Pith. sign in

REVIEW 3 major objections 4 minor 46 references

A vision-language pipeline called KYA can watch dashcam video, detect a sudden cut-in, and reply out loud with an emotionally matched reaction—humorous, analytical, angry, or forgiving—and the paper reports that users prefer such responses,

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-01 21:07 UTC pith:BMET3E34

load-bearing objection The 4.29 headline score is a constructed redistribution of a single global mean, not a measured rating — the prototype and preference data are real, but the evaluation as described does not support the abstract. the 3 major comments →

arxiv 2607.16181 v1 pith:BMET3E34 submitted 2026-07-17 cs.CV

Vision-Language Assistant for Emotional Reactions to Risky Driving

Harine Choi , Eun Hak Lee , Zhengzhong Tu This is my paper
classification cs.CV
keywords KYArisky drivingcut-in detectionvision-language modelsadvanced driver assistance systemsemotional AIprojected reach time (PRT)user study
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

This paper tries to establish that an in-car assistant can do more than recite neutral warnings: a modular pipeline called KYA watches dashcam video, spots a sudden cut-in, and replies out loud in a personality-matched voice—humorous, analytical, angry, or forgiving—chosen by the driver. The case is built with two modules. A YOLOv8-based vision module tracks the closest threatening vehicle and compresses the interaction into a small structured behavior log centered on PRT, a risk metric equal to the distance to that vehicle divided by the ego vehicle's speed. That log, plus the driver's persona choice, is fed to a large language model (ChatGPT-4o, Claude 3, Gemini 2.5, or Copilot), which turns it into a spoken reaction. On 20 real dashcam clips and 108 survey participants, the paper finds that users most like humorous and analytical personas and that the YOLOv8s-plus-ChatGPT-4o pairing reaches the highest overall score, 4.29 out of 5. If taken at face value, the contribution is a template for emotionally intelligent driving support: the language model never sees pixels, only interpretable risk metrics, so the same design could be adapted to other dangerous maneuvers.

Core claim

On the paper's own terms, the finding is that sudden cut-ins can be detected from ordinary dashcam video and translated into persona-matched emotional speech with no trajectory prediction and no extra sensors. The paper defines cut-in risk through PRT—the time the ego vehicle would need to reach the aggressor's current position—and treats a sharp drop in PRT as the signature of a risky maneuver. It then converts detector outputs (distance, relative speed, vehicle type and color, cut-in direction) into a normalized behavior log; the LLM reads only this log, not the image, and produces a reaction aligned with one of seven user-selected personas. In a blind 108-participant evaluation, ChatGPT-4

What carries the argument

The load-bearing mechanism is PRT (Projected Reach Time), Eq. 1: PRT = d_rel/(v_ego + ε), where d_rel is the distance from the ego vehicle to the closest threatening vehicle and v_ego is the ego vehicle's speed, with a small ε preventing division by zero. PRT is deliberately ego-centric—it ignores the aggressor's future trajectory—which makes it computable from a single dashcam and sensitive to the sudden distance drops that define cut-ins. Around PRT, the paper builds a structured behavior log: frame-by-frame records of the aggressor's relative distance, relative speed, vehicle type, color, and cut-in direction, normalized for scale invariance across cameras. This log is the only channel be

Load-bearing premise

The whole pipeline rests on the assumption that relative distance in meters and speeds in km/h can be estimated accurately enough from an uncalibrated dashcam to make PRT meaningful; the paper never states how pixel positions become metric distances and speeds, and its own Results (Table 2) caution that extreme speed values are frame-level positional estimates rather than calibrated measurements.

What would settle it

Instrument a vehicle with GPS and radar (or a depth camera) and drive the same cut-in scenarios; compute PRT from YOLOv8-derived bounding boxes and compare each value with the measured time-to-reach. If the two disagree systematically, the 1,189 'high-risk' frames—and every LLM input built from them—are untrustworthy.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • An in-car warning system could flag sudden cut-ins from a single dashcam, without radar, lidar, or motion prediction, because PRT only needs ego speed and observed distance.
  • Users' stated preference for humorous and analytical personas over angry ones suggests that persona-matched phrasing could make warnings more acceptable and less irritating than a fixed neutral alert.
  • Because the language module ingests structured logs rather than images, tailgating (time headway), sudden braking (deceleration), and other events could be added by swapping the behavioral metric.
  • The survey's gender and voice preferences indicate that one emotional style will not fit all drivers, pushing future deployments toward adaptive personalization.
  • All four LLMs received favorable ratings, which the paper reads as evidence that emotional alignment is achievable across models, with tone matching mattering more than whichever LLM is used.

Where Pith is reading between the lines

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

  • A careful reading of the Results shows that 4.29 is a constructed allocation: only overall mean ratings (3.35 zero-shot, 3.49 few-shot) were observed, then distributed across models proportionally to preference weights; it is a ranking index, not an independently measured 5-point score for ChatGPT-4o.
  • If PRT is validated against calibrated sensors (GPS, radar, or a depth camera), the behavior-log interface would become a general 'risk-to-language' adapter for many maneuvers, not just cut-ins.
  • The preference data imply a testable extension: an adaptive assistant that shifts tone with event severity or driver state (humor for minor annoyances, calm analysis for real danger) may outperform any fixed persona; the paper did not test this.
  • The structured, human-readable log also suggests a route to explainability and audit: since the LLM's input is already a small risk record, users and regulators could inspect exactly which distance/speed values triggered a given reaction.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 4 minor

Summary. The paper introduces KYA, a two-module vision-language pipeline for detecting risky driving behaviors (specifically sudden cut-ins) from dashcam video and generating persona-aligned emotional responses. The vision module uses YOLOv8 variants to detect vehicles and computes relative distance, relative speed, and a proposed risk indicator, Projected Reach Time (PRT), from uncalibrated monocular video. The language module feeds the resulting behavior log to LLMs (ChatGPT-4o, Claude 3, Gemini 2.5, Copilot) with user-selected emotional personas. The system is evaluated on 20 YouTube dashcam videos and a 108-participant survey. The headline claim is that 'the combination of YOLOv8s and ChatGPT-4o achieved the highest score of 4.29 out of 5.00.'

Significance. If the evaluation were valid, the paper would make a useful contribution to emotionally adaptive in-vehicle assistants, an underexplored area relative to perception-only ADAS. The modular design is simple and interpretable, and the survey-based comparison across personas is a reasonable way to assess subjective response quality. The authors also deserve credit for reporting the allocation procedure used to compute the overall scores, which makes the central validity problem transparent rather than hidden. However, the headline quantitative claim is not supported by the data as described: no participant rated any model 4.29, and the vision module's metric estimation rests on unspecified calibration. Because the central claims of the paper depend on these points, the current manuscript cannot be accepted as is.

major comments (3)
  1. [Results of Language Model, Table 3] The reported 'Overall Score' values (e.g., 4.29 for ChatGPT-4o, few-shot) are not measured ratings. The text states: 'Because only the overall mean rating was available for each prompting condition (3.35 for zero-shot and 3.49 for few-shot), we proportionally distributed this observed average across models according to their weighted preference scores.' Thus each 'overall score' is an arithmetic partition of a single global mean, not an independent rating of any model. No participant ever assigned 4.29 to YOLOv8s+ChatGPT-4o, and that combination was never tested as an integrated system in the survey. The Abstract's claim that this combination 'achieved the highest score of 4.29 out of 5.00' is therefore unsupported by the experiment. The authors should either conduct a survey with per-model ratings and report measured means, or clearly label these values as derived allocation scores and
  2. [Methods, Vehicle Detection; Eq. (1); Table 2] The system requires absolute units: 'estimated speed in kilometers per hour' and 'distance from the ego vehicle' (in meters) feed Eq. (1) for PRT and the structured behavior log. No calibration procedure, depth model, or monocular speed-estimation method is described; the 'normalization procedure' preserves only relative proximity, not absolute metric values. The paper itself acknowledges that relative speeds up to 168 km/h are 'instantaneous relative velocity estimates derived from frame-level positional changes rather than calibrated vehicle speed measurements.' Without a validated mapping from pixels to meters and kph, the PRT values, the 1,189-frame risk labeling, and every behavioral input to the LLM are quantitatively ungrounded. The authors must either provide a calibration and error analysis or explicitly reframe the work as using only relative, unit-free visual cues and remove a
  3. [Results, Driving Behavior Estimation] The identification of 1,189 high-risk frames is a load-bearing step, but the selection process is under-specified. The text says frames were 'selected based on thresholding relative behavior indicators and verified through visual inspection,' yet the thresholds are not given, no ground-truth labeling protocol is described, and no inter-rater reliability is reported. The paper also states 'we are not defining cut-in events ourselves' while simultaneously relying on a scene-selection criterion ('search terms related to sudden cut-ins') and a frame-level risk threshold. This leaves the reader unable to assess the quality of the risk labels or to reproduce the pipeline. The authors should report the thresholds and provide a validation of the labeling method.
minor comments (4)
  1. [Results of Language Model, statistical test] The chi-square test on raw selection frequencies is reported as 'statistically significant' (χ² = 12.20, p = 0.0067), but the authors acknowledge that participants could select multiple responses, violating the independence assumption. This should be either omitted or replaced with a test appropriate for multiple-selection data (e.g., a permutation test).
  2. [Data, Video dataset] The text says 'Out of the 20 vehicles observed' in the Results; this appears to mean 20 videos/sequences, not vehicles. Please correct the wording.
  3. [Throughout] There are occasional typos and reference inconsistencies, e.g., 'vison-language' in the Introduction, and the reference list entries are not always matched to in-text citations with consistent numbering. A careful proofreading pass is needed.
  4. [Figure 3] The histograms lack axis labels for all subplots and do not report the bin width or the number of frames. Adding these details would improve interpretability.

Circularity Check

1 steps flagged

Headline '4.29/5.00' is a constructed reallocation of the survey's own global mean, so the top-model claim is not an independent measurement.

specific steps
  1. fitted input called prediction [Results of Language Model, 'Performance of Emotional Response Generation' (Table 3); echoed in Abstract]
    "Because only the overall mean rating was available for each prompting condition (3.35 for zero-shot and 3.49 for few-shot), we proportionally distributed this observed average across models according to their weighted preference scores."

    The 'Overall Score' column (e.g., ChatGPT-4o = 4.29 in few-shot) is not a measured model rating. It is exactly the single observed 3.49 group mean, deterministically partitioned among models in proportion to weighted preference scores. Those weighted preference scores are computed from the same 108-participant survey's selection rates and from persona-importance weights that the paper states are 'derived from observed user preferences' in that same survey. Therefore the reported ranking and the abstract's 'highest score of 4.29' reduce, by construction, to the survey's own preference data. No participant rated ChatGPT-4o as 4.29, and no integrated YOLOv8s+ChatGPT-4o system was measured in this language-model evaluation.

full rationale

The KYA pipeline has independent components: Table 1 reports YOLOv8 detection metrics from dashcam frames, and Table 3 reports language-model preference statistics from 108 participants. The detection metrics are self-contained and not circular. The circularity is concentrated in the headline '4.29/5.00' claim. The paper explicitly says only a single overall mean rating per prompting condition (3.35 for zero-shot, 3.49 for few-shot) was available and that this mean was 'proportionally distributed' across models according to weighted preference scores. Because the weighted preference scores and the persona-importance weights both come from the same survey whose data they re-express, the model-level 'Overall Score' is not an independent measurement; it is a reparametrization of the survey's own preference structure. Thus the central claim that 'the combination of YOLOv8s and ChatGPT-4o achieved the highest score of 4.29' is forced by construction rather than observed. The unresolved monocular distance/speed estimation and the absence of an integrated vision+language experiment are validity concerns rather than circularity, so they do not increase the score further. Since the vision-module results provide independent content, the paper is not wholly circular.

Axiom & Free-Parameter Ledger

6 free parameters · 6 axioms · 0 invented entities

The central claims rest on: (1) an undocumented monocular speed/distance estimation that produces the meters and kph feeding PRT; (2) a self-constructed evaluation label — risk frames defined by thresholding the paper's own PRT indicator; (3) a within-survey weighting scheme that manufactures the headline 4.29 overall score; and (4) hand-set thresholds (confidence 0.3, sampling 5–10 fps, unreported risk thresholds). Nearly every quantitative output (PRT values, risky-frame counts, overall scores) depends on parameters that are unreported or derived from the same data being evaluated, so the reader cannot tell how much of the result is contributed by the paper vs. imported from prior work or the survey itself. No invented entities: PRT is a new named metric (Eq. 1) but not an entity with a falsifiable handle; no particles, forces, mediators, dimensions, or conserved quantities are introduced.

free parameters (6)
  • YOLOv8n/s confidence threshold = 0.3
    Lowered from 0.5 to 0.3 to boost detection of small/occluded vehicles (Methods, Vehicle Detection); hand-chosen and affects which detections enter the behavior log.
  • Persona importance weights = neutral 0.083, light anger 0.083, noticeable anger 0.133, intense anger 0.110, analytical 0.232, humorous 0.265, magnani
    'Relative importance of each persona derived from observed user preferences' from the same 108-participant survey; these weights determine the constructed overall scores (Results, Performance of Emotional Response Generation).
  • High-risk frame thresholds = not reported
    The 1,189 risky frames were 'selected based on thresholding relative behavior indicators and verified through visual inspection' (Results, Driving Behavior Estimation); threshold values are never given, so the evaluation labels are not reproducible.
  • Per-video scale anchor for normalization = not specified
    A reference 'scale anchor' rescales all distance/speed features per video (Methods, Vehicle Detection); how the reference is chosen is unstated, and PRT magnitudes depend on it.
  • Video sampling rate = 5–10 fps
    Set 'depending on scene complexity' (Methods, Vision Model); affects temporal derivatives used for relative speed and PRT.
  • PRT denominator epsilon = unspecified small constant
    Added to prevent division by zero in Eq. (1); negligible but unstated.
axioms (6)
  • domain assumption Each scene contains at least one vehicle performing a risky cut-in toward the ego vehicle
    Stated explicitly: 'Since our dataset exclusively consists of cut-in scenarios, we assume that each scene contains at least one vehicle performing a risky maneuver toward the ego vehicle' (Methods, Driving Behavior Recognition).
  • domain assumption Monocular dashcam frames can be converted to absolute meters and kph
    The behavior log and PRT (Eq. 1) require relative distance in meters and ego speed in kph, yet no calibration or monocular depth/speed estimation method is described (Methods, Vehicle Detection).
  • domain assumption PRT is a valid proxy for collision risk
    PRT = d_rel/(v_ego+ε) ignores the aggressor's velocity and trajectory; the paper asserts it 'signals an emerging proximity hazard' without validating it against TTC, PET, or crash/near-crash data (Methods, Driving Behavior Recognition).
  • domain assumption The 'three-second rule' is the correct normative threshold for cut-in reaction time
    NHTSA's three-second rule is a following-distance guideline; it is applied here to cut-in PRT values to conclude that reaction windows are 'dangerously narrow' (Results, Driving Behavior Estimation).
  • domain assumption YOLOv8 pretrained on BDD100K/nuScenes transfers to curated YouTube dashcam footage
    Fine-tuning is mentioned for n/s variants but no training data or protocol is given; detection metrics depend on this transfer (Methods, Vehicle Detection).
  • domain assumption Forced-choice preference selections measure 'emotional alignment' of LLM responses
    The user study treats preference among anonymized responses as a measure of persona-consistency/emotional alignment without a validated instrument (Data; Results).

pith-pipeline@v1.3.0-alltime-deepseek · 14403 in / 27223 out tokens · 214019 ms · 2026-08-01T21:07:18.240849+00:00 · methodology

0 comments
read the original abstract

This study introduces a vision-language pipeline that detects risky driving behaviors and generates emotionally expressive responses to support driver awareness and comfort. Although vision-language models have advanced perception and reasoning in autonomous driving, existing systems rarely consider the emotional dimension or real-world user experience. Keep Yelling Assistant (KYA) detects high-risk driving maneuvers in real time, such as sudden cut-ins. It then produces emotional responses through a large language model tailored to driver preferences. The framework comprises two core modules. The vision module uses YOLOv8 variants to detect nearby vehicles and identify risky behaviors such as sudden cut-ins. Key driving metrics, including relative distance, speed, and projected reach time, are extracted and normalized to produce a structured behavior log. The language module processes this log with user-defined emotional tone settings, such as neutral, humorous, and analytical, and generates verbal reactions using state-of-the-art large language models, including ChatGPT-4o, Claude 3, Gemini 2.5, and Copilot. We evaluated the proposed system using dashcam videos containing risky driving behaviors and a user study involving 108 participants. Participants selected preferred response styles, and the large language models were evaluated based on emotional alignment. All models received favorable ratings, although preferences varied across personas. Notably, the combination of YOLOv8s and ChatGPT-4o achieved the highest score of 4.29 out of 5.00. By integrating real-world perception with emotionally adaptive dialogue, KYA introduces a new paradigm for emotionally intelligent in-vehicle artificial intelligence. It offers promising directions for improving safety, trust, and emotional well-being in both conventional and autonomous vehicles.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

46 extracted references · 2 canonical work pages

  1. [1]

    H., Yun, H., Cho, S

    Lee, E. H., Yun, H., Cho, S. H., & Lee, E. (2025). Aggressive driving in ride-hailing: Work hours and road conditions. Journal of Transportation Safety & Security, 1-24

    2025

  2. [2]

    Jafarpour, S., & Rahimi-Movaghar, V. (2014). Determinants of risky driving behavior: a narrative review. Medical journal of the Islamic Republic of Iran, 28, 142

    2014

  3. [3]

    Department of Transportation, National Highway Traffic Safety Administration

    U.S. Department of Transportation, National Highway Traffic Safety Administration. (2024, November 25). NHTSA releases early estimates showing nationwide decreases in traffic fatalities in priority safety areas during first half of 2024, including pedestrian and speeding-related crashes. https://www.nhtsa.gov/press-releases/nhtsa-releases-2024-early-estim...

    2024

  4. [4]

    H., Yang, X

    Jiang, Z. H., Yang, X. G., Sun, T., Wang, T., & Yang, Z. (2021). Investigating the relationship between traffic violations and crashes at signalized intersections: an empirical study in China. Journal of advanced transportation, 2021(1), 4317214

    2021

  5. [5]

    (2024) Road traffic accident fatality rate

    Statistics Korea. (2024) Road traffic accident fatality rate. KOSIS Index (JipyoNuri). Retrieved July 29, 2025, https://www.index.go.kr/unify/idx-info.do?idxCd=4261

    2024

  6. [6]

    M., Xavier, C., & Pereira, P

    Morgado, S. M., Xavier, C., & Pereira, P. (2023, November). Prevention and Road Safety Campaigns Through Emotion in Portugal: A Comprehensive Empirical Analysis. In International Conference on Marketing and Technologies (pp. 139-155). Singapore: Springer Nature Singapore

    2023

  7. [7]

    V., & Karthikeyan, J

    Chand, H. V., & Karthikeyan, J. (2022). CNN Based Driver Drowsiness Detection System Using Emotion Analysis. Intelligent Automation & Soft Computing, 31(2)

    2022

  8. [8]

    & Wortmann, F

    Liu, S., Koch, K., Zhou, Z., Föll, S., He, X., Menke, T., ... & Wortmann, F. (2021). The empathetic car: Exploring emotion inference via driver behaviour and traffic context. Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies, 5(3), 1-34. Choi, Lee and Tu 17

    2021

  9. [9]

    & Cao, D

    Li, W., Wu, L., Wang, C., Xue, J., Hu, W., Li, S., ... & Cao, D. (2022). Intelligent cockpit for intelligent vehicle in metaverse: A case study of empathetic auditory regulation of human emotion. IEEE Transactions on Systems, Man, and Cybernetics: Systems, 53(4), 2173-2187

    2022

  10. [10]

    Jadhav, S. S. (2024). Advancing Safety in Vehicles with AI-Driven Emotion Recognition (Doctoral dissertation, Dublin, National College of Ireland)

    2024

  11. [11]

    Pandya, D., & Thakkar, A. (2024). Sentiment analysis of self driving car dataset: A comparative study of deep learning approaches. Procedia Computer Science, 235, 12-21

    2024

  12. [12]

    Huang, B., Lv, J., & Qiang, L. (2025). Influencing driving safety by matching AI assistant's verbal emotions to driver: A randomized controlled trial on performance, attention, and emotion. Computers in Human Behavior, 169, 108667

    2025

  13. [13]

    Wang, Y., Gu, Y., Quan, T., Yang, J., Dong, M., An, N., & Ren, F. (2025). ViE-Take: A Vision- Driven Multi-Modal Dataset for Exploring the Emotional Landscape in Takeover Safety of Autonomous Driving. Research, 8, 0603

    2025

  14. [14]

    & Kompatsiaris, I

    Stappen, L., Baird, A., Rizos, G., Tzirakis, P., Du, X., Hafner, F., ... & Kompatsiaris, I. (2020, October). Muse 2020 challenge and workshop: Multimodal sentiment analysis, emotion-target engagement and trustworthiness detection in real-life media: Emotional car reviews in-the-wild. In Proceedings of the 1st International on Multimodal Sentiment Analysis...

    2020

  15. [15]

    Zou, Z., Khan, A., Lwin, M., Alnajjar, F., & Mubin, O. (2025). Investigating the impacts of auditory and visual feedback in advanced driver assistance systems: a pilot study on driver behavior and emotional response. Frontiers in Computer Science, 6, 1499165

    2025

  16. [16]

    Wachter, F. (2019). Emotional driving in AI-powered Cars; Driver and traffic safety. Emotion tracking and route prediction, alongside rewarding and education for good driving behaviour

    2019

  17. [17]

    (2023, April)

    Giri, M., Bansal, M., Ramesh, A., & Satvik, D. (2023, April). Enhancing safety in vehicles using emotion recognition with artificial intelligence. In 2023 IEEE 8th International Conference for Convergence in Technology (I2CT) (pp. 1-10). IEEE

    2023

  18. [18]

    H., Feng, S., & Li, F

    Zhang, Z., Li, D., Sun, Y., Lee, C. H., Feng, S., & Li, F. (2025, May). How Do We Team Up? Human-Machine Co-driving Style Assessment Through Visual Dynamic Analysis and Vision- Language Model. In International Conference on Human-Computer Interaction (pp. 287-304). Cham: Springer Nature Switzerland

    2025

  19. [19]

    Zheng, P., Zhao, Y., Gong, Z., Zhu, H., & Wu, S. (2024). Simplellm4ad: An end-to-end vision- language model with graph visual question answering for autonomous driving. arXiv preprint arXiv:2407.21293

    Pith/arXiv arXiv 2024

  20. [20]

    (2025, April)

    Zhao, G., Wang, X., Zhu, Z., Chen, X., Huang, G., Bao, X., & Wang, X. (2025, April). Drivedreamer-2: Llm-enhanced world models for diverse driving video generation. In Proceedings of the AAAI Conference on Artificial Intelligence (Vol. 39, No. 10, pp. 10412-10420)

    2025

  21. [21]

    & Gong, J

    Jin, Y., Yang, R., Yi, Z., Shen, X., Peng, H., Liu, X., ... & Gong, J. (2024, October). Surrealdriver: Designing llm-powered generative driver agent framework based on human drivers’ driving-thinking data. In 2024 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) (pp. 966-971). IEEE. Choi, Lee and Tu 18

    2024

  22. [22]

    Liu, C., Liu, G., Wang, Z., Yang, J., & Chen, S. (2025). CoLMDriver: LLM-based Negotiation Benefits Cooperative Autonomous Driving. arXiv preprint arXiv:2503.08683

    Pith/arXiv arXiv 2025

  23. [23]

    Chahe, A., & Zhou, L. (2025). ReasonDrive: Efficient Visual Question Answering for Autonomous Vehicles with Reasoning-Enhanced Small Vision-Language Models. In Proceedings of the Computer Vision and Pattern Recognition Conference (pp. 3870-3879)

    2025

  24. [24]

    (2024, September)

    Nie, M., Peng, R., Wang, C., Cai, X., Han, J., Xu, H., & Zhang, L. (2024, September). Reason2drive: Towards interpretable and chain-based reasoning for autonomous driving. In European Conference on Computer Vision (pp. 292-308). Cham: Springer Nature Switzerland

    2024

  25. [25]

    Sima, C., Renz, K., Chitta, K., Chen, L., Zhang, H., Xie, C., ... & Li, H. (2024, September). Drivelm: Driving with graph visual question answering. In European conference on computer vision (pp. 256- 274). Cham: Springer Nature Switzerland

    2024

  26. [26]

    M., Chen, L., Hünermann, J., Karnsund, A., Hanotte, B., Chidananda, P.,

    Marcu, A. M., Chen, L., Hünermann, J., Karnsund, A., Hanotte, B., Chidananda, P., ... & Sinavski, O. (2024, September). Lingoqa: Visual question answering for autonomous driving. In European Conference on Computer Vision (pp. 252-269). Cham: Springer Nature Switzerland

    2024

  27. [27]

    Qian, T., Chen, J., Zhuo, L., Jiao, Y., & Jiang, Y. G. (2024, March). Nuscenes-qa: A multi-modal visual question answering benchmark for autonomous driving scenario. In Proceedings of the AAAI Conference on Artificial Intelligence (Vol. 38, No. 5, pp. 4542-4550)

    2024

  28. [28]

    Y., Cui, C., Ma, Y., Moradipari, A., Gupta, R., Han, K., & Wang, Z

    Park, S. Y., Cui, C., Ma, Y., Moradipari, A., Gupta, R., Han, K., & Wang, Z. (2025). Nuplanqa: A large-scale dataset and benchmark for multi-view driving scene understanding in multi-modal large language models. arXiv preprint arXiv:2503.12772

    Pith/arXiv arXiv 2025

  29. [29]

    Wang, Y., Xing, S., Can, C., Li, R., Hua, H., Tian, K., ... & Tu, Z. (2025). Generative ai for autonomous driving: Frontiers and opportunities. arXiv preprint arXiv:2505.08854

    Pith/arXiv arXiv 2025

  30. [30]

    Jiang, P., Ergu, D., Liu, F., Cai, Y., & Ma, B. (2022). A Review of Yolo algorithm developments. Procedia computer science, 199, 1066-1073

    2022

  31. [31]

    M., & Romero-González, J

    Terven, J., Córdova-Esparza, D. M., & Romero-González, J. A. (2023). A comprehensive review of yolo architectures in computer vision: From yolov1 to yolov8 and yolo-nas. Machine learning and knowledge extraction, 5(4), 1680-1716

    2023

  32. [32]

    H., Huang, N

    Hsu, H. H., Huang, N. F., & Han, C. H. (2020, June). Collision analysis to motor dashcam videos with yolo and mask r-cnn for auto insurance. In 2020 International Conference on Intelligent Engineering and Management (ICIEM) (pp. 311-315). IEEE

    2020

  33. [33]

    OpenAI. (2024). ChatGPT-4o. https://openai.com/chatgpt

    2024

  34. [34]

    Anthropic. (2024). Claude 3. https://www.anthropic.com/index/introducing-claude

    2024

  35. [35]

    Google. (2024). Gemini 2.5 by Google DeepMind. https://deepmind.google/technologies/gemini/

    2024

  36. [36]

    Microsoft. (2024). Microsoft Copilot. https://copilot.microsoft.com/

    2024

  37. [37]

    National Highway Traffic Safety Administration. (2013). Traffic safety facts 2011: A compilation of motor vehicle crash data from the Fatality Analysis Reporting System and the General Estimates System (Report No. DOT HS 811 754). U.S. Department of Transportation. http://www- nrd.nhtsa.dot.gov/Pubs/811754AR.pdf Choi, Lee and Tu 19

    2013

  38. [38]

    Yun, H., & Lee, E. H. (2025). Party politics in transport policy with a large language model. Transport Policy, 171, 487-496

    2025

  39. [39]

    H., & Lee, E

    Lee, E. H., & Lee, E. (2026). Electric vehicle charging station location selection using generative artificial intelligence. Transportation Research Part E: Logistics and Transportation Review, 213, 104930. https://doi.org/10.1016/j.tre.2026.104930

    arXiv 2026

  40. [40]

    H., Moon, S., Cho, S

    Lee, E. H., Moon, S., Cho, S. H., & Lee, H. (2026). Passenger to train assignment using only smart card data. Transportation Research Record. Advance online publication. https://doi.org/10.1177/03611981251407915

    work page doi:10.1177/03611981251407915 2026

  41. [41]

    Lee, E. H. (2025). Understanding gender gap in bike-sharing services via explainable artificial intelligence. Transportation Research Record, 2679(8). https://doi.org/10.1177/03611981251335900

    work page doi:10.1177/03611981251335900 2025

  42. [42]

    Lee, E. H. (2024). eXplainable DEA approach for evaluating performance of public transport origin– destination pairs. Research in Transportation Economics, 108, 101491

    2024

  43. [43]

    Lee, E. H. (2023). Traffic speed prediction of urban road network based on high-importance links using XGB and SHAP. IEEE Access, 11, 113217–113226

    2023

  44. [44]

    H., Ham, S

    Min, J. H., Ham, S. W., Kim, D. K., & Lee, E. H. (2023). Deep multimodal learning for traffic speed estimation combining dedicated short-range communication and vehicle detection system data. Transportation Research Record, 2677(5), 247–259

    2023

  45. [45]

    H., & Lee, E

    Lee, E. H., & Lee, E. (2024). Congestion boundary approach for phase transitions in traffic flow. Transportmetrica B: Transport Dynamics, 12(1), 2379377

    2024

  46. [46]

    Kwak, K., & Lee, E. H. (2024). Impact of road transport system on groundwater quality inferred from explainable artificial intelligence (XAI). Science of the Total Environment, 917, 170388

    2024