Pith. sign in

REVIEW 4 major objections 5 minor 48 references

DeepUKF-VIN: Adaptively-tuned Deep Unscented Kalman Filter for 3D Visual-Inertial Navigation based on IMU-Vision-Net

T0 review · 4 major / 5 minor · reviewed 2026-08-09 · deepseek-v4-flash

Pith's one-line read This paper claims that a quaternion-based Unscented Kalman Filter whose noise covariances are set by two small neural networks—one reading recent IMU measurements and one reading stereo images—consistently beats the standard UKF for…

desk verdict The derivation is clean and the idea is plausible, but the paper's own Table III contradicts the central claim of consistent superiority on held-out data, and the training/validation split is misreported. read the letter →

arxiv 2502.00575 v2 pith:Z2HVVKZ2 submitted 2025-02-01 cs.RO cs.SYeess.SY

classification cs.ROcs.SYeess.SY
keywords UnscentedKalmanFilterVisual-InertialNavigationAdaptiveCovarianceTuningDeepLearningQuaternionSensorFusionGPS-denied
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper tries to establish that adaptive noise-covariance tuning by a deep network can replace manual filter tuning in visual-inertial navigation. Its proposed filter, DeepUKF-VIN, wraps a quaternion-based Unscented Kalman Filter with two learned components—IMU-Net, which reads the last ten IMU readings, and Vision-Net, which reads the latest stereo pair—that output multiplicative scaling factors for the filter's noise covariance matrices. On two real-world sequences from the EuRoC dataset, including one never seen in training, the adaptively tuned UKF reports lower loss and lower mean squared error than the standard UKF and than a learning-augmented EKF, across orientation, position, and velocity simultaneously. This matters because manual covariance tuning is a time-consuming bottleneck in Kalman-filter deployment, and the results suggest the tuning step can be learned and transferred across filters.

What carries the argument

The central machinery is a quaternion-based Unscented Kalman Filter whose noise covariance matrices are not fixed but recomputed at each step by the Deep Learning-based Adaptation Mechanism (DLAM). IMU-Net is a two-layer bidirectional GRU that maps the last ten IMU measurements to twelve scaling factors for the gyroscope, accelerometer, and bias covariances; Vision-Net is a small convolutional network that maps the current stereo pair to a single scaling factor for the landmark measurement covariance. The scaling rule c_i = c̄_i $10^{{υ tanh γ_i}}$ keeps the learned covariances positive and bounded around offline-tuned nominal values. The filter itself handles quaternions with the ⊞/⊟ operations, weighted quaternion averaging via the largest eigenvalue of the weighted outer-product matrix, and intermittent vision updates by aggregating IMU steps between image frames.

What would settle it

Run DeepUKF-VIN and the standard UKF on a visual-inertial sequence where landmark world coordinates are not supplied ahead of time and must be estimated online together with pose (e.g., a mapless or SLAM-style setup), using identical nominal covariances and training data; if the learned filter no longer beats the standard UKF, the improvement is attributable to the map, not the learned covariances.

Watch

Extended reading notes

Core claim

The central claim is that learned covariance scaling transfers from an EKF-trained model to UKF inference and improves all navigation outputs. The filter represents attitude as a unit quaternion, propagates $\sigma$ points through the discrete navigation kinematics, computes quaternion-weighted means, and corrects with a Kalman update; the noise covariances in that correction are driven by scaling factors γ from the two networks via c_i = c̄_i $10^{{υ tanh γ_i}}$. The paper reports that on the V1_02 medium sequence (used only for validation, not training for the UKF case) and the V2_02 medium sequence (never seen in training), DeepUKF-VIN outperforms both the standard UKF-VIN and the DeepEKF in orientation, position, and velocity MSE.

Load-bearing premise

The reported gains depend on the assumptions that the world-frame positions of tracked visual landmarks are known well enough (they are said to be triangulated from stereo, but the paper does not show how these coordinates are initialized or corrected without using the pose being estimated) and that noise covariances learned under an EKF improve an UKF without retraining.

Editorial extensions

If this is right

  • If the reported comparison holds, noise-covariance matrices for a UKF can be produced by a small learned network instead of manual trial-and-error tuning.
  • The same DLAM training, done with an EKF for gradient simplicity, improves a different filter type at inference, suggesting the learned noise model is not filter-specific.
  • The improvement is simultaneous across orientation, position, and velocity, not a trade-off among state components.
  • The filter remains stable and accurate with low-rate vision (20 Hz) and low-cost IMU data, the regime where manual tuning is most fragile.

Reading between the lines

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

  • Because the scaling rule only multiplies nominal covariances, applying the same DLAM to other Kalman-type filters, as the authors suggest, is straightforward; the main risk is not architecture but the assumption that the noise model learned from one trajectory family transfers to different motion profiles.
  • A direct testable extension would use the same IMU-Net and Vision-Net outputs to adapt covariances in a system that estimates landmark positions online; the paper's own validation uses pre-triangulated landmark coordinates, so the claim currently covers fixed-map navigation.
  • The loss weights wq = 1000, wp = 600, wv = 100 imply the method prioritizes orientation accuracy; deployments that care equally about position would need those weights rebalanced, which the paper does not explore.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper proposes DeepUKF-VIN, a quaternion-based Unscented Kalman Filter for 3D visual-inertial navigation in which the measurement and process noise covariance matrices are scaled at each time step by two neural networks: IMU-Net, a bidirectional GRU stack, and Vision-Net, a convolutional network. The networks are trained end-to-end by differentiating a weighted mean-square-error loss through the filtering recursion; to avoid differentiation through eigenvalue and singular-value decompositions, an EKF is substituted for the UKF during training. The paper reports experiments on the EuRoC V1_02_medium and V2_02_medium sequences and claims that DeepUKF-VIN consistently outperforms a standard UKF-VIN and a DeepEKF across orientation, position, and velocity estimation.

Significance. The idea of learning covariance corrections from raw sensor data and transferring them to a nonlinear filter is potentially useful, and the quaternion-based UKF formulation in Section IV is technically detailed and mostly coherent. The paper also evaluates on real-world data with a public dataset, which is a strength. However, the central empirical claim of consistent superiority is contradicted by the paper's own held-out results, and the training/evaluation protocol is internally inconsistent. Because the main contribution is empirical validation of an adaptive filtering mechanism, these problems are load-bearing rather than cosmetic.

major comments (4)
  1. [Table III; Section VII; Abstract] The claim that DeepUKF-VIN consistently outperforms UKF-VIN across all navigation components is contradicted by Table III on the held-out V2_02_medium sequence: the orientation MSE is 0.0080 for DeepUKF-VIN versus 0.0026 for UKF-VIN, a factor-of-three degradation. Table II shows essentially no loss improvement on that sequence (250 vs. 251). Since Section VII explicitly identifies V2_02_medium as not used in training, this is not a training-set artifact; the central empirical claim fails on the paper's own test data.
  2. [Section VI vs. Section VII] Section VI states that 'The V1_02_medium part of the EuRoC dataset has been utilized for training,' yet Section VII says that 'DeepUKF-VIN was never trained on either experiment' and presents V1_02_medium as a validation experiment. These statements are mutually inconsistent. Moreover, the loss weights, nominal standard deviations, deviation bound, and loss transient cutoff were chosen with access to V1 data, so the V1 comparison in Table II and Table III is a fit-versus-baseline comparison rather than an independent evaluation.
  3. [Section VI, Algorithm 1] The training procedure explicitly replaces the UKF with an EKF to make gradient computation tractable and then hypothesizes that the learned covariance model transfers to the UKF. This hypothesis is never tested: Section VII does not retrain the network with a UKF, does not compare EKF-trained versus UKF-trained covariance models, and does not analyze whether the learned covariances are consistent with the UKF's noise assumptions. Without such evidence, the reported UKF results cannot be attributed to the DLAM mechanism.
  4. [Eq. (25); Section VI] The measurement update in Eq. (25) uses world-frame landmark coordinates l_w,i, but the manuscript only says that these points are computed by stereo triangulation. It does not specify how l_w,i is initialized or updated relative to the pose being estimated. If l_w,i is derived using the estimated trajectory, Eq. (25) introduces a hidden coupling that can bias the innovation; if l_w,i comes from a prior map, the system is not operating in the fully map-free setting implied by the paper. This point must be clarified for the experiments to be interpretable.
minor comments (5)
  1. [Section IV-A] The first sentence reads 'he filter is initialized' and is missing the leading 'T'; this should be corrected.
  2. [Section V, Eq. (47)] The notation for the IMU-Net input is inconsistent: the text around Eq. (47) uses 'uk−11:k−1', while the definition and Fig. 3 use 'uk−1−dGRU:k−1'; one notation should be used throughout.
  3. [Section VII, Table III] No error bars, repeated-run statistics, or significance tests are reported, so differences such as position MSE 0.3011 versus 0.3070 on V2_02_medium are within plausible random variation and should not be described as consistent improvement.
  4. [Algorithm 1, step 10] The gradient clipping step is written as 'Gradientmini-batch ← max(Gradientmini-batch, 1)', which reads as clipping from below at 1; if a norm cap of 1 is intended, the operation should be stated precisely.
  5. [Section VI] The numerical values of the nominal standard deviations, the deviation bound upsilon, and the UKF tuning parameters (lambda, alpha, beta) are not reported, which hinders reproducibility even though the loss weights are stated.

Circularity Check

1 steps flagged · score 6.0 of 10

V1 validation is a training-set fit, and the held-out V2 result contradicts the claimed consistent superiority.

  1. fitted input called prediction [Section VI, 'Training and Implementation'; Section VII, 'Experimental Validation'; Eq. (58); Tables II and III]
    "The V1 02 medium part of the EuRoC dataset [43] has been utilized for training. ... the total loss is computed as the weighted sum of the mean square errors (MSE) of the individual errors defined in (57). ... To validate the effectiveness of quaternion-based DeepUKF-VIN, the algorithm is tested using the real-world V1 02 medium EuRoC dataset."

    The networks are trained end-to-end on V1 02 medium to minimize the loss in Eq. (58), which is a weighted MSE of exactly the orientation, position, and velocity errors defined in Eq. (57) and reported in Table III. Therefore the V1 row of Tables II and III is a comparison of the trained model on its own training set against untrained baselines; any V1 improvement is a fitted outcome, not a prediction. The paper further states that 'DeepUKF-VIN was never trained on either experiment,' which is contradicted by the Section VI training sentence. Hence part of the central claim of consistent superiority reduces, by construction, to evaluating a fit on the data used to produce that fit.

full rationale

The filter derivation itself is largely self-contained: the quaternion UKF equations are standard, and the DLAM covariance-scaling formula in Eq. (50) is explicitly attributed to external prior work [32,33]. The significant circularity enters at the evaluation stage. The networks are trained on V1 02 medium to minimize Eq. (58), which is the weighted MSE of exactly the orientation, position, and velocity errors reported in Table III; consequently, the V1 comparison in Tables II and III is a training-set comparison rather than an independent prediction. The paper's assertion that DeepUKF-VIN 'was never trained on either experiment' is contradicted by Section VI. The genuinely held-out V2 experiment is the only independent test, and it does not rescue the central claim: Table III shows DeepUKF-VIN orientation MSE 0.0080 versus 0.0026 for UKF-VIN, about a factor of three worse, while the overall loss is essentially tied (250 versus 251). Thus the claimed consistent superiority over all components and both scenarios is not established; for one of the two experiments it is a fitted input being presented as validation. The EKF-for-training substitution is a plausible transfer hypothesis rather than circularity, because V2 could in principle test it, but the V2 results provide mixed to negative evidence. Overall, the circularity is partial rather than total, located in the evaluation evidence chain rather than the filter mathematics.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The central claim rests on hand-tuned nominal covariance scales and loss weights, an unverified transfer from EKF training to UKF inference, and an underspecified landmark model. The V1 evaluation is trained data, so the only out-of-sample evidence (V2) fails to confirm the headline.

free parameters (5)
  • nominal standard deviations cbar_i (13 values) = not disclosed
    Baseline noise levels from traditional offline tuning (Eq. 50, Section VI); every learned scale multiplies these values.
  • deviation bound upsilon = not disclosed
    Controls how far learned covariances may move from nominal values in Eq. (50).
  • loss weights wq, wp, wv = wq=1000, wp=600, wv=100
    Hand-tuned to balance orientation, position, and velocity MSE in Eq. (58).
  • UKF tuning parameters lambda, alpha, beta = not disclosed
    Sigma point spread and weighting parameters in Eqs. (31) and (37).
  • loss transient cutoff = first 50 time steps excluded
    Loss is evaluated from the 51st data point onward, by design ignoring transient errors.
assumptions (4)
  • domain assumption IMU noise is zero-mean white Gaussian with diagonal covariance
    Eqs. (18)-(20); standard but not verified on the dataset.
  • domain assumption Landmark coordinates in the world frame l_w,i are known
    Measurement function Eq. (25) requires l_w; Section VI's triangulation description does not explain how l_w is obtained independently of the estimated pose.
  • ad hoc to paper Covariance model learned with an EKF transfers to the UKF
    Section VI: 'we hypothesize that the model can learn the optimal covariance matrices ... independently of the filter type used'; this premise is load-bearing for all UKF results.
  • domain assumption Ground truth from Vicon is accurate enough for training labels
    Training loss Eq. (58) uses q_k, p_k, v_k from EuRoC motion capture; errors are assumed negligible.

how reviews work

0 comments
Cite this review

Pith. "Pith review of DeepUKF-VIN: Adaptively-tuned Deep Unscented Kalman Filter for 3D Visual-Inertial Navigation based on IMU-Vision-Net." pith.science (2026). https://pith.science/paper/Z2HVVKZ2

@misc{pith2026250200575,
  author       = {Pith},
  title        = {Pith review of: DeepUKF-VIN: Adaptively-tuned Deep Unscented Kalman Filter for 3D Visual-Inertial Navigation based on IMU-Vision-Net},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/Z2HVVKZ2}},
  note         = {Machine review of arXiv:2502.00575}
}
read the original abstract

This paper addresses the challenge of estimating the orientation, position, and velocity of a vehicle operating in three-dimensional (3D) space with six degrees of freedom (6-DoF). A Deep Learning-based Adaptation Mechanism (DLAM) is proposed to adaptively tune the noise covariance matrices of Kalman-type filters for the Visual-Inertial Navigation (VIN) problem, leveraging IMU-Vision-Net. Subsequently, an adaptively tuned Deep Learning Unscented Kalman Filter for 3D VIN (DeepUKF-VIN) is introduced to utilize the proposed DLAM, thereby robustly estimating key navigation components, including orientation, position, and linear velocity. The proposed DeepUKF-VIN integrates data from onboard sensors, specifically an inertial measurement unit (IMU) and visual feature points extracted from a camera, and is applicable for GPS-denied navigation. Its quaternion-based design effectively captures navigation nonlinearities and avoids the singularities commonly encountered with Euler-angle-based filters. Implemented in discrete space, the DeepUKF-VIN facilitates practical filter deployment. The filter's performance is evaluated using real-world data collected from an IMU and a stereo camera at low sampling rates. The results demonstrate filter stability and rapid attenuation of estimation errors, highlighting its high estimation accuracy. Furthermore, comparative testing against the standard Unscented Kalman Filter (UKF) in two scenarios consistently shows superior performance across all navigation components, thereby validating the efficacy and robustness of the proposed DeepUKF-VIN. Keywords: Deep Learning, Unscented Kalman Filter, Adaptive tuning, Estimation, Navigation, Unmanned Aerial Vehicle, Sensor-fusion.

Figures

Figures reproduced from arXiv: 2502.00575 by the authors.

Figure 1
Figure 1. IMU-Net Architecture Schematics A. IMU-Net It is assumed that the covariance matrices Cηw,k and Cηa,k can be optimized for each batch by considering the input vector, which comprises the last db IMU measurements and practically it is a feasible and realizable condition. Recur￾rent deep learning frameworks, particularly Recurrent Neural Networks (RNNs) and their advanced variants, have proven effective in modeling se… view at source ↗
Figure 2
Figure 2. Vision-Net Architecture Schematics Filter Aggregate Predict Update Sensor Data Filter Adaption Mechanism Aggregate IMU Data IMU Net Vision Data Aggregate Predict Vision Net Update Current State Vector Estimate Sigma Points Construction Aggregate IMU Sigma Points Propagation Augmentation yes No If at end of Aggregate Vision Data Vision Noise Covariance Matrix Calculate Measurement Sigma Points And Its Statistics IMU … view at source ↗
Figure 3
Figure 3. Summary schematic architecture of quaternion-based DeepUKF-VIN. First, the Aggregate Predict step of the filter is [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figures from the paper (3 more)
Figure 5
Figure 5. Figure 5: Matched feature points between the left and right [PITH_FULL_IMAGE:figures/full_fig_p010_5.png]
Figure 6
Figure 6. Figure 6: Validation results of quaternion-based DeepUKF-VIN: The algorithm is evaluated using the V1 [PITH_FULL_IMAGE:figures/full_fig_p011_6.png]
Figure 7
Figure 7. Figure 7: Components of the orientation (left), position (mid [PITH_FULL_IMAGE:figures/full_fig_p011_7.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

48 extracted references · 38 canonical work pages

  1. [1]

    Advances in UA V Avionics Systems Architecture, Classification and Integration: A Comprehensive Review and Future Perspectives,

    H. A. Hashim, “Advances in UA V Avionics Systems Architecture, Classification and Integration: A Comprehensive Review and Future Perspectives,” Results in Engineering , vol. 25, p. 103786, 2025

  2. [2]

    Geometric Stochastic Filter with Guaranteed Performance for Autonomous Navigation based on IMU and Feature Sensor Fusion,

    H. A. Hashim, M. Abouheaf, and M. A. Abido, “Geometric Stochastic Filter with Guaranteed Performance for Autonomous Navigation based on IMU and Feature Sensor Fusion,” Control Engineering Practice, vol. 116, p. 104926, 2021

  3. [3]

    Mtrajplan- ner: A multiple-trajectory planning algorithm for autonomous underwa- ter vehicles,

    Y .-J. Gong, T. Huang, Y .-N. Ma, S.-W. Jeon, and J. Zhang, “Mtrajplan- ner: A multiple-trajectory planning algorithm for autonomous underwa- ter vehicles,” IEEE Transactions on Intelligent Transportation Systems , vol. 24, no. 4, pp. 3714–3727, 2023

  4. [4]

    Homecare robotic systems for healthcare 4.0: Visions and enabling technologies,

    G. Yang and et al., “Homecare robotic systems for healthcare 4.0: Visions and enabling technologies,” IEEE journal of biomedical and health informatics, vol. 24, no. 9, pp. 2535–2549, 2020

  5. [5]

    Efficient package delivery task assignment for truck and high capacity drone,

    X. Bai, Y . Ye, B. Zhang, and S. S. Ge, “Efficient package delivery task assignment for truck and high capacity drone,” IEEE Transactions on Intelligent Transportation Systems , vol. 24, no. 11, pp. 13 422–13 435, 2024

  6. [6]

    Fault-tolerant cooperative navigation of networked uav swarms for forest fire moni- toring,

    J. Hu, H. Niu, J. Carrasco, B. Lennox, and F. Arvin, “Fault-tolerant cooperative navigation of networked uav swarms for forest fire moni- toring,” Aerospace Science and Technology , vol. 123, p. 107494, 2022

  7. [7]

    Heterogeneity in ice-wedge per- mafrost degradation revealed across spatial scales,

    K. N. Braun and C. G. Andresen, “Heterogeneity in ice-wedge per- mafrost degradation revealed across spatial scales,” Remote Sensing of Environment, vol. 311, p. 114299, 2024

  8. [8]

    Augmented reality navigation for minimally invasive knee surgery using enhanced arthroscopy,

    F. Chen and wt al., “Augmented reality navigation for minimally invasive knee surgery using enhanced arthroscopy,” Computer Methods and Programs in Biomedicine , vol. 201, p. 105952, 2021

Show all 48 references
  1. [9]

    Multiparticle kalman filter for object localization in symmetric environments,

    R. Korkin, I. Oseledets, and A. Katrutsa, “Multiparticle kalman filter for object localization in symmetric environments,” Expert Systems with Applications, vol. 237, p. 121408, 2024

  2. [10]

    Magnetometer-based attitude determination extended kalman filter and optimization tech- niques,

    S. Wattanarungsan, T. Kuwahara, and S. Fujita, “Magnetometer-based attitude determination extended kalman filter and optimization tech- niques,” IEEE Transactions on Aerospace and Electronic Systems , vol. 59, no. 6, pp. 7993–8004, 2023

  3. [11]

    Pedestrian dead reckoning with wearable sensors: A systematic review,

    X. Hou and J. Bergmann, “Pedestrian dead reckoning with wearable sensors: A systematic review,” IEEE Sensors Journal, vol. 21, no. 1, pp. 143–152, 2021. 13

  4. [12]

    UWB Ranging and IMU Data Fusion: Overview and Nonlinear Stochastic Filter for Inertial Navigation,

    H. A. Hashim, A. E. Eltoukhy, and K. G. Vamvoudakis, “UWB Ranging and IMU Data Fusion: Overview and Nonlinear Stochastic Filter for Inertial Navigation,” IEEE Transactions on Intelligent Transportation Systems, vol. 25, no. 1, pp. 359–369, 2024

  5. [13]

    Airtag of the clones: Shenanigans with liberated item finders,

    T. M. Roth, F. Freyer, M. Hollick, and J. Classen, “Airtag of the clones: Shenanigans with liberated item finders,” 2022 IEEE Security and Privacy Workshops (SPW) , pp. 301–311, 2022

  6. [14]

    Exponentially Stable Observer-based Controller for VTOL-UA Vs without Velocity Measurements,

    H. A. Hashim, “Exponentially Stable Observer-based Controller for VTOL-UA Vs without Velocity Measurements,”International Journal of Control, vol. 96, no. 8, pp. 1946–1960, 2023

  7. [15]

    Method for registration of 3-d shapes,

    P. J. Besl and N. D. McKay, “Method for registration of 3-d shapes,” in Sensor fusion IV: control paradigms and data structures , vol. 1611. Spie, 1992, pp. 586–606

  8. [16]

    Point set registration: Coherent point drift,

    A. Myronenko and X. Song, “Point set registration: Coherent point drift,” IEEE transactions on pattern analysis and machine intelligence, vol. 32, no. 12, pp. 2262–2275, 2010

  9. [17]

    Robust inertial-aided underwa- ter localization based on imaging sonar keyframes,

    Y . Xu, R. Zheng, S. Zhang, and M. Liu, “Robust inertial-aided underwa- ter localization based on imaging sonar keyframes,” IEEE Transactions on Instrumentation and Measurement , vol. 71, pp. 1–12, 2022

  10. [18]

    A survey of lidar technology and its use in spacecraft relative navigation,

    J. A. Christian and S. Cryan, “A survey of lidar technology and its use in spacecraft relative navigation,” in AIAA Guidance, Navigation, and Control (GNC) Conference, 2013, p. 4641

  11. [19]

    GPS-denied Navigation: Attitude, Position, linear Ve- locity, and Gravity Estimation with Nonlinear Stochastic Observer,

    H. A. Hashim, “GPS-denied Navigation: Attitude, Position, linear Ve- locity, and Gravity Estimation with Nonlinear Stochastic Observer,” in 2021 American Control Conference (ACC) . IEEE, 2021, pp. 1146– 1151

  12. [20]

    Distinctive image features from scale-invariant keypoints,

    D. G. Lowe, “Distinctive image features from scale-invariant keypoints,” International journal of computer vision , vol. 60, pp. 91–110, 2004

  13. [21]

    Good features to track,

    J. Shi et al. , “Good features to track,” in 1994 Proceedings of IEEE conference on computer vision and pattern recognition . IEEE, 1994, pp. 593–600

  14. [22]

    Visual–inertial naviga- tion systems for aerial robotics: Sensor fusion and technology,

    F. Santoso, M. A. Garratt, and S. G. Anavatti, “Visual–inertial naviga- tion systems for aerial robotics: Sensor fusion and technology,” IEEE Transactions on Automation Science and Engineering , vol. 14, no. 1, pp. 260–275, 2016

  15. [23]

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

    A. I. Mourikis and S. I. Roumeliotis, “A multi-state constraint kalman filter for vision-aided inertial navigation,” in Proceedings 2007 IEEE international conference on robotics and automation . IEEE, 2007, pp. 3565–3572

  16. [24]

    Robust stereo visual inertial odometry for fast autonomous flight,

    K. Sun and et al., “Robust stereo visual inertial odometry for fast autonomous flight,” IEEE Robotics and Automation Letters, vol. 3, no. 2, pp. 965–972, 2018

  17. [25]

    Kalman filter for mobile- robot attitude estimation: Novel optimized and adaptive solutions,

    A. Odry, R. Fuller, I. J. Rudas, and P. Odry, “Kalman filter for mobile- robot attitude estimation: Novel optimized and adaptive solutions,” Mechanical systems and signal processing, vol. 110, pp. 569–589, 2018

  18. [26]

    Quaternion-based Unscented Kalman Filter for 6-DoF Vision-based Inertial Navigation in GPS-denied Regions,

    K. Ghanizadegan and H. A. Hashim, “Quaternion-based Unscented Kalman Filter for 6-DoF Vision-based Inertial Navigation in GPS-denied Regions,” IEEE Transactions on Instrumentation and Measurement , vol. 74, no. 1, pp. 1–13, 2025

  19. [27]

    Nonlinear Stochastic At- titude Filters on the Special Orthogonal Group 3: Ito and Stratonovich,

    H. A. Hashim, L. J. Brown, and K. McIsaac, “Nonlinear Stochastic At- titude Filters on the Special Orthogonal Group 3: Ito and Stratonovich,” IEEE Transactions on Systems, Man, and Cybernetics: Systems , vol. 49, no. 9, pp. 1853–1865, 2019

  20. [28]

    Systematic Convergence of Nonlinear Stochastic Esti- mators on the Special Orthogonal Group SO(3),

    H. A. Hashim, “Systematic Convergence of Nonlinear Stochastic Esti- mators on the Special Orthogonal Group SO(3),” International Journal of Robust and Nonlinear Control , vol. 30, no. 10, pp. 3848–3870, 2020

  21. [29]

    Fusing inertial sensor data in an extended kalman filter for 3d camera tracking,

    A. T. Erdem and A. O. Ercan, “Fusing inertial sensor data in an extended kalman filter for 3d camera tracking,” IEEE Transactions on Image Processing, vol. 24, no. 2, pp. 538–548, 2014

  22. [30]

    The unscented kalman filter,

    E. A. Wan and R. Van Der Merwe, “The unscented kalman filter,” Kalman filtering and neural networks , pp. 221–280, 2001

  23. [31]

    On noise covariance estimation for kalman filter-based damage localization,

    S. Wernitz, E. Chatzi, B. Hofmeister, M. Wolniak, W. Shen, and R. Rolfes, “On noise covariance estimation for kalman filter-based damage localization,” Mechanical Systems and Signal Processing , vol. 170, p. 108808, 2022

  24. [32]

    Ai-imu dead-reckoning,

    M. Brossard, A. Barrau, and S. Bonnabel, “Ai-imu dead-reckoning,” IEEE Transactions on Intelligent Vehicles , vol. 5, no. 4, pp. 585–595, 2020

  25. [33]

    Imu dead-reckoning localization with rnn-iekf algorithm,

    H. Zhou and et al., “Imu dead-reckoning localization with rnn-iekf algorithm,” in 2022 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) . IEEE, 2022, pp. 11 382–11 387

  26. [34]

    Learning vehicle trajectory uncertainty,

    B. Or and I. Klein, “Learning vehicle trajectory uncertainty,” Engineer- ing Applications of Artificial Intelligence , vol. 122, p. 106101, 2023

  27. [35]

    Multi-level deep learning kalman filter,

    S. Yan, Y . Liang, and B. Wang, “Multi-level deep learning kalman filter,” in 2023 International Conference on Advanced Robotics and Mechatronics (ICARM). IEEE, 2023, pp. 1113–1118

  28. [36]

    Special Orthogonal Group SO(3), Euler Angles, Angle- axis, Rodriguez Vector and Unit-quaternion: Overview, Mapping and Challenges,

    H. A. Hashim, “Special Orthogonal Group SO(3), Euler Angles, Angle- axis, Rodriguez Vector and Unit-quaternion: Overview, Mapping and Challenges,” arXiv preprint arXiv:1909.06669 , 2019

  29. [37]

    A review of irregular time series data handling with gated recurrent neural networks,

    P. B. Weerakody, K. W. Wong, G. Wang, and W. Ela, “A review of irregular time series data handling with gated recurrent neural networks,” Neurocomputing, vol. 441, pp. 161–178, 2021

  30. [38]

    Unicornn: A recurrent model for learning very long time dependencies,

    T. K. Rusch and S. Mishra, “Unicornn: A recurrent model for learning very long time dependencies,” in International Conference on Machine Learning. PMLR, 2021, pp. 9168–9178

  31. [39]

    Learning long-term dependencies in irregularly-sampled time series,

    M. Lechner and R. M. Hasani, “Learning long-term dependencies in irregularly-sampled time series,” ArXiv, vol. abs/2006.04418,

  32. [40]

    Remaining useful life prognosis based on ensemble long short-term memory neural network,

    Y . Cheng, J. Wu, H. Zhu, S. W. Or, and X. Shao, “Remaining useful life prognosis based on ensemble long short-term memory neural network,” IEEE Transactions on Instrumentation and Measurement , vol. 70, pp. 1–12, 2020

  33. [41]

    Gate-variants of gated recurrent unit (gru) neural networks,

    R. Dey and F. M. Salem, “Gate-variants of gated recurrent unit (gru) neural networks,” 2017 IEEE 60th International Midwest Symposium on Circuits and Systems (MWSCAS) , pp. 1597–1600, 2017. [Online]. Available: https://api.semanticscholar.org/CorpusID:8492900

  34. [42]

    Performance evaluation of deep neural networks applied to speech recognition: Rnn, lstm and gru,

    A. N. Shewalkar, D. Nyavanandi, and S. A. Ludwig, “Performance evaluation of deep neural networks applied to speech recognition: Rnn, lstm and gru,” Journal of Artificial Intelligence and Soft Computing Research, vol. 9, pp. 235 – 245, 2019

  35. [43]

    The EuRoC micro aerial vehicle datasets,

    M. Burri and et al., “The EuRoC micro aerial vehicle datasets,” The International Journal of Robotics Research , vol. 35, no. 10, pp. 1157– 1163, 2016

  36. [44]

    Adam: A method for stochastic optimization,

    D. P. Kingma, “Adam: A method for stochastic optimization,” arXiv preprint arXiv:1412.6980, 2014

  37. [45]

    Towards understanding con- vergence and generalization of adamw,

    P. Zhou, X. Xie, Z. Lin, and S. Yan, “Towards understanding con- vergence and generalization of adamw,” IEEE Transactions on Pattern Analysis and Machine Intelligence , 2024

  38. [46]

    Accelerating 3d deep learning with pytorch3d,

    N. Ravi, J. Reizenstein, D. Novotny, T. Gordon, W.-Y . Lo, J. Johnson, and G. Gkioxari, “Accelerating 3d deep learning with pytorch3d,” arXiv:2007.08501, 2020

  39. [47]

    Hartley and A

    R. Hartley and A. Zisserman, Multiple view geometry in computer vision. Cambridge university press, 2003

  40. [2020]

    Available: https://api.semanticscholar.org/CorpusID: 219530825

    [Online]. Available: https://api.semanticscholar.org/CorpusID: 219530825

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.