Pith. sign in

REVIEW 3 major objections 6 minor 45 references

Neural Augmented Kalman Filters for Road Network assisted GNSS positioning

T0 review · 3 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read A Temporal Graph Neural Network learns which road a vehicle is on and how much to trust that choice, cutting 95th-percentile urban GNSS positioning error by 29% compared with a GNSS-only Kalman filter.

desk verdict A well-executed engineering contribution with a plausible 29% tail-error reduction on one dataset; the main uncertainty is how much of that gain is an artifact of the heuristic teacher labels. read the letter →

arxiv 2507.00654 v1 pith:TA7G3ZXN submitted 2025-07-01 cs.LG cs.SYeess.SPeess.SY

Hans van Gorp , Davide Belli , Amir Jalalirad , Bence Major This is my paper
classification cs.LGcs.SYeess.SPeess.SY
keywords GNSSpositioningroadnetworkassistancetemporalgraphneuralnetworksKalmanfiltersurbanmultipathmitigationuncertaintypredictionmap-aidednavigation
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

GNSS positioning degrades sharply in dense cities because signals bounce off buildings, yet services like ride-hailing and lane-level navigation need 2-10 m accuracy. The paper tries to show that a Temporal Graph Neural Network (a graph network with memory over time) can learn the two choices a road-assisted Kalman filter (a standard recursive estimator that fuses noisy measurements) needs to make: which road segment the vehicle is on, and how much to trust that choice as a per-frame variance. Given real GNSS pseudoranges and an open road map, the network's selected segment is inserted into the Kalman filter as an extra measurement, and the whole pipeline is trained end-to-end on final position error. The authors report that this cuts 95th-percentile horizontal error from 77.23 m to 55.02 m, a 29% reduction over a GNSS-only Kalman filter, and also beats a Viterbi map-matching baseline (a dynamic-programming sequence decoder), on held-out urban drives from four cities. If the claim holds, meaningful urban accuracy gains are available with no new sensors, just the map as prior knowledge.

What carries the argument

The central object is the road-network observation model that is inserted into the Kalman filter's measurement update. The road network is converted to a graph in which nodes are road segments and edges are intersections; for a selected segment $r^*$, the model constructs a two-dimensional Gaussian observation $z$ with covariance $V = \mathrm{diag}(\sigma^2_\parallel, \sigma^2_\perp)$, expressed in coordinates parallel and perpendicular to the road, and fuses it through the standard update $K = P^+ H^T (H P^+ H^T + V)^{-1}$. The TGNN supplies both inputs to that update: a softmax head over segment probabilities (with the cross-entropy target provided by an offline bidirectional Viterbi oracle) and an exponential head predicting $\sigma^2_\parallel$ and $\sigma^2_\perp$. Because the KF is differentiable, the covariance head is trained end-to-end on the final position error, which is what allows the filter to down-weight the road measurement in ambiguous geometries rather than using a fixed tuned variance.

What would settle it

Take a set of urban GNSS drives with ground-truth road-segment labels. Compute where the bidirectional Viterbi oracle disagrees with ground truth, then check whether the TGNN's selected segments match the oracle's wrong choices on those cases; if the TGNN follows the oracle into error and the 95th-percentile improvement vanishes when the training labels are replaced by ground-truth segments, the reported 29% reduction is a property of the oracle's error pattern rather than a general learned road-selection ability.

Watch

Extended reading notes

Core claim

The paper claims that a road network can be turned into a trainable measurement source for a Kalman filter. A Temporal Graph Neural Network scores every candidate road segment in a field of view around the current position; the best-scoring segment is turned into a Gaussian observation aligned with the road, with variances predicted by a second output head. Both the segment probability and the variances are trained jointly, with the segment loss supervised by a bidirectional Viterbi oracle and the covariance loss backpropagated through the differentiable Kalman update against the ground-truth trajectory. On real drives from four cities evaluated by leave-one-out cross-validation, this yields a 95th-percentile horizontal error of 55.02 m versus 77.23 m for a GNSS-only KF (28.8% lower) and 68.27 m for a KF fused with Viterbi road selection. The authors state this is the first deep-learning method to jointly exploit road-network data and GNSS measurements to determine the user's position on Earth.

Load-bearing premise

The learned road selector is trained to imitate a bidirectional Viterbi oracle with hand-set parameters that the paper admits occasionally fails, so if the oracle's mistakes are systematic the neural network inherits them and the 29% tail gain partly reflects the oracle's error pattern.

Editorial extensions

If this is right

  • Road-assisted positioning can be trained end-to-end and run in real time: the TGNN has under 50k parameters and costs about 1.7 MFLOPs per step, so the accuracy gain does not require a separate offline or heavy computation stage.
  • The per-step predicted covariance lets the filter trust the map when it helps and ignore it when the road structure is ambiguous, which is the mechanism behind the tail-error reduction over fixed-variance grid-search tuning.
  • The learned selector can be trained offline with the benefit of future information (the oracle) and then deployed online without future data, closing the gap between offline smoothing and real-time filtering.
  • Using only open road-network data and the receiver's own pseudoranges, the approach improves the error regime that matters for lane and side-of-street estimation, so it is a candidate low-cost complement to 3D-city-model assistance.

Reading between the lines

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

  • The same recipe may generalize beyond roads: any discrete spatial prior that can be turned into a graph—indoor floor plans, railway topology, pedestrian walkways—could be injected into a KF as a learned Gaussian pseudo-measurement with a per-timestep covariance head.
  • The reported gain is concentrated in the error tail: at the median the TGNN (8.74 m) is actually slightly worse than the Viterbi baseline (8.02 m), so a fair reading is that the method trades a small median loss for a large reduction in rare large errors rather than uniformly improving accuracy.
  • A testable extension the authors do not run: train the same architecture against ground-truth road labels instead of the bidirectional Viterbi oracle; if the 95th-percentile error drops further, the oracle's occasional failures are a bottleneck, and if it rises, the oracle's smoothing is doing essential work.
  • The qualitative failure cases suggest the next bottleneck is map fidelity: snapping to road centerlines on multi-lane parallel roads produces consistent offsets whose magnitude scales with road width, so lane-level maps or an IMU heading constraint would be the natural follow-up rather than a larger network.
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

3 major / 6 minor

Summary. The paper proposes augmenting a GNSS Kalman filter with a temporal graph neural network (TGNN) that, at each timestep, selects a road segment from OpenStreetMap data and predicts a diagonal observation covariance for a road-network measurement update. The selection head is trained with cross-entropy against labels from a bidirectional Viterbi 'Oracle', and the covariance head is trained with an MSE loss on the final KF position through a differentiable Kalman filter. On a real-world GNSS dataset across four cities, the method reduces the 95th-percentile horizontal error from 77.23 m (GNSS-only KF) to 55.02 m, a 28.8% reduction, and also improves over a KF+Viterbi baseline (68.27 m). Ablations show that replacing either learned component with a non-learned counterpart degrades performance.

Significance. If the result holds, this is a useful and reasonably novel contribution: it is, to my knowledge, the first learned road-network augmentation of a GNSS KF that outputs both road selection and observation covariance; the model is lightweight (<50k parameters, 1.7 MFLOPs); and the paper includes careful ablations (Tables 2–4) and a sensitivity analysis (Fig. 5). The headline 28.8% improvement is internally consistent with Table 1, and the fully differentiable KF training is a sound approach. The main caveats are that the evidence is a single dataset with no released code or data, the teacher (Oracle) labels are heuristic and unquantified, and the training/inference input distributions differ because the KF state seen by the TGNN at test time has already been road-updated. These caveats do not invalidate the core idea, but they limit how strongly the 29% claim can be generalized.

major comments (3)
  1. [§3.3, Eq. (14); §4.2] The road-selection head is trained by cross-entropy against the bidirectional Viterbi Oracle, whose emission weighting β=0.01 and k-hop transition k=2 are hand-set (Eqs. 5–6, Appendix C). Section 4.2 acknowledges 'occasional failures' of the Oracle, but the paper never quantifies Oracle accuracy or shows how its errors correlate with the 95th-percentile error tail that the headline metric targets. If the Oracle systematically selects a wrong parallel road in large-error urban events, CE training would teach the TGNN to reproduce that bias, and part of the measured gain over the GNSS-only KF could be an artifact of the teacher rather than of correct road selection. Please report road-selection accuracy (Oracle, Viterbi, TGNN) against a ground-truth road reference, stratified by KF error magnitude; also report sensitivity of the Table 1 result to β and k, and/or train the selector with forward-only Viterbi labels as a control.
  2. [§3.3 and Figure 1] The training and inference regimes for the TGNN differ in a way the paper does not address. The Oracle labels are generated from the GNSS-only KF trajectory (no road updates), while at inference the KF state fed to the TGNN has already been updated by road measurements at previous timesteps (Figure 1), so the input distribution at test time is not the one seen during label generation. This closed-loop shift could degrade selection quality in a way that is not captured by the ablations. Please specify whether training inputs are the same closed-loop states or the GNSS-only states, and, if the latter, quantify the effect of using GNSS-only states at inference.
  3. [§3.4, Eq. (16)] The paper states that the covariance head is optimized 'end-to-end to minimize the positioning error,' but the road selection uses a hard argmax (Eq. 1 with J_TGNN = -Pφ), which is non-differentiable. The text should clarify that the selection head is trained solely by L_CE and that the MSE loss trains only the covariance head with the selected road treated as fixed; if a soft or straight-through variant is used, it should be described. As written, the claim of end-to-end training is stronger than what the equations show.
minor comments (6)
  1. [§4.2] The phrase 'GNN-only KF' should be 'GNSS-only KF'.
  2. [§4.2] The statement 'with less than 1 m error increase at the 50th percentile' is ambiguous: relative to the GNSS-only KF the median error decreases, while relative to KF+Viterbi it increases by 0.72 m; please rephrase.
  3. [Eq. (5)] The phrase 'β is a hyper-parameter to balance between the two cost functions' is misleading because β multiplies both Jpos and Jθ; it scales the combined cost rather than balancing the two terms. Consider writing p(x|ri)=max(1 - (β Jpos + Jθ)/2, ε) if a true balance is intended.
  4. [Appendix B.1] There are typos in the feature list: 'V ehicle Headingand Position Uncertainty' and 'desribes' should be corrected.
  5. [Table 1] The Oracle row uses zero covariance; the text should state explicitly that this is a hard-constraint oracle and not a realistic learned upper bound.
  6. [General] The paper does not release code or data; a reproducibility statement describing the exact split, feature preprocessing, and road-network extraction would strengthen the manuscript.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: TGNN predictions are evaluated on held-out folds against ground-truth position, and the bidirectional Viterbi Oracle is a training-label source, not the reported predictive claim.

full rationale

The claimed result is a 29% reduction in 95th-percentile horizontal error when augmenting a KF with a TGNN. The derivation chain is: the TGNN selects a road segment (trained by cross-entropy against a bidirectional Viterbi Oracle, Eq. 14) and predicts a road-measurement covariance (trained by MSE between the differentiable KF output and ground truth, Eqs. 15-17). The final metric is the held-out horizontal error, not the Oracle agreement or the training loss. The data is split into three regionally non-overlapping folds with leave-one-out cross-validation (Section 4.1), so the reported 55.02 m HE@95 is not obtained by fitting the test set. The covariance head is optimized on a training split, which is standard supervised fitting, not a circular prediction. The Oracle itself is only a teacher for road-selection labels and is acknowledged to have occasional failures (Section 4.2); those failures are a label-quality concern, not a circularity. The ablations in Tables 2 and 3 show that both the learned selector and the learned covariance contribute beyond non-learned alternatives, consistent with a genuine mechanism rather than a by-construction equivalence. The only self-citations are the GNSS dataset and KF initialization from Jalalirad et al. (2023), which are real external data artifacts with ground-truth positions and are not load-bearing assumptions that force the target result. No equation in the paper reduces to its own input, and no fitted parameter is renamed as a prediction.

Assumptions & free parameters 6 free parameters · 4 assumptions · 0 invented entities

No new physical entities or forces are introduced; the road measurement is a modeling construct, not a new entity. The main ledger entries are training hyperparameters and assumptions about map and Oracle quality.

free parameters (6)
  • TGNN network weights = ~50,000 parameters
    Trained end-to-end by CE + MSE (Equations 14 to 17); the central claim depends on this learned mapping.
  • beta (emission cost weight) = 0.01
    Balances position and heading cost in the Viterbi emission PDF (Equation 5); chosen in preliminary experiments and fixed for all experiments (Appendix C). Used by the Oracle and as a feature source for the TGNN.
  • k (Viterbi k-hop transition) = 2
    Transition probability in Equation 6 considers a k-hop neighborhood to handle high-speed traversal; set to 2 (Appendix C).
  • field of view = 50 m
    Radius around the position estimate defining candidate road segments; selected based on the ablation in Section 4.3.
  • lambda (loss weight) = 0.01
    Balances cross-entropy and MSE losses in Equation 17; set so 100 m of position distance becomes unit cost (Appendix C).
  • baseline road covariances (sigma_perp^2, sigma_par^2) = grid-searched over 0-10 and 0-10/100-1000/inf
    Used for Instant and Viterbi baselines; selected on the training set by lowest HE@95 (Appendix C). Not used by the proposed TGNN method.
assumptions (4)
  • domain assumption GNSS pseudo-ranges and corrections in the dataset are accurate enough to initialize and maintain the KF state.
    Section 4.1 and Appendix B describe corrections (Klobuchar, Saastamoinen, Sagnac) but the paper does not analyze residual errors.
  • domain assumption OpenStreetMap road graph, split into segments up to 25 m, correctly represents connectivity and drivable roads in the test areas.
    Used to build graph G = (R, A) for the GCN and Viterbi; map-vehicle mismatches are acknowledged in Section 4.2 as one reason the Oracle error cannot reach zero.
  • domain assumption Road network observations can be modeled as uni-modal Gaussian measurements in the KF update.
    Equations 3 and 18 to 22 assume a Gaussian observation with covariance V; the authors state this assumption is not natural for road data because nearby roads create multi-modality, and the selector is meant to enforce uni-modality.
  • domain assumption The bidirectional Viterbi Oracle provides correct-enough road segment labels for supervised training.
    Section 3.3 trains with labels r*_oracle; Section 4.2 admits occasional Oracle failures, so label quality is load-bearing.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Neural Augmented Kalman Filters for Road Network assisted GNSS positioning." pith.science (2026). https://pith.science/paper/TA7G3ZXN

@misc{pith2026250700654,
  author       = {Pith},
  title        = {Pith review of: Neural Augmented Kalman Filters for Road Network assisted GNSS positioning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/TA7G3ZXN}},
  note         = {Machine review of arXiv:2507.00654}
}
read the original abstract

The Global Navigation Satellite System (GNSS) provides critical positioning information globally, but its accuracy in dense urban environments is often compromised by multipath and non-line-of-sight errors. Road network data can be used to reduce the impact of these errors and enhance the accuracy of a positioning system. Previous works employing road network data are either limited to offline applications, or rely on Kalman Filter (KF) heuristics with little flexibility and robustness. We instead propose training a Temporal Graph Neural Network (TGNN) to integrate road network information into a KF. The TGNN is designed to predict the correct road segment and its associated uncertainty to be used in the measurement update step of the KF. We validate our approach with real-world GNSS data and open-source road networks, observing a 29% decrease in positioning error for challenging scenarios compared to a GNSS-only KF. To the best of our knowledge, ours is the first deep learning-based approach jointly employing road network data and GNSS measurements to determine the user position on Earth.

Figures

Figures reproduced from arXiv: 2507.00654 by the authors.

Figure 1
Figure 1. In order to leverage road network data in a GNSS-based KF, a Gaussian observation (mean z, covariance Q) needs to be constructed by selecting the correct road segment. In our proposed method, this road selection is performed by a Temporal Graph Convolutional Network which also predicts the appropriate covari￾ance (left). The position and uncertainty predicted by the KF can be improved using this additional data sour… view at source ↗
Figure 2
Figure 2. Overview of different road selection algorithms. In (a) we show an example user trajectory estimated from noisy GNSS measurements, with past (x0, x1), current (x2), and the future (x3) locations. The road network has 3 segments including two parallel roads r0 and r1, with the true user trajectory following r1 and r2, marked in green. We compare three road selection methods, highlighting in green correct predictions … view at source ↗
Figure 3
Figure 3. Architecture of the proposed TGNN. along the road. The information about one-way road di￾rectionality, if available, will be encoded in the transition probabilities. Finally, the emission PDF is calculated using a weighted average of the heading and positions costs, as: p(x|ri) = max 1 − βJpos(ri) + Jθ(ri) 2 , ϵ! , (5) where β is a hyper-parameter to balance between the two cost functions, and ϵ = 0.01 ensures that … view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: Horizontal Error CDF for different methods. 4.2. Quantitative Results We report the end-to-end positioning error averaged over folds for different positioning algorithms in [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 6
Figure 6. Figure 6: Qualitative results for the proposed TGNN model compared against the Viterbi baseline, zoomed in to interesting parts of the drive. Full images can be found in supplementary material Appendix E . only the Viterbi prior input features, which describe the prob￾ability of…
Figure 7
Figure 7. Figure 7: Architecture for the GNN and MLP ablations. We provide diagrams describing the neural network for the ablations of the TGNN architecture. The GNN model ( [PITH_FULL_IMAGE:figures/full_fig_p014_7.png]
Figure 8
Figure 8. Figure 8: Full drive where TGNN outperforms GNSS-only. -3000 -2000 -1000 0 1000 2000 3000 East [m] -1500 -1000 -500 0 500 1000 1500 North [m] KF + TGNN KF Ground Truth [PITH_FULL_IMAGE:figures/full_fig_p015_8.png]
Figure 9
Figure 9. Figure 9: Full drive where TGNN underperforms GNSS-only. 15 [PITH_FULL_IMAGE:figures/full_fig_p015_9.png]
Figure 10
Figure 10. Figure 10: Full drive where TGNN outperforms Viterbi. -800 -600 -400 -200 0 200 400 600 800 East [m] -400 -300 -200 -100 0 100 200 300 400 North [m] KF + TGNN KF + Viterbi Ground Truth [PITH_FULL_IMAGE:figures/full_fig_p016_10.png]
Figure 11
Figure 11. Figure 11: Full drive where TGNN underperforms Viterbi. 16 [PITH_FULL_IMAGE:figures/full_fig_p016_11.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

45 extracted references · 38 canonical work pages

  1. [1]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION format.date year duplicate empty "emp...

  2. [2]

    M., Hilal, A

    Atia, M. M., Hilal, A. R., Stellings, C., Hartwell, E., Toonstra, J., Miners, W. B., and Basir, O. A. A low-cost lane-determination system using gnss/imu fusion and hmm-based multistage map matching. IEEE Transactions on Intelligent Transportation Systems, 18 0 (11): 0 3027--3037, 2017

    work page 2017

  3. [3]

    A., Anthes, R

    Bevis, M., Businger, S., Chiswell, S., Herring, T. A., Anthes, R. A., Rocken, C., and Ware, R. H. Gps meteorology: Mapping zenith wet delays onto precipitable water. Journal of Applied Meteorology (1988-2005), pp.\ 379--386, 1994

    work page 1988

  4. [4]

    S., and Ganesh, L

    Bidikar, B., Rao, G. S., and Ganesh, L. Sagnac effect and set error based pseudorange modeling for gps applications. Procedia Computer Science, 87: 0 172--177, 2016

    work page 2016

  5. [5]

    Osmnx: A python package to work with graph-theoretic openstreetmap street networks

    Boeing, G. Osmnx: A python package to work with graph-theoretic openstreetmap street networks. Journal of Open Source Software, 2 0 (12): 0 215, 2017. doi:10.21105/joss.00215. URL https://doi.org/10.21105/joss.00215

    work page doi:10.21105/joss.00215 2017

  6. [6]

    Spectral temporal graph neural network for multivariate time-series forecasting

    Cao, D., Wang, Y., Duan, J., Zhang, C., Zhu, X., Huang, C., Tong, Y., Xu, B., Bai, J., Tong, J., et al. Spectral temporal graph neural network for multivariate time-series forecasting. Advances in neural information processing systems, 33: 0 17766--17778, 2020

    work page 2020

  7. [7]

    El Najjar, M. E. and Bonnifait, P. A road-matching method for precise vehicle localization using belief theory and kalman filtering. Autonomous Robots, 19: 0 173--191, 2005

    work page 2005

  8. [8]

    Sigmoid-weighted linear units for neural network function approximation in reinforcement learning

    Elfwing, S., Uchibe, E., and Doya, K. Sigmoid-weighted linear units for neural network function approximation in reinforcement learning. Neural networks, 107: 0 3--11, 2018

    2018

Show all 45 references
  1. [9]

    EO and GNSS Market Report

    EUSPA. EO and GNSS Market Report. Publications Office of the European Union, 2024. doi:10.2878/73092

    2024 doi

  2. [10]

    Deepmm: Deep learning based map matching with data augmentation

    Feng, J., Li, Y., Zhao, K., Xu, Z., Xia, T., Zhang, J., and Jin, D. Deepmm: Deep learning based map matching with data augmentation. IEEE Transactions on Mobile Computing, 21 0 (7): 0 2372--2384, 2020

    2020

  3. [11]

    Rl-akf: An adaptive kalman filter navigation algorithm based on reinforcement learning for ground vehicles

    Gao, X., Luo, H., Ning, B., Zhao, F., Bao, L., Gong, Y., Xiao, Y., and Jiang, J. Rl-akf: An adaptive kalman filter navigation algorithm based on reinforcement learning for ground vehicles. Remote Sensing, 12 0 (11): 0 1704, 2020

    2020

  4. [12]

    Y., Dauwels, J., Mitrovic, N., Asif, M

    Goh, C. Y., Dauwels, J., Mitrovic, N., Asif, M. T., Oran, A., and Jaillet, P. Online map-matching based on hidden markov model for real-time traffic sensing applications. In 2012 15th International IEEE Conference on Intelligent Transportation Systems, pp.\ 776--781. IEEE, 2012

    2012

  5. [13]

    and Tu, W

    Guo, C. and Tu, W. A novel self-learning gnss/ins integrated navigation method. In Proceedings of the 34th International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GNSS+ 2021), pp.\ 168--179, 2021

    2021

  6. [14]

    Precise positioning with machine learning based kalman filter using gnss/imu measurements from android smartphone

    Han, K., Lee, S., Song, Y.-J., Lee, H.-B., Park, D.-H., and Won, J.-H. Precise positioning with machine learning based kalman filter using gnss/imu measurements from android smartphone. In Proceedings of the 34th International Technical Meeting of the Satellite Division of The...

    2021

  7. [15]

    Long short-term memory

    Hochreiter, S. Long short-term memory. Neural Computation MIT-Press, 1997

    1997

  8. [16]

    Amm: an adaptive online map matching algorithm

    Hu, H., Qian, S., Ouyang, J., Cao, J., Han, H., Wang, J., and Chen, Y. Amm: an adaptive online map matching algorithm. IEEE Transactions on Intelligent Transportation Systems, 24 0 (5): 0 5039--5051, 2023

    2023

  9. [17]

    Batch normalization: Accelerating deep network training by reducing internal covariate shift

    Ioffe, S. Batch normalization: Accelerating deep network training by reducing internal covariate shift. arXiv preprint arXiv:1502.03167, 2015

    2015 arXiv

  10. [18]

    Jagadeesh, G. R. and Srikanthan, T. Online map-matching of noisy and sparse location data with hidden markov and route choice models. IEEE Transactions on Intelligent Transportation Systems, 18 0 (9): 0 2423--2434, 2017

    2017

  11. [19]

    Gnss positioning using cost function regulated multilateration and graph neural networks

    Jalalirad, A., Belli, D., Major, B., Jee, S., Shah, H., and Morrison, W. Gnss positioning using cost function regulated multilateration and graph neural networks. In Proceedings of the 36th International Technical Meeting of the Satellite Division of The Institute of Navigatio...

    2023

  12. [20]

    V., Gupta, S., Shetty, A., and Gao, G

    Kanhere, A. V., Gupta, S., Shetty, A., and Gao, G. Improving gnss positioning using neural-network-based corrections. NAVIGATION: Journal of the Institute of Navigation, 69 0 (4), 2022

    2022

  13. [21]

    Kingma, D. P. and Ba, J. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980, 2014

    2014 arXiv

  14. [22]

    Kipf, T. N. and Welling, M. Semi-supervised classification with graph convolutional networks. arXiv preprint arXiv:1609.02907, 2016

    2016 arXiv

  15. [23]

    Ionospheric time-delay algorithm for single-frequency gps users

    Klobuchar, J. Ionospheric time-delay algorithm for single-frequency gps users. IEEE transactions on aerospace and electronic systems, 23 0 (3): 0 325--331, 1987

    1987

  16. [24]

    Deep learning based kalman filter for gnss/ins integration: Neural network architecture and feature selection

    Li, S., Mikhaylov, M., Mikhaylov, N., and Pany, T. Deep learning based kalman filter for gnss/ins integration: Neural network architecture and feature selection. In 2023 International Conference on Localization and GNSS (ICL-GNSS), pp.\ 1--7. IEEE, 2023

    2023

  17. [25]

    A novel map matching method based on improved hidden markov and conditional random fields model

    Li, W., Chen, Y., Wang, S., Li, H., and Fan, Q. A novel map matching method based on improved hidden markov and conditional random fields model. International Journal of Digital Earth, 17 0 (1): 0 2328366, 2024

    2024

  18. [26]

    and Gao, G

    Mohanty, A. and Gao, G. Tightly coupled graph neural network and kalman filter for smartphone positioning. NAVIGATION: Journal of the Institute of Navigation, 71 0 (4), 2024

    2024

  19. [27]

    Neural city maps for gnss nlos prediction

    Neamati, D., Gupta, S., Partha, M., and Gao, G. Neural city maps for gnss nlos prediction. In Proceedings of the 36th International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GNSS+ 2023), pp.\ 2073--2087, 2023

    2023

  20. [28]

    Planet dump retrieved from https://planet.osm.org

    OpenStreetMap contributors . Planet dump retrieved from https://planet.osm.org . https://www.openstreetmap.org , 2017

    2017

  21. [29]

    and Washington, S

    Quddus, M. and Washington, S. Shortest path and vehicle trajectory aided map-matching for low frequency gps data. Transportation Research Part C: Emerging Technologies, 55: 0 328--339, 2015

    2015

  22. [30]

    L., Van Sloun, R

    Revach, G., Shlezinger, N., Ni, X., Escoriza, A. L., Van Sloun, R. J., and Eldar, Y. C. Kalmannet: Neural network aided kalman filtering for partially known dynamics. IEEE Transactions on Signal Processing, 70: 0 1532--1547, 2022

    2022

  23. [31]

    Temporal graph networks for deep learning on dynamic graphs

    Rossi, E., Chamberlain, B., Frasca, F., Eynard, D., Monti, F., and Bronstein, M. Temporal graph networks for deep learning on dynamic graphs. arXiv preprint arXiv:2006.10637, 2020

    2006 arXiv

  24. [32]

    C., Hagenbuchner, M., and Monfardini, G

    Scarselli, F., Gori, M., Tsoi, A. C., Hagenbuchner, M., and Monfardini, G. The graph neural network model. IEEE transactions on neural networks, 20 0 (1): 0 61--80, 2008

    2008

  25. [33]

    Shlezinger, N., Revach, G., Ghosh, A., Chatterjee, S., Tang, S., Imbiriba, T., Dunik, J., Straka, O., Closas, P., and Eldar, Y. C. Ai-aided kalman filters. arXiv preprint arXiv:2410.12289, 2024

    2024 arXiv

  26. [34]

    Siemuri, A., Selvan, K., Kuusniemi, H., V \"a lisuo, P., and Elmusrati, M. S. Improving precision gnss positioning and navigation accuracy on smartphones using machine learning. In Proceedings of the 34th International Technical Meeting of the Satellite Division of The Institu...

    2021

  27. [35]

    and Handel, P

    Skog, I. and Handel, P. In-car positioning and navigation technologies—a survey. IEEE Transactions on Intelligent Transportation Systems, 10 0 (1): 0 4--21, 2009

    2009

  28. [36]

    Song, H. Y. and Lee, J. H. A map matching algorithm based on modified hidden markov model considering time series dependency over larger time span. Heliyon, 9 0 (11), 2023

    2023

  29. [37]

    and Amano, Y

    Suzuki, T. and Amano, Y. Nlos multipath classification of gnss signal correlation output using machine learning. Sensors, 21 0 (7): 0 2503, 2021

    2021

  30. [38]

    A gru and akf-based hybrid algorithm for improving ins/gnss navigation accuracy during gnss outage

    Tang, Y., Jiang, J., Liu, J., Yan, P., Tao, Y., and Liu, J. A gru and akf-based hybrid algorithm for improving ins/gnss navigation accuracy during gnss outage. Remote Sensing, 14 0 (3): 0 752, 2022

    2022

  31. [39]

    End game for urban gnss: Google’s use of 3d building models

    van Diggelen, F. End game for urban gnss: Google’s use of 3d building models. Inside GNSS, 2021

    2021

  32. [40]

    R., Quddus, M

    Velaga, N. R., Quddus, M. A., Bristow, A. L., and Zheng, Y. Map-aided integrity monitoring of a land vehicle navigation system. IEEE Transactions on Intelligent Transportation Systems, 13 0 (2): 0 848--858, 2012

    2012

  33. [41]

    Prediction on the urban gnss measurement uncertainty based on deep learning networks with long short-term memory

    Zhang, G., Xu, P., Xu, H., and Hsu, L.-T. Prediction on the urban gnss measurement uncertainty based on deep learning networks with long short-term memory. IEEE Sensors Journal, 21 0 (18): 0 20563--20577, 2021

    2021

  34. [42]

    T-gcn: A temporal graph convolutional network for traffic prediction

    Zhao, L., Song, Y., Zhang, C., Liu, Y., Wang, P., Lin, T., Deng, M., and Li, H. T-gcn: A temporal graph convolutional network for traffic prediction. IEEE transactions on intelligent transportation systems, 21 0 (9): 0 3848--3858, 2019

    2019

  35. [43]

    and Groves, P

    Zhong, Q. and Groves, P. Optimizing los/nlos modeling and solution determination for 3d-mapping-aided gnss positioning. In ION GNSS+ 2023, 2023

    2023

  36. [44]

    Enhancing gnss positioning accuracy for road monitoring systems: A factor graph optimization approach aided by geospatial information

    Zhong, Y., Hu, R., Bai, X., Li, X., Hsu, L.-T., and Wen, W. Enhancing gnss positioning accuracy for road monitoring systems: A factor graph optimization approach aided by geospatial information. IEEE Transactions on Instrumentation and Measurement, 2024

    2024

  37. [45]

    Graph neural networks: A review of methods and applications

    Zhou, J., Cui, G., Hu, S., Zhang, Z., Yang, C., Liu, Z., Wang, L., Li, C., and Sun, M. Graph neural networks: A review of methods and applications. AI open, 1: 0 57--81, 2020

    2020

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.