tification of process and measurement noise covariance for state and parameter estimation using extended Kalman filter,

V . A. Bavdekar, A. P. Deshpande, and S. C. Patwardhan, “Iden- tification of process and measurement noise covariance for state and parameter estimation using extended Kalman filter,” *Jour- nal of Process Control*, vol. 21, no. 4, pp. 585–601, 2011, doi: 10.1016/j.jprocont.2011.01.001

work page doi:10.1016/j.jprocont.2011.01.001 2011

[16] [16]

Adam: A Method for Stochastic Optimization

D. P. Kingma and J. Ba, “Adam: A Method for Stochastic Optimiza- tion,” *arXiv preprint*, arXiv:1412.6980, 2017. [Online]. Available: [https://arxiv.org/abs/1412.6980] (https://arxiv.org/abs/1412.6980)

work page internal anchor Pith review arXiv 2017

[17] [17]

Julier, J

S. Julier, J. Uhlmann and H. F. Durrant-Whyte, ”A new method for the nonlinear transformation of means and covariances in filters and estimators,” in IEEE Transactions on Automatic Control, vol. 45, no. 3, pp. 477-482, March 2000, doi: 10.1109/9.847726. keywords: Filtering;Additive noise;Nonlinear systems;Nonlinear filters;State es- timation;Probability d...

work page doi:10.1109/9.847726 2000

[18] [18]

Culotta, Fabrizio. (2023). Given N forecasting models, what to do?. 317-322. 10.36253/979-12-215-0106-3.55

work page doi:10.36253/979-12-215-0106-3.55 2023

[19] [19]

Revach, N

G. Revach, N. Shlezinger, R. J. G. van Sloun and Y . C. El- dar, ”Kalmannet: Data-Driven Kalman Filtering,” ICASSP 2021 - 2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), Toronto, ON, Canada, 2021, pp. 3905-3909, doi: 10.1109/ICASSP39728.2021.9413750. keywords: Training;Computational modeling;Signal processing algori...

work page doi:10.1109/icassp39728.2021.9413750 2021

[20] [20]

Coskun, Huseyin & Achilles, Felix & DiPietro, Robert & Navab, Nassir & Tombari, Federico. (2017). Long Short-Term Memory Kalman Filters:Recurrent Neural Estimators for Pose Regularization. 10.48550/arXiv.1708.01885

work page doi:10.48550/arxiv.1708.01885 2017

[21] [21]

Ding, Di & He, Kai & Qian, Wei. (2021). A Bayesian Adaptive Unscented Kalman Filter for Aircraft Parameter and Noise Estimation. Journal of Sensors. 2021. 10.1155/2021/9002643

work page doi:10.1155/2021/9002643 2021

[22] [22]

Lee, Sungyoon & Kim, Hoki & Lee, Jaewook. (2022). Graddiv: Adversarial robustness of randomized neural networks via gradient diversity regularization. IEEE Transactions on Pattern Analysis and Machine Intelligence. 2022

2022

[23] [23]

A first insight into Bayesian Neural Networks (BNN),

A. Costa Leirbag, “A first insight into Bayesian Neural Networks (BNN),” Medium, 2019. [Online]. Available: https://medium.com/@costaleirbag/a-first-insight-into-bayesian- neural-networks-bnn-c767551e9526

2019

[24] [24]

Ge, B., Zhang, H., Jiang, L., Li, Z., & Butt, M. M. (2019). Adaptive Unscented Kalman Filter for Target Tracking with Un- known Time-Varying Noise Covariance. Sensors, 19(6), 1371. https://doi.org/10.3390/s19061371

work page doi:10.3390/s19061371 2019

[25] [25]

Perrusqu ´ıa, A., Guo, W., Fraser, B. et al. Uncovering drone intentions using control physics informed machine learning. Commun Eng 3, 36 (2024). https://doi.org/10.1038/s44172-024-00179-3

work page doi:10.1038/s44172-024-00179-3 2024

[26] [26]

Khazraj, F

H. Khazraj, F. Faria da Silva and C. L. Bak, ”A performance comparison between extended Kalman Filter and unscented Kalman Filter in power system dynamic state estimation,” 2016 51st Interna- tional Universities Power Engineering Conference (UPEC), Coimbra, Portugal, 2016, pp. 1-6, doi: 10.1109/UPEC.2016.8114125

work page doi:10.1109/upec.2016.8114125 2016

[27] [27]

V ., Buntine, W

Jospin, L. V ., Buntine, W. L., Boussa ¨ıd, F., Laga, H., & Bennamoun, M. (2020). Hands-on Bayesian neural networks: A tutorial for deep learning users. arXiv. https://arxiv.org/abs/2007.06823

work page arXiv 2020

[28] [28]

Akhlaghi, Shahrokh & Zhou, Ning & Huang, Zhenyu. (2017). Adap- tive Adjustment of Noise Covariance in Kalman Filter for Dynamic State Estimation. 10.48550/arXiv.1702.00884

work page doi:10.48550/arxiv.1702.00884 2017