Pith. sign in

REVIEW 5 major objections 6 minor 38 references

Geo-ORBIT: A Federated Digital Twin Framework for Scene-Adaptive Lane Geometry Detection

T0 review · 5 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 The paper claims that FedMeta-GeoLane, a federated black-box meta-learning framework, lets roadside cameras learn lane geometry from vehicle trajectories while outperforming centralized meta-learning on both seen and unseen locations and…

desk verdict Plausible framework, but Table I cannot carry the central claim: component losses don't sum to totals and the meta-learner tunes its own evaluation weights. read the letter →

arxiv 2507.08743 v1 pith:7W4DEF4X submitted 2025-07-11 cs.CV cs.AI

Rei Tamaru , Pei Li , Bin Ran This is my paper · ORCID
classification cs.CVcs.AI
keywords lanedetectionfederatedlearningmeta-learningdigitaltwinroadsidecamerasvehicletrajectoryclusteringgeometrycommunicationcost
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 show that a digital twin of roadway geometry can be built cheaply and privately by letting each roadside camera learn its own lane-detection settings instead of relying on a single centralized model. Its central claim is that FedMeta-GeoLane, a federated meta-learning strategy, produces lower geometric error and better transfer to unseen camera locations than both a fixed baseline and a centrally trained meta-learner, while cutting communication cost by over 98%. The proposed Geo-ORBIT framework couples this federated meta-learner with a trajectory-based lane detection pipeline and a SUMO-CARLA digital twin so that detected lanes and real traffic flows are mirrored in simulation. If the claim holds, transportation agencies could scale lane-aware digital twins across hundreds of cameras without shipping raw video to a central server.

What carries the argument

The load-bearing mechanism is the black-box meta-learner $f_\phi$, a two-layer MLP with parameter-specific output heads that maps scene features $x_i$ to a dictionary of detection parameters $\theta_i = f_\phi(x_i)$ used by the knowledge-based lane detection pipeline. This design deliberately avoids backpropagating through non-differentiable components (histogram peak detection, KMeans, spline fitting); instead the meta-learner is supervised by geometric losses between detected and reference lane geometries, and its parameters are aggregated across roadside cameras via FedAvg. The composite objective $L_{\text{total}} = \lambda_1 L_{\text{consistency}} + \lambda_2 L_{\text{geometry}} + \lambda_3 L_{\text{center}} + \lambda_4 L_{\text{lane num}}$ weights Frechet-distance shape similarity, lane-width agreement, a triplet centerline embedding loss, and lane-count error, with the $\lambda_i$ themselves meta-learned. The system is completed by a SUMO-CARLA digital twin that maps detected trajectories into the simulation for validation.

What would settle it

Recompute Table I with all $\lambda_i$ fixed to 1.0 in Eq. (9) for every method; if FedMeta-GeoLane's $L_{\text{consistency}}$ is again exactly 0.0 and its $L_{\text{total}}$ advantage over Meta-GeoLane largely vanishes, then the claimed geometric superiority is an artifact of self-tuned evaluation weights rather than better lane reconstruction.

Watch

Extended reading notes

Core claim

The central discovery is that a non-differentiable, knowledge-based lane detection pipeline can be personalized to a scene by a black-box meta-learner that outputs the pipeline's control parameters (smoothing factor, histogram bin count, lane width estimate, and loss weights) from cheap contextual features such as trajectory speed and time of day. Because the meta-learner operates at the parameter level rather than through the detector, it can be trained in a federated loop: each roadside camera computes its own task loss and gradient, and the server averages the updates with FedAvg. In the paper's Table I, this FedMeta-GeoLane configuration reports a total validation loss of 6.94 on seen locations and 32.38 on unseen locations, compared to 12.16 and 69.61 for the centrally trained Meta-GeoLane, plus a reduction in transmission throughput from 3418 Mbps to 47.2 Mbps. The authors interpret this as evidence that federated meta-learning generalizes across locations and preserves privacy without sacrificing geometric fidelity.

Load-bearing premise

The load-bearing premise is that the composite validation loss $L_{\text{total}}$ in Eq. (9), whose weights the meta-model itself learns, is a faithful measure of lane geometry quality; if the model merely learns to down-weight the terms it finds hardest, the reported drop in $L_{\text{total}}$ does not prove better geometry.

Editorial extensions

If this is right

  • Deploying FedMeta-GeoLane to a new roadside camera requires only computing scene features and a forward pass of the meta-learner, so lane detection can be bootstrapped without retraining.
  • Agencies running hundreds of cameras could synchronize a lane-aware digital twin while transmitting only small model updates, not raw video streams.
  • Because raw images and trajectories never leave the local entity, the framework preserves privacy for drivers while still improving a shared global model.
  • The paper's reported 98% reduction in communication cost rests on exchanging 0.2 MB model parameters per client instead of 427 MB file uploads.
  • Lane-count estimation remains an open failure mode, so geometry-only trajectory clustering is not yet sufficient for complex interchanges or sparse-traffic lanes.

Reading between the lines

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

  • The near-exact $L_{\text{consistency}} = 0$ reported for FedMeta suggests that the meta-learned weights $\lambda_i$ may have driven the consistency term to near zero; a fairer comparison would freeze all loss weights at fixed equal values before comparing models.
  • A testable extension would be to run the same federated meta-learner on synthetic scenes with known ground-truth lane counts, isolating whether the remaining lane-count error is a clustering limitation or an OSM ground-truth mismatch.
  • The 3418 Mbps baseline mixes raw video uploads (427.3 MB per client) with a single training round, whereas the federated method runs 20 rounds of small updates; a more direct comparison would hold total bits and rounds constant.
  • Because the meta-learner's parameters include the loss weights themselves, the reported 'total loss' is partly self-assessed; an external evaluator using fixed geometric metrics (e.g., Frechet distance in meters) would give a less circular measure.
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

5 major / 6 minor

Summary. The paper introduces Geo-ORBIT, a federated digital twin framework for lane geometry detection from roadside cameras. The core component is GeoLane, a knowledge-based detector that reconstructs lanes from vehicle trajectories, extended by Meta-GeoLane, a black-box meta-learner that predicts detection parameters from scene features, and FedMeta-GeoLane, a federated variant that trains the meta-learner across camera sites without sharing raw data. The framework is integrated with SUMO and CARLA for digital twin synchronization. Experiments on four Wisconsin roadside cameras compare GeoLane, Meta-GeoLane, and FedMeta-GeoLane on seen and unseen locations, and report communication cost reductions relative to centralized training. The central claim is that FedMeta-GeoLane consistently outperforms baselines in geometric error and generalization while drastically reducing communication overhead.

Significance. If the empirical claims were adequately supported, the paper would make a useful contribution: it combines federated learning with black-box meta-learning for a non-differentiable lane detection pipeline, addresses privacy and communication costs in a digital twin setting, and provides a trajectory-based lane geometry sensing approach that avoids HD maps and expensive sensors. The communication cost analysis (Table II) is a genuine strength, and the public code release is commendable. However, the main quantitative evidence for the headline claim is currently not reliable: the primary evaluation metric is the meta-learner's own weighted objective, the reported table is internally inconsistent, the unseen-location comparison lacks the fixed baseline, and all results come from a single run without error bars. These issues undermine the paper's central claims as written, although they appear addressable with a revised evaluation protocol.

major comments (5)
  1. [Section V-B1, Eq. (9), Table I] The headline metric L_total is not a neutral evaluation metric: Eq. (9) defines it as a weighted sum with weights lambda_i, Fig. 3 lists 'loss weight' among the meta-learned parameters, and Section V-B1 explicitly concedes that the zero L_consistency may mean the model 'assigns negligible weight to this term during optimization.' A model can therefore lower L_total by down-weighting loss terms it cannot reduce, so the reported improvements (e.g., 99.1% and 37.5%) do not by themselves establish better lane geometry. The authors should report the learned lambda_i for each model, evaluate with a fixed set of weights, and present the unweighted component losses as the primary comparison.
  2. [Table I, Eq. (9)] Table I cannot be audited as reported because the lambda_i values are never given. With unit weights, the component losses do not sum to L_total in any row: for Seen Baseline, 5.45+15.12+6.78+5.00=32.35 but L_total is 77.84; for Seen Meta, 7.04+11.76+4.73+2.67=26.20 but L_total is 12.16; for Seen FedMeta, 0.0+2.65+3.16+2.67=8.48 but L_total is 6.94. The discrepancies are not a uniform scaling, since some rows require weights greater than one and others less than one. The authors must either report the weights used or replace L_total with the unweighted sum so that the table is internally consistent.
  3. [Table I, Section V-B1] The claim of 'stronger generalization to unseen locations' is not supported against the fixed-parameter baseline because Table I has no Baseline row for unseen locations; only Meta and FedMeta are shown. The text states that FedMeta 'reduces total error by over 50% relative to meta-learning' on unseen locations, but this does not establish that FedMeta outperforms the non-adaptive baseline there. Add a Baseline row for unseen locations and report the relative improvement over it.
  4. [Table I, Figs. 7-8] All quantitative results in Table I appear to come from a single training run. No standard deviations, multiple seeds, or statistical significance tests are reported; Figs. 7-8 show mean and standard deviation of the parameter alignment loss over epochs, not over independent runs. The claimed improvements may be within run-to-run variance, especially given the small number of camera sites. Report mean and standard deviation over at least several seeds for all reported losses.
  5. [Section IV-A vs. Section V-B1, Eq. (5)] The definition of L_consistency is inconsistent with its interpretation. Eq. (5) defines L_consistency as the Fréchet distance between detected and OSM reference centerlines, but Section V-B1 interprets a zero value as indicating 'stable predictions across video frames,' which is a different quantity. This makes the zero entry ambiguous and the discussion self-contradictory. Please clarify which quantity is actually measured, how it is computed, and why a zero value is plausible.
minor comments (6)
  1. [Section V-B1] The sentence 'improving total loss by %' contains a missing numerical value; please fill in the percentage or rephrase.
  2. [Section V-C, Table II] The text says the federated meta-learning framework exchanges parameters 'over 10 training rounds,' but Table II lists 20 rounds for Federated Meta. Please make these consistent.
  3. [Eq. (8)] The lane-number loss is defined as L without a subscript; rename it to L_lane_num for consistency with Table I and Eq. (9).
  4. [Eq. (7)] The centerline embedding loss uses a learned feature mapping f(·), but the paper never describes how f is trained or whether it is held fixed during evaluation. Please specify this.
  5. [Fig. 3, Section III-A2] Fig. 3 lists 'loss weight' among the meta-learned parameters, but Section III-A2 only describes smoothing and angle parameters. Please list all meta-learned parameters explicitly in one place.
  6. [References] Reference [26] contains stray text 'tLDR:' in the bibliography entry; remove it.

Circularity Check

1 steps flagged · score 6.0 of 10

The headline L_total metric in Table I is Eq. (9)'s training objective with loss weights among the meta-learner's outputs, so the reported 99.1%/37.5% improvements are partly self-scored rather than independent geometric error.

  1. self definitional [Eq. (9); Fig. 3 architecture diagram; Table I; Section V-B1.]
    "The overall training objective combines all previously defined loss terms into a single expression, weighted by hyper-parameters λi: Ltotal = λ1Lconsistency + λ2Lgeometry + λ3Lcenter + λ4Llane num. ... θ = {lane width, loss weight, etc…} ... leading the model to assign negligible weight to this term during optimization."

    Table I uses L_total as the headline geometric-error metric and bases the 99.1% and 37.5% improvement claims on it. But Eq. (9) defines L_total as the weighted sum of the component losses, and Fig. 3 lists 'loss weight' among the parameters θ that the meta-learner fφ outputs for each task. The model can therefore reduce L_total by predicting small λi for components it fits poorly, rather than by improving geometry. The paper explicitly concedes this: the FedMeta L_consistency = 0.0 entry may mean the model 'assigns negligible weight to this term during optimization.' The unreported λi also explain why the component losses do not sum to L_total in any Table I row (e.g., Seen FedMeta: 0.0 + 2.65 + 3.16 + 2.67 = 8.48 vs.

full rationale

The principal circular step is that Table I's headline performance metric is the same weighted objective the meta-learner is trained on, and the weights are themselves part of the predicted parameter set (Fig. 3). Thus part of FedMeta's reported advantage can be achieved by downweighting difficult terms, exactly as the paper's own 'negligible weight' caveat acknowledges. The unweighted component losses (e.g., L_geometry 2.65 vs. 15.12 for seen locations) and the 98% communication reduction provide some independent support, so the central claim is not entirely forced; hence the score is 6 rather than 8. The missing error bars, absent unseen-location baseline row, and arithmetic mismatch between components and L_total are correctness concerns rather than circularity, but they compound the problem that the headline numbers rest on an unreported, self-tuned weighting.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The central claim rests on the assumption that OSM pseudo-ground-truth is a valid reference, that trajectory-derived features are sufficient to predict detection parameters, and that the composite loss with learnable weights measures geometry. The last assumption is undermined by the zeroed consistency weight. No new physical entities are introduced.

free parameters (3)
  • theta_smoothing = not reported
    Spline smoothing parameter predicted by the meta-learner; controls lane centerline fitting in Eq. (2).
  • theta_angle = not reported
    Histogram bin count and peak prominence parameter used for lane count estimation in Section III-B.
  • Loss weights lambda_1 to lambda_4 = not reported
    Weights in the composite objective Eq. (9) are predicted by the meta-learner and can be set to zero, directly affecting the reported validation losses.
assumptions (4)
  • domain assumption OSM-derived lane centerlines are an accurate pseudo-ground-truth reference.
    The paper uses OSM data as reference for consistency and geometry losses (Section IV-A). If OSM is inaccurate, the loss values do not reflect true lane geometry.
  • ad hoc to paper Trajectory-derived features such as speed and time of day are sufficient to predict good detection parameters for a camera location.
    The meta-learner takes 'high-level contextual features' as input (Section III-A2), but no evidence is given that these features carry enough information to set optimal smoothing and histogram parameters.
  • ad hoc to paper The composite loss with learnable weights is a valid measure of lane geometry quality.
    Eq. (9) defines L_total with weights lambda_i that the meta-learner outputs; the paper's own discussion of L_consistency = 0 shows the weights can game the metric. Cited in Section V-B1.
  • standard math Standard algorithms (KMeans, spline fitting, Frechet distance) behave as assumed.
    These are well-established tools used in the pipeline and metrics.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Geo-ORBIT: A Federated Digital Twin Framework for Scene-Adaptive Lane Geometry Detection." pith.science (2026). https://pith.science/paper/7W4DEF4X

@misc{pith2026250708743,
  author       = {Pith},
  title        = {Pith review of: Geo-ORBIT: A Federated Digital Twin Framework for Scene-Adaptive Lane Geometry Detection},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/7W4DEF4X}},
  note         = {Machine review of arXiv:2507.08743}
}
read the original abstract

Digital Twins (DT) have the potential to transform traffic management and operations by creating dynamic, virtual representations of transportation systems that sense conditions, analyze operations, and support decision-making. A key component for DT of the transportation system is dynamic roadway geometry sensing. However, existing approaches often rely on static maps or costly sensors, limiting scalability and adaptability. Additionally, large-scale DTs that collect and analyze data from multiple sources face challenges in privacy, communication, and computational efficiency. To address these challenges, we introduce Geo-ORBIT (Geometrical Operational Roadway Blueprint with Integrated Twin), a unified framework that combines real-time lane detection, DT synchronization, and federated meta-learning. At the core of Geo-ORBIT is GeoLane, a lightweight lane detection model that learns lane geometries from vehicle trajectory data using roadside cameras. We extend this model through Meta-GeoLane, which learns to personalize detection parameters for local entities, and FedMeta-GeoLane, a federated learning strategy that ensures scalable and privacy-preserving adaptation across roadside deployments. Our system is integrated with CARLA and SUMO to create a high-fidelity DT that renders highway scenarios and captures traffic flows in real-time. Extensive experiments across diverse urban scenes show that FedMeta-GeoLane consistently outperforms baseline and meta-learning approaches, achieving lower geometric error and stronger generalization to unseen locations while drastically reducing communication overhead. This work lays the foundation for flexible, context-aware infrastructure modeling in DTs. The framework is publicly available at https://github.com/raynbowy23/FedMeta-GeoLane.git.

Figures

Figures reproduced from arXiv: 2507.08743 by the authors.

Figure 1
Figure 1. Transportation systems and digital twins. (a) Five-layer transportation system model adopted from [22], with exemplary entities on the different layers, [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Comparison of existing digital twin studies based on system architec [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Architecture of the federated meta-learning framework. The framework detects roadway geometry at local entities with local GeoLane models. The [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Overview of Knowledge-Based Lane Detection Algorithm. (a) Video [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: Locations of four cameras on US 12, Madison, Wisconsin, the US, [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: Qualitative comparison across multiple locations. The camera at Park is treated as an unseen location for Meta-GeoLane and FedMeta-GeoLane. Blue [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: FedMeta-GeoLane: the av￾erage and standard deviation of pa￾rameter alignment losses across ten epochs [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]
Figure 9
Figure 9. Figure 9: Digital twin synchronization with SUMO and CARLA at multiple [PITH_FULL_IMAGE:figures/full_fig_p009_9.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

38 extracted references · 37 canonical work pages

  1. [1]

    Digital twin: Generalization, characterization and implementation,

    E. VanDerHorn and S. Mahadevan, “Digital twin: Generalization, characterization and implementation,” 10 Decision Support Systems , vol. 145, p. 113524,

    work page

  2. [2]

    Digital twin in manufacturing: A categorical literature review and classification,

    W. Kritzinger, M. Karner, G. Traar, J. Henjes, and W. Sihn, “Digital twin in manufacturing: A categorical literature review and classification,” Ifac-PapersOnline, vol. 51, no. 11, pp. 1016–1022, 2018

    work page 2018

  3. [3]

    Characterising the digital twin: A systematic literature review,

    D. Jones, C. Snider, A. Nassehi, J. Yon, and B. Hicks, “Characterising the digital twin: A systematic literature review,” CIRP journal of manufacturing science and technology, vol. 29, pp. 36–52, 2020

    work page 2020

  4. [4]

    Digital twin: Enabling technologies, challenges and open research,

    A. Fuller, Z. Fan, C. Day, and C. Barlow, “Digital twin: Enabling technologies, challenges and open research,” IEEE access, vol. 8, pp. 108 952–108 971, 2020

    work page 2020

  5. [5]

    Digital twin- assisted cooperative driving at non-signalized intersec- tions,

    Z. Wang, K. Han, and P. Tiwari, “Digital twin- assisted cooperative driving at non-signalized intersec- tions,” IEEE Transactions on Intelligent Vehicles, vol. 7, no. 2, pp. 198–209, 2021

    work page 2021

  6. [6]

    Collaboration as a service: Digital-twin- enabled collaborative and distributed autonomous driv- ing,

    Y . Hui, X. Ma, Z. Su, N. Cheng, Z. Yin, T. Luan, and Y . Chen, “Collaboration as a service: Digital-twin- enabled collaborative and distributed autonomous driv- ing,” IEEE Internet Of Things Journal , vol. 9, pp. 18 607–18 619, 2022

    work page 2022

  7. [7]

    Mixed cloud control testbed: Validating vehicle-road-cloud integration via mixed dig- ital twin,

    J. Dong, Q. Xu, J. Wang, C. Yang, M. Cai, C. Chen, Y . Liu, J. Wang, and K. Li, “Mixed cloud control testbed: Validating vehicle-road-cloud integration via mixed dig- ital twin,” IEEE Transactions on Intelligent Vehicles , vol. 8, no. 4, pp. 2723–2736, 2023

    work page 2023

  8. [8]

    Digital twin-based cloud-native vehicular net- works architecture for intelligent driving,

    X. Tan, Q. Meng, M. Wang, Q. Zheng, J. Wu, and J. Yang, “Digital twin-based cloud-native vehicular net- works architecture for intelligent driving,” IEEE Net- work, 2023

    work page 2023

Show all 38 references
  1. [9]

    A digital twin-based traffic guidance scheme for autonomous driving,

    X. Liao, S. Leng, Y . Sun, K. Zhang, and M. A. Im- ran, “A digital twin-based traffic guidance scheme for autonomous driving,” IEEE Internet of Things Journal , 2024

    2024

  2. [10]

    Smart mobility digital twin based automated vehicle navigation system: A proof of concept,

    K. Wang, Z. Li, K. Nonomura, T. Yu, K. Sakaguchi, O. Hashash, and W. Saad, “Smart mobility digital twin based automated vehicle navigation system: A proof of concept,” IEEE Transactions on Intelligent Vehicles , 2024

    2024

  3. [11]

    Task-efficiency oriented v2x communications: Digital twin meets mobile edge computing,

    G. Cai, B. Fan, Y . Dong, T. Li, Y . Wu, and Y . Zhang, “Task-efficiency oriented v2x communications: Digital twin meets mobile edge computing,” IEEE Wireless Communications, pp. 1–8, 2023

    2023

  4. [12]

    Wireless digital twin for assessing the reliability of vehicular communication links,

    S. Zelenbaba, B. Rainer, M. Hofer, and T. Zemen, “Wireless digital twin for assessing the reliability of vehicular communication links,” in 2022 IEEE Globecom Workshops (GC Wkshps). IEEE, 2022, pp. 1034–1039

    2022

  5. [13]

    Application of lightweight digital twin system in intel- ligent transportation,

    Q. Liu, X. Qi, S. Liu, X. Cheng, X. Ke, and F. Wang, “Application of lightweight digital twin system in intel- ligent transportation,” IEEE Journal of Radio Frequency Identification, vol. 6, pp. 729–732, 2022

    2022

  6. [14]

    A multi-modal simulation framework to enable digital twin-based v2x communications in dynamic environments,

    L. Cazzella, F. Linsalata, M. Magarini, M. Matteucci, and U. Spagnolini, “A multi-modal simulation framework to enable digital twin-based v2x communications in dynamic environments,” in 2024 IEEE 100th Vehicular Technology Conference (VTC2024-Fall) . IEEE, 2024, pp. 1–6

    2024

  7. [15]

    Spat/map v2x communication between traffic light and vehicles and a realization with digital twin,

    T. W ´agner, T. Orm ´andi, T. Tettamanti, and I. Varga, “Spat/map v2x communication between traffic light and vehicles and a realization with digital twin,” Computers And Electrical Engineering , vol. 106, p. 108560, 2023

    2023

  8. [16]

    A Transportation Digital-Twin Approach for Adaptive Traffic Control Systems,

    S. Dasgupta, M. Rahman, A. D. Lidbe, W. Lu, and S. Jones, “A Transportation Digital-Twin Approach for Adaptive Traffic Control Systems,” Jul. 2023, arXiv:2109.10863 [physics]. [Online]. Available: http: //arxiv.org/abs/2109.10863

    2023 arXiv

  9. [17]

    Digital twin for pedestrian safety warning at a single urban traffic intersection,

    Y . Fu, M. K. Turkcan, V . Anantha, Z. Kostic, G. Zuss- man, and X. Di, “Digital twin for pedestrian safety warning at a single urban traffic intersection,” in 2024 IEEE Intelligent Vehicles Symposium (IV). IEEE, 2024, pp. 2640–2645

    2024

  10. [18]

    Scan-to-graph: automatic generation and representation of highway geometric digital twins from point cloud data,

    Y . Pan, M. Wang, L. Lu, R. Wei, S. Cavazzi, M. Peck, and I. Brilakis, “Scan-to-graph: automatic generation and representation of highway geometric digital twins from point cloud data,” Automation in Construction , vol. 166, p. 105654, 2024

    2024

  11. [19]

    Building digital twins of existing highways using map data based on engineering expertise,

    F. Jiang, L. Ma, T. Broyd, W. Chen, and H. Luo, “Building digital twins of existing highways using map data based on engineering expertise,” Automation in Construction, vol. 134, p. 104081, 2022

    2022

  12. [20]

    3d highway curve reconstruction from mobile laser scanning point clouds,

    Z. Zhang, J. Li, Y . Guo, C. Yang, and C. Wang, “3d highway curve reconstruction from mobile laser scanning point clouds,” IEEE Transactions on Intelligent Trans- portation Systems, vol. 21, no. 11, pp. 4762–4772, 2019

    2019

  13. [21]

    Automating construction of road digital twin geometry using context and location aware segmentation,

    D. Davletshina, V . K. Reja, and I. Brilakis, “Automating construction of road digital twin geometry using context and location aware segmentation,” Automation in Con- struction, vol. 168, p. 105795, 2024

    2024

  14. [22]

    6-layer model for a struc- tured description and categorization of urban traffic and environment,

    M. Scholtes, L. Westhofen, L. R. Turner, K. Lotto, M. Schuldes, H. Weber, N. Wagener, C. Neurohr, M. H. Bollmann, F. K ¨ortke et al. , “6-layer model for a struc- tured description and categorization of urban traffic and environment,” iEEE Access , vol. 9, pp. 59 131–59 147, 2021

    2021

  15. [23]

    Advances and open problems in federated learning,

    P. Kairouz, H. B. McMahan, B. Avent, A. Bellet, M. Ben- nis, A. N. Bhagoji, K. Bonawitz, Z. Charles, G. Cormode, R. Cummings et al. , “Advances and open problems in federated learning,” Foundations and trends® in machine learning, vol. 14, no. 1–2, pp. 1–210, 2021

    2021

  16. [24]

    A survey on federated learning in intelligent transportation systems,

    R. Zhang, J. Mao, H. Wang, B. Li, X. Cheng, and L. Yang, “A survey on federated learning in intelligent transportation systems,” IEEE Transactions on Intelligent Vehicles, 2024

    2024

  17. [25]

    A digital twin-based traffic light management system using BIRCH algorithm,

    H. Y . Adarbah, M. Sookhak, and M. Atiquzzaman, “A digital twin-based traffic light management system using BIRCH algorithm,” Ad Hoc Networks , vol. 164, p. 103613, Nov. 2024. [Online]. Available: https://www.sciencedirect.com/science/article/ pii/S1570870524002245

    2024

  18. [26]

    A Digital Twin Environment for 5G Vehicle-to-Everything: Architecture and Open Issues,

    ——, “A Digital Twin Environment for 5G Vehicle-to-Everything: Architecture and Open Issues,” in Proceedings of the Int’l ACM Symposium on Performance Evaluation of Wireless Ad Hoc, Sensor, & Ubiquitous Networks . Montreal Quebec 11 Canada: ACM, Oct. 2023, pp. 115–122, tLDR: Th...

    2023

  19. [27]

    Towards next generation of pedestrian and connected vehicle in-the-loop research: A digital twin co- simulation framework,

    Z. Wang, O. Zheng, L. Li, M. Abdel-Aty, C. Cruz-Neira, and Z. Islam, “Towards next generation of pedestrian and connected vehicle in-the-loop research: A digital twin co- simulation framework,” IEEE Transactions on Intelligent Vehicles, vol. 8, no. 4, pp. 2674–2683, 2023

    2023

  20. [28]

    Intelligent highway adaptive lane learning system in multiple rois of surveillance camera video,

    M. Qiu, L. Christopher, S. Y .-P. Chien, and Y . Chen, “Intelligent highway adaptive lane learning system in multiple rois of surveillance camera video,” IEEE Trans- actions on Intelligent Transportation Systems , vol. 25, no. 8, pp. 8591–8601, 2024

    2024

  21. [29]

    Lane detection in video-based intelligent transportation monitoring via fast extracting and clustering of vehicle motion trajectories,

    J. Ren, Y . Chen, L. Xin, and J. Shi, “Lane detection in video-based intelligent transportation monitoring via fast extracting and clustering of vehicle motion trajectories,” Mathematical Problems in Engineering, vol. 2014, no. 1, p. 156296, 2014

    2014

  22. [30]

    A review of ap- plications in federated learning,

    L. Li, Y . Fan, M. Tse, and K.-Y . Lin, “A review of ap- plications in federated learning,” Computers & Industrial Engineering, vol. 149, p. 106854, 2020

    2020

  23. [31]

    Adaptive federated learning and digital twin for industrial internet of things,

    W. Sun, S. Lei, L. Wang, Z. Liu, and Y . Zhang, “Adaptive federated learning and digital twin for industrial internet of things,” IEEE Transactions on Industrial Informatics , vol. 17, no. 8, pp. 5605–5614, 2021

    2021

  24. [32]

    Digital twins-enabled federated learning in mo- bile networks: From the perspective of communication- assisted sensing,

    J. Mu, W. Ouyang, T. Hong, W. Yuan, Y . Cui, and Z. Jing, “Digital twins-enabled federated learning in mo- bile networks: From the perspective of communication- assisted sensing,” IEEE Journal on Selected Areas in Communications, vol. 41, no. 10, pp. 3230–3241, 2023

    2023

  25. [33]

    Digital twin-enabled efficient federated learning for col- lision warning in intelligent driving,

    L. Tang, M. Wen, Z. Shan, L. Li, Q. Liu, and Q. Chen, “Digital twin-enabled efficient federated learning for col- lision warning in intelligent driving,” IEEE Transactions on Intelligent Transportation Systems, vol. 25, no. 3, pp. 2573–2585, 2024

    2024

  26. [34]

    Federated learning for digital twin-based vehicular networks: Architecture and chal- lenges,

    L. U. Khan, E. Mustafa, J. Shuja, F. Rehman, K. Bilal, Z. Han, and C. S. Hong, “Federated learning for digital twin-based vehicular networks: Architecture and chal- lenges,” IEEE Wireless Communications , vol. 31, no. 2, pp. 156–162, 2024

    2024

  27. [35]

    Learning to learn from apis: Black-box data-free meta-learning,

    Z. Hu, L. Shen, Z. Wang, B. Wu, C. Yuan, and D. Tao, “Learning to learn from apis: Black-box data-free meta-learning,” in International Conference on Machine Learning. PMLR, 2023, pp. 13 610–13 627

    2023

  28. [36]

    Model-agnostic meta-learning for fast adaptation of deep networks,

    C. Finn, P. Abbeel, and S. Levine, “Model-agnostic meta-learning for fast adaptation of deep networks,” in International conference on machine learning . PMLR, 2017, pp. 1126–1135

    2017

  29. [37]

    Ultralytics YOLO,

    G. Jocher, J. Qiu, and A. Chaurasia, “Ultralytics YOLO,” Jan. 2023. [Online]. Available: https://github. com/ultralytics/ultralytics

    2023

  30. [2021]

    Available: https://www.sciencedirect

    [Online]. Available: https://www.sciencedirect. com/science/article/pii/S0167923621000348

Pith tools

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