Pith. sign in

REVIEW 4 major objections 6 minor 51 references

Vision-Integrated LLMs for Autonomous Driving Assistance : Human Performance Comparison and Trust Evaluation

T0 review · 4 major / 6 minor · reviewed 2026-08-08 · deepseek-v4-flash

Pith's one-line read A vision-integrated LLM driving assistant can describe crash scenes nearly as well as human drivers, and a single interaction raises drivers' trust in automation.

desk verdict A useful evaluation template trapped in a paper that never proves its own system produced the outputs it compares to humans. read the letter →

arxiv 2502.06843 v1 pith:3FV2OJTO submitted 2025-02-06 cs.CV cs.AIcs.HC

classification cs.CVcs.AIcs.HC
keywords autonomousdrivinglargelanguagemodelsvision-languagemodelspatialreasoninghuman-AIsimilaritytrustinautomationGPT-4YOLOv4
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

This paper sets out to show that a driving-assistance system built by pairing a vision module with a large language model can describe unexpected road situations about as well as experienced human drivers, and that seeing the system's outputs makes drivers trust automation more. The pipeline uses YOLOv4 with a Vision Transformer to extract objects and spatial layout, projects those features into GPT-4's embedding space, and lets GPT-4 generate situation descriptions and suggested responses. In a study of 45 experienced drivers, expert raters judged the system's scene descriptions highly similar to human descriptions (4.20/5) and its responses moderately similar (3.38/5); reported trust rose from 50.70 to 59.97 on the Trust in Automation scale after one interaction (t(44)=5.030, p<.001). If these numbers hold, they suggest that relatively simple vision-to-language integration can already make autonomous-driving assistance legible and acceptable to human drivers, while leaving open whether the system is truly reasoning about scenes or producing fluent, standardized text.

What carries the argument

The load-bearing mechanism is the vision adapter plus the linear projection into the language model's embedding space: YOLOv4 detects objects on a grid, ViT splits the image into patches and encodes relationships between them, and a linear projection layer is said to map the resulting visual features into GPT-4's embedding space so the language model can reason about the scene. GPT-4, run with temperature 0.7, then produces the situation description and recommended actions. On the evaluation side, the claims are carried by the Trust in Automation scale for trust change, expert semantic-similarity ratings for human-likeness, and METEOR/BERT scores for textual alignment, which together turn 'is this like what a human would say?' into quantitative outcomes.

What would settle it

Re-run the experiment with the visual pathway disabled: feed GPT-4 only the YOLO detection labels, or a constant vector, and check whether the situation descriptions and trust ratings stay the same. If they do, the claimed vision-to-language alignment is not doing the work; if they collapse, the adapter is necessary. A second check is to try to reproduce the pipeline from the method section—without a loss function for the linear projection or paired image-text data, the described system may not be implementable as specified.

Watch

Extended reading notes

Core claim

The paper's central claim is that the proposed system—a vision adapter combining YOLOv4 and ViT, a linear projection layer, and GPT-4 as the reasoning module—closely mirrors human performance in describing situations and moderately aligns with human decisions in generating appropriate responses. The evidence has two strands: semantic similarity between AI and human text, with expert-rated similarity of 4.20/5 for situation descriptions and 3.38/5 for appropriate responses, alongside METEOR scores near 0.75 and BERT scores near 0.74; and a pre-post Trust in Automation measurement in which mean trust rose from 50.70 to 59.97 (t(44)=5.030, p<.001, Cohen's d=0.750). The authors frame the contribution as evidence that a vision-integrated LLM can augment rather than replicate human decision-making, with the system strongest at structured perception tasks and weaker at open-ended, experience-dependent response generation.

Load-bearing premise

The whole demonstration rests on the assumption that a linear projection layer can align YOLOv4/ViT visual features with GPT-4's embedding space well enough for GPT-4 to reason about the scene, but the paper states this alignment without describing how it was trained, on what data, or with what objective.

Editorial extensions

If this is right

  • If the similarity scores generalize, a practical assistance system could narrate hazards to drivers in human-like language rather than emitting raw alerts, because perception is the part that already matches human descriptions.
  • The moderate alignment on responses (3.38/5) implies the assistant should be a recommender with the driver in the loop, not an autonomous decision-maker, in novel or high-stakes situations.
  • The trust increase of 9.27 points on the TiA scale after a single exposure suggests that explaining an AI's reasoning in natural language is a viable route to improving acceptance of driving automation.
  • Because descriptions scored higher than responses, near-term deployments should emphasize perception-and-explanation features and treat driving-action suggestions as draft advisories that need human confirmation.

Reading between the lines

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

  • A direct ablation would test whether the vision adapter earns its place: feed GPT-4 only the YOLO detection labels, or a constant input, and see how much of the 4.20 similarity score remains; the paper reports no such comparison.
  • The trust measurement captures one scripted exposure to three scenarios, so whether the gain persists, generalizes to unfamiliar roads, or survives a system mistake is a separate question the paper leaves open.
  • The METEOR and BERT scores measure fluency and word-overlap with human references, not factual correctness, so a fluent but wrong description could score well; a next study could score object counts and spatial claims against ground truth.
  • Because the linear projection is asserted rather than specified, a replication would need a concrete training objective and paired image-text data before the reported similarity and trust numbers can be attributed to the architecture.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. The paper proposes a vision-integrated LLM-based autonomous driving assistance system that combines a YOLOv4/ViT vision adapter with GPT-4 through a linear projection layer. The authors report an evaluation with 45 experienced drivers: expert ratings of semantic similarity between AI-generated and human-written situation descriptions and appropriate responses, plus METEOR/BERT scores and a pre-post Trust in Automation (TiA) assessment. The central claims are that the system closely mirrors human performance in describing situations, moderately aligns with human decisions in generating responses, and that interaction with the system significantly increases user trust.

Significance. If the central claims were properly evidenced, the work would be a useful contribution to human-centered evaluation of vision-language models for driving assistance, combining objective text-similarity metrics, human expert ratings, and a standardized trust instrument. The use of 45 experienced drivers and the TiA scale are strengths, as is the attempt to compare AI outputs against human reasoning rather than only against ground-truth labels. However, the significance is conditional on establishing that the reported outputs actually came from the described architecture and that the similarity and trust measurements are properly calibrated; as written, the evidence for these claims is incomplete.

major comments (4)
  1. [Section II-A, II-B, Table V] The manuscript does not establish that the outputs evaluated in Section IV-C were generated by the described YOLOv4/ViT-to-linear-projection-to-GPT-4 pipeline. Section II-A states that visual features are 'aligned with the LLM's embedding space using a linear projection layer' but supplies no training objective, no paired image-text data, no prompt template, and no sanity check. Section II-B repeats the alignment claim without further detail. Without example raw outputs, code, or a reproducible procedure, Table V's narrative examples cannot be verified as system outputs. The author(s) must provide implementation details, sample outputs generated by the full pipeline, and an experiment demonstrating that the linear projection enables GPT-4 to perform spatial reasoning.
  2. [Section IV-C, Table III] The expert similarity ratings lack a human-human baseline and inter-rater reliability statistics. Two experts produced 540 scores, but no agreement measure (e.g., Cohen's kappa or ICC) is reported, and the mean similarity of 4.20 for situation descriptions is not calibrated against how similar two humans would be on the same task. Without a human-human baseline, the abstract's claim that the system 'closely mirrors human performance' is not supported. The authors should add a human-human similarity condition and report rater agreement.
  3. [Section IV-D, Tables VI-VIII] The trust increase (pre 50.70 to post 59.97, t(44)=5.03, p<.001) cannot be attributed to the system because there is no control condition. Any interaction with a plausible driving-assistance output could raise trust through demand characteristics, exposure, or a general positive attitude toward automation. A control group that does not receive the system's outputs, or that receives a non-AI baseline intervention, is necessary to support the claim that the system itself increased trust. Additionally, because the system outputs were not shown to originate from the proposed architecture (see first major comment), the trust stimulus is inadequately specified.
  4. [Section IV-C, Table IV] The METEOR and BERT scores are reported without specifying the reference texts. The text says these metrics evaluate 'semantic similarity for AI-generated text,' but METEOR and BERT scores require a reference (e.g., human responses). If the references are the participants' responses, the construction of the reference set must be stated; if the references are a separate gold standard, that is a different claim. This ambiguity affects the interpretation of Table IV and should be clarified in revision.
minor comments (6)
  1. [Section III-A, III-B] The TiA scale is described inconsistently: Section III-A says it measures six components, while Section III-B says it consists of 19 items covering five dimensions. Please reconcile these descriptions with the cited original scale.
  2. [References] Several references appear miscited or contain questionable DOIs. For example, reference [40] is YOLOv3, not YOLOv4; reference [52] is YOLOP, not BDD100k; and reference [41] has a DOI (10.1109/ACCESS.2024.1234567) that looks like a placeholder. The reference list should be thoroughly verified.
  3. [Table II] The precision, recall, and F1-score in Table II are reported without the number of test images or detected objects, making it impossible to interpret the results. Please provide the evaluation set size and any confidence intervals.
  4. [Section II-B] Training details are incomplete: the batch size, number of epochs, and the exact fine-tuning procedure for ViT on BDD100k are not given. Adding these would improve reproducibility.
  5. [Section II-A] The prompt template or the exact input format to GPT-4 is not described, which is essential for reproducing the reported outputs. Please include the prompt and any interface details.
  6. [Section IV-D] The paper states that the trust increase corresponds to '9.27 percentage points,' but the TiA scale is scored in raw points, not percentages. Please clarify the interpretation or report the possible range of the scale.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the human-similarity and trust measurements are external to the model's construction, and no fitted parameter is renamed as a prediction.

full rationale

The paper's central empirical claims are that the vision-integrated GPT-4 system produces situation descriptions rated similar to human descriptions (M=4.20 on a 5-point scale), appropriate responses moderately similar (M=3.38), and that TiA trust scores increased after exposure (t(44)=5.030, p<.001). None of these quantities is fed back into the model. The vision adapter metrics (precision 89.5%, recall 91.2%, F1 90.3%) are standard object-detection measures computed independently of the human-AI similarity ratings and trust scores. METEOR and BERT scores are external text-similarity metrics, not losses used to train or fit the LLM, and the expert Likert ratings are independent human judgments. Trust is measured with the validated TiA scale before and after exposure; although the design lacks a control condition, this is a causal-inference limitation rather than a circular derivation. The paper's principal weakness is an attribution gap: Section II-A/B describes the linear projection alignment of YOLOv4/ViT features with GPT-4 without supplying a training objective, paired data, or a sanity check, so the reader cannot verify that the quoted AI outputs actually came from the described architecture. That is a reproducibility or verification problem, not a case where the prediction is equivalent to an input by construction. There are no self-citations used as load-bearing evidence, no imported uniqueness theorem, and no fitted parameter repackaged as a prediction. Accordingly, the circularity score is 0.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

The system is an assembly of existing models, so there are no invented entities. The central claim rests on several unverified assumptions: BDD100k generalizes to the test scenarios, expert similarity ratings are valid without a human-human baseline, the TiA scale is used consistently, and the linear projection into GPT-4's embedding space enables spatial reasoning. The GPT-4 temperature is a hand-chosen parameter that affects output variability and hence similarity scores.

free parameters (3)
  • GPT-4 temperature = 0.7
    Chosen by hand in Section II-A to balance creativity and precision; it directly affects response variability and therefore the similarity scores.
  • Train/validation split = 80/20
    Used in Section II-B for BDD100k training; no sensitivity analysis or separate test set is reported.
  • Learning rate = 0.001
    Adam optimizer setting in Section II-B; the reported vision adapter metrics are not shown to be robust to this choice.
assumptions (5)
  • domain assumption BDD100k is representative of diverse real-world driving scenarios and supports generalization to unexpected situations.
    Stated in Section II-A; the three emergency test scenarios are assumed to be covered by this dataset, but the paper does not show that the test images come from a distribution similar to the training set.
  • domain assumption Expert Likert ratings of semantic similarity are valid and sufficient measures of human-AI alignment.
    Section IV-A2 and IV-C rely on two experts' 1-5 scores without reporting inter-rater reliability or a human-human agreement baseline.
  • domain assumption The Trust in Automation (TiA) scale measures trust consistently before and after exposure.
    Section III-A/B uses the TiA scale, but the paper gives inconsistent descriptions of its components (six vs five) and item count, raising construct validity concerns.
  • ad hoc to paper A linear projection of YOLOv4/ViT features into GPT-4's embedding space is sufficient for GPT-4 to perform spatial reasoning.
    Section II-A/B asserts this alignment without giving a training loss, paired data, or empirical validation that the projected features are actually usable by GPT-4.
  • ad hoc to paper Temperature 0.7 produces sufficiently consistent outputs for a fair comparison with human responses.
    The temperature is set in Section II-A; no repeated sampling or variance analysis is reported, so the stability of the AI outputs is unknown.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Vision-Integrated LLMs for Autonomous Driving Assistance : Human Performance Comparison and Trust Evaluation." pith.science (2026). https://pith.science/paper/3FV2OJTO

@misc{pith2026250206843,
  author       = {Pith},
  title        = {Pith review of: Vision-Integrated LLMs for Autonomous Driving Assistance : Human Performance Comparison and Trust Evaluation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3FV2OJTO}},
  note         = {Machine review of arXiv:2502.06843}
}
read the original abstract

Traditional autonomous driving systems often struggle with reasoning in complex, unexpected scenarios due to limited comprehension of spatial relationships. In response, this study introduces a Large Language Model (LLM)-based Autonomous Driving (AD) assistance system that integrates a vision adapter and an LLM reasoning module to enhance visual understanding and decision-making. The vision adapter, combining YOLOv4 and Vision Transformer (ViT), extracts comprehensive visual features, while GPT-4 enables human-like spatial reasoning and response generation. Experimental evaluations with 45 experienced drivers revealed that the system closely mirrors human performance in describing situations and moderately aligns with human decisions in generating appropriate responses.

Figures

Figures reproduced from arXiv: 2502.06843 by the authors.

Figure 1
Figure 1. Model Response Results by Cases [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. visualizes these results, emphasizing the AI system’s [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

51 extracted references · 26 canonical work pages

  1. [1]

    Survey of deep reinforcement learning for motion planning of autonomous vehicles,

    Aradi, S., “Survey of deep reinforcement learning for motion planning of autonomous vehicles,” IEEE Transactions on Intelligent Transportation Systems, vol. 22, no. 2, Feb. 2020, pp. 464–478. doi: 10.1109/TITS.2020.3041234

  2. [2]

    Three decades of driver assistance systems: Review and future perspectives,

    Bengler, K., Dietmayer, K., Färber, B., Maurer, M., Stiller, C., & Winner, H., “Three decades of driver assistance systems: Review and future perspectives,” IEEE Intelligent Transportation Systems Magazine, vol. 6, no. 4, Winter 2014, pp. 6–22. doi: 10.1109/MITS.2014.2357643

  3. [3]

    End to end learning for self-driving cars,

    Bojarski, M., Del Testa, D., Dworakowski, D., Firner, B., Flepp, B., Goyal, P., ... & Zieba, K., “End to end learning for self-driving cars,” arXiv preprint, arXiv:1604.07316,

  4. [4]

    On the opportunities and risks of foundation models,

    Bommasani, R., Hudson, D. A., Adeli, E., Altman, R., Arora, S., von Arx, S., ... & Liang, P., “On the opportunities and risks of foundation models,” arXiv preprint, arXiv:2108.07258,

  5. [5]

    Language models are few-shot learners,

    Brown, T. B., Mann, B., Ryder, N., Subbiah, M., Kaplan, J., Dhariwal, P., ... & Amodei, D., “Language models are few-shot learners,” arXiv preprint, arXiv:2005.14165,

  6. [6]

    DQ-GAT: Towards safe and efficient autonomous driving with deep Q-learning and graph attention networks,

    Cai, P., Wang, H., Sun, Y., & Liu, M., “DQ-GAT: Towards safe and efficient autonomous driving with deep Q-learning and graph attention networks,” IEEE Transactions on Intelligent Transportation Systems, vol. 23, no. 11, Nov. 2022, pp. 21102–21112. doi: 10.1109/TITS.2022.3181112

  7. [7]

    DeepDriving: Learning affordance for direct perception in autonomous driving,

    Chen, C., Seff, A., Kornhauser, A., & Xiao, J., “DeepDriving: Learning affordance for direct perception in autonomous driving,” in Proc. IEEE Int. Conf. Comput. Vis., Santiago, Chile, 2015, pp. 2722–2730. doi: 10.1109/ICCV.2015.312

  8. [8]

    Interpretable end-to-end urban autonomous driving with latent deep reinforcement learning,

    Chen, J., Li, S. E., & Tomizuka, M., “Interpretable end-to-end urban autonomous driving with latent deep reinforcement learning,” IEEE Transactions on Intelligent Transportation Systems, vol. 23, no. 6, Jun. 2021, pp. 5068–5078. doi: 10.1109/TITS.2021.3081112

Show all 51 references
  1. [9]

    Driving with LLMs: Fusing object-level vector modality for explainable autonomous driving,

    Chen, L., Sinavski, O., Hünermann, J., Karnsund, A., Willmott, A. J., Birch, D., ... & Shotton, J., “Driving with LLMs: Fusing object-level vector modality for explainable autonomous driving,” in 2024 IEEE Int. Conf. Robotics Autom. (ICRA), Yokohama, Japan, May 2024, pp. 14093...

  2. [10]

    Multi-task learning for multimodal data fusion in 3D object detection,

    Cheng, Y., Yin, X., Shen, F., Tao, D., & Li, X., “Multi-task learning for multimodal data fusion in 3D object detection,” IEEE Transactions on Circuits and Systems for Video Technology, vol. 30, no. 4, Apr. 2020, pp. 966–978. doi: 10.1109/TCSVT.2020.2973942

  3. [11]

    A review on safety failures, security attacks, and available countermeasures for autonomous vehicles,

    Cui, J., Liew, L. S., Sabaliauskaite, G., & Zhou, F., “A review on safety failures, security attacks, and available countermeasures for autonomous vehicles,” Ad Hoc Networks, vol. 90, Nov. 2019, Art. no. 101823. doi: 10.1016/j.adhoc.2019.101823

  4. [12]

    Explaining autonomous driving with visual attention and end-to-end trainable region proposals,

    Cultrera, L., Becattini, F., Seidenari, L., Pala, P., & Del Bimbo, A., “Explaining autonomous driving with visual attention and end-to-end trainable region proposals,” Journal of Ambient Intelligence and Humanized Computing, Jan. 2023, pp. 1–13. doi: 10.1007/s12652-023-04090-x

  5. [13]

    An image is worth 16x16 words: Transformers for image recognition at scale,

    Dosovitskiy, A., Beyer, L., Kolesnikov, A., Weissenborn, D., Zhai, X., Unterthiner, T., ... & Houlsby, N., “An image is worth 16x16 words: Transformers for image recognition at scale,” arXiv preprint, arXiv:2010.11929,

  6. [14]

    The Pascal Visual Object Classes Challenge: A retrospective,

    Everingham, M., Gool, L. V., Williams, C. K., Winn, J., & Zisserman, A., “The Pascal Visual Object Classes Challenge: A retrospective,” International Journal of Computer Vision, vol. 111, no. 1, Jan. 2015, pp. 98–136. doi: 10.1007/s11263-014-0733-5

  7. [15]

    Preparing a nation for autonomous vehicles: Opportunities, barriers and policy recommendations,

    Fagnant, D. J., & Kockelman, K., “Preparing a nation for autonomous vehicles: Opportunities, barriers and policy recommendations,” Transportation Research Part A: Policy and Practice, vol. 77, Jul. 2015, pp. 167–181. doi: 10.1016/j.tra.2015.04.003

  8. [17]

    Learning from long-tail data with no bias towards head categories,

    Gao, Y., Guo, Y., & Zhu, Z., “Learning from long-tail data with no bias towards head categories,” IEEE Transactions on Neural Networks and Learning Systems, vol. 30, no. 5, May 2019, pp. 1489–1500. doi: 10.1109/TNNLS.2019.2903926

  9. [18]

    Are we ready for autonomous driving? The KITTI vision benchmark suite,

    Geiger, A., Lenz, P., & Urtasun, R., “Are we ready for autonomous driving? The KITTI vision benchmark suite,” in Proc. IEEE Conf. Comput. Vis. Pattern Recognit., Portland, OR, USA, 2013, pp. 3354–3361. doi: 10.1109/CVPR.2013.424

  10. [19]

    Machine ethics and automated vehicles,

    Goodall, N. J., “Machine ethics and automated vehicles,” in Road Vehicle Automation, Cham, Switzerland: Springer, 2014, pp. 93–102. doi: 10.1007/978-3-319-05990-7_9

  11. [20]

    Long-tail visual recognition by learning from frequent and rare classes via selective joint fine-tuning,

    Guo, Y., Lei, A., Liu, H., Zhu, Z., & He, X., “Long-tail visual recognition by learning from frequent and rare classes via selective joint fine-tuning,” in Proc. ACM Multimedia Conf., Seoul, South Korea, 2018, pp. 57–65. doi: 10.1145/3240508.3240537

  12. [21]

    Interpretation and robustness in autonomous driving,

    Hawkins, A. J., Shah, J. A., & Abbeel, P., “Interpretation and robustness in autonomous driving,” IEEE Transactions on Neural Networks and Learning Systems, vol. 31, no. 9, Sept. 2020, pp. 3478–3492. doi: 10.1109/TNNLS.2020.2995290

  13. [22]

    Toward trustworthy decision-making for autonomous vehicles: A robust reinforcement learning approach with safety guarantees,

    He, X., Huang, W., & Lv, C., “Toward trustworthy decision-making for autonomous vehicles: A robust reinforcement learning approach with safety guarantees,” Engineering, vol. 33, Jan. 2024, pp. 77–89. doi: 10.1016/j.eng.2023.06.012

  14. [23]

    Computer vision for autonomous vehicles: Problems, datasets and state of the art,

    Janai, J., Güney, F., Behl, A., & Geiger, A., “Computer vision for autonomous vehicles: Problems, datasets and state of the art,” Foundations and Trends in Computer Graphics and Vision, vol. 12, nos. 1–3, 2020, pp. 1–308. doi: 10.1561/0600000076

  15. [24]

    Autonomous land vehicle project at CMU,

    Kanade, T., Thorpe, C., & Whittaker, W., “Autonomous land vehicle project at CMU,” in Proc. 1986 ACM Fourteenth Annu. Conf. Comput. Sci., Cincinnati, OH, USA, 1986, pp. 71–80. doi: 10.1145/1006440.1006453

  16. [25]

    Learning to drive in a day,

    Kendall, A., Hawke, J., Janz, D., Mazur, P., Reda, D., Allen, J.-M., ... Shah, A., “Learning to drive in a day,” in Proc. 2019 Int. Conf. Robotics Autom., Montreal, Canada, 2019, pp. 8248–8254. doi: 10.1109/ICRA.2019.8793742

  17. [26]

    Long-tail distribution adaptation,

    Kim, J., Lee, J., & Kim, J., “Long-tail distribution adaptation,” in Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit., Seattle, WA, USA, 2020, pp. 14334–14343. doi: 10.1109/CVPR.2020.01434

  18. [27]

    Deep reinforcement learning for autonomous driving: A survey,

    Kiran, B. R., Sobh, I., Talpaert, V., Mannion, P., Al Sallab, A. A., Yogamani, S., & Pérez, P., “Deep reinforcement learning for autonomous driving: A survey,” IEEE Transactions on Intelligent Transportation Systems, vol. 23, no. 6, Jun. 2022, pp. 4909–4926. doi: 10.1109/TITS....

  19. [29]

    Pointpillars: Fast encoders for object detection from point clouds,

    Lang, A. H., Vora, S., Caesar, H., Zhou, L., Yang, J., & Beijbom, O., “Pointpillars: Fast encoders for object detection from point clouds,” IEEE Transactions on Pattern Analysis and Machine Intelligence, Oct. 2019, doi: 10.1109/TPAMI.2019.2932102

  20. [30]

    Towards fully autonomous driving: Systems and algorithms,

    Levinson, J., Askeland, J., Becker, J., Dolson, J., Held, D., Kammel, S., ... & Thrun, S., “Towards fully autonomous driving: Systems and algorithms,” in Proc. 2011 IEEE Intell. Vehicles Symp., Baden-Baden, Germany, 2011, pp. 163–168. doi: 10.1109/IVS.2011.5940536

  21. [31]

    Blip-2: Bootstrapping language-image pre-training with frozen image encoders and large language models,

    Li, J., Li, D., Savarese, S., & Hoi, S., “Blip-2: Bootstrapping language-image pre-training with frozen image encoders and large language models,” in Proc. Int. Conf. Mach. Learn., Honolulu, HI, USA, Jul. 2023, pp. 19730–19742. doi: 10.48550/arXiv.2301.02111

  22. [32]

    Autonomous vehicle implementation predictions: Implications for transport planning,

    Litman, T., “Autonomous vehicle implementation predictions: Implications for transport planning,” Victoria Transport Policy Institute, Victoria, Canada, Rep. AVIP-2020,

  23. [34]

    Deep learning: A critical appraisal,

    Marcus, G., “Deep learning: A critical appraisal,” arXiv preprint, arXiv:1801.00631,

  24. [37]

    ALVINN: An autonomous land vehicle in a neural network,

    Pomerleau, D. A., “ALVINN: An autonomous land vehicle in a neural network,” in Adv. Neural Inf. Process. Syst., vol. 1, 1988, pp. 305–313. doi: 10.5555/2986293.2986348

  25. [38]

    PointNet: Deep learning on point sets for 3D classification and segmentation,

    Qi, C. R., Su, H., Mo, K., & Guibas, L. J., “PointNet: Deep learning on point sets for 3D classification and segmentation,” in Proc. IEEE Conf. Comput. Vis. Pattern Recognit., Honolulu, HI, USA, 2017, pp. 652–660. doi: 10.1109/CVPR.2017.660

  26. [39]

    Learning transferable visual models from natural language supervision,

    Radford, A., Wu, J., Child, R., Luan, D., Amodei, D., & Sutskever, I., “Learning transferable visual models from natural language supervision,” in Proc. Int. Conf. Mach. Learn., Vienna, Austria, 2021, pp. 8748–8763. doi: 10.48550/arXiv.2103.00020

  27. [40]

    YOLOv3: An incremental improvement,

    Redmon, J., & Farhadi, A., “YOLOv3: An incremental improvement,” arXiv preprint, arXiv:1804.02767,

  28. [42]

    Multimodal sensor fusion in uncertain environments for self-driving cars,

    Schneider, P., Kümmerle, J., & Burgard, W., “Multimodal sensor fusion in uncertain environments for self-driving cars,” Journal of Field Robotics, vol. 34, no. 6, Sep. 2017, pp. 1263–1288. doi: 10.1002/rob.21705

  29. [43]

    LanguageMPC: Large language models as decision makers for autonomous driving,

    Sha, H., Mu, Y., Jiang, Y., Chen, L., Xu, C., Luo, P., Li, S. E., Tomizuka, M., Zhan, W., & Ding, M., “LanguageMPC: Large language models as decision makers for autonomous driving,” arXiv preprint, arXiv:2310.03026,

  30. [44]

    Lmdrive: Closed-loop end-to-end driving with large language models,

    Shao, H., Hu, Y., Wang, L., Song, G., Waslander, S. L., Liu, Y., & Li, H., “Lmdrive: Closed-loop end-to-end driving with large language models,” in Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit., Vancouver, BC, Canada, 2024, pp. 15120–15130. doi: 10.1109/CVPR.2024.3456123

  31. [45]

    Lm-nav: Robotic navigation with large pre-trained models of language, vision, and action,

    Shah, D., Osiński, B., & Levine, S., “Lm-nav: Robotic navigation with large pre-trained models of language, vision, and action,” in Conf. Robot Learn., San Francisco, CA, USA, Mar. 2023, pp. 492–504. doi: 10.48550/arXiv.2301.12345

  32. [46]

    Drivelm: Driving with graph visual question answering,

    Sima, C., Renz, K., Chitta, K., Chen, L., Zhang, H., Xie, C., ... & Li, H., “Drivelm: Driving with graph visual question answering,” in Proc. Eur. Conf. Comput. Vis., Cham, Switzerland: Springer, 2025, pp. 256–274. doi: 10.1007/12345-2025-12345

  33. [47]

    Scalability in perception for autonomous driving: Waymo open dataset,

    Sun, P., Kretzschmar, H., Dotiwalla, X., Chouard, T., Patnaik, V., Tsui, P., ... & Zeng, W., “Scalability in perception for autonomous driving: Waymo open dataset,” in Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit., Seattle, WA, USA, 2020, pp. 2446–2454. doi: 10.1109/CVPR...

  34. [49]

    Drivemlm: Aligning multi-modal large language models with behavioral planning states for autonomous driving,

    Wang, W., Xie, J., Hu, C., Zou, H., Fan, J., Tong, W., ... & Dai, J., “Drivemlm: Aligning multi-modal large language models with behavioral planning states for autonomous driving,” arXiv preprint, arXiv:2312.09245,

  35. [51]

    Dilu: A knowledge-driven approach to autonomous driving with large language models,

    Wen, L., Fu, D., Li, X., Cai, X., Ma, T., Cai, P., ... & Qiao, Y., “Dilu: A knowledge-driven approach to autonomous driving with large language models,” arXiv preprint, arXiv:2309.16292,

  36. [52]

    Yolop: You only look once for panoptic driving perception,

    Wu, D., Liao, M. W., Zhang, W. T., Wang, X. G., Bai, X., Cheng, W. Q., & Liu, W. Y., “Yolop: You only look once for panoptic driving perception,” Machine Intelligence Research, vol. 19, no. 6, 2022, pp. 550–562. doi: 10.1007/2022-12345

  37. [54]

    Rag-driver: Generalisable driving explanations with retrieval-augmented in-context learning in multi-modal large language model,

    Yuan, J., Sun, S., Omeiza, D., Zhao, B., Newman, P., Kunze, L., & Gadd, M., “Rag-driver: Generalisable driving explanations with retrieval-augmented in-context learning in multi-modal large language model,” arXiv preprint, arXiv:2402.10828,

  38. [55]

    Explainability of deep vision-based autonomous driving systems: Review and challenges,

    Zablocki, É., Ben-Younes, H., Pérez, P., & Cord, M., “Explainability of deep vision-based autonomous driving systems: Review and challenges,” International Journal of Computer Vision, vol. 130, no. 10, Oct. 2022, pp. 2425–2452. doi: 10.1007/s11263-022-01610-y

  39. [56]

    Minigpt-4: Enhancing vision-language understanding with advanced large language models,

    Zhu, D., Chen, J., Shen, X., Li, X., & Elhoseiny, M., “Minigpt-4: Enhancing vision-language understanding with advanced large language models,” arXiv preprint, arXiv:2304.10592,

  40. [57]

    Human-like autonomous car-following model with deep reinforcement learning,

    Zhu, M., Wang, X., & Wang, Y., “Human-like autonomous car-following model with deep reinforcement learning,” Transportation Research Part C: Emerging Technologies, vol. 97, Jul. 2018, pp. 348–368. doi: 10.1016/j.trc.2018.10.003

  41. [58]

    Cross-domain long-tail recognition via dynamic class prototype,

    Zhu, Z., Liang, C., Zhang, J., & Zhang, C., “Cross-domain long-tail recognition via dynamic class prototype,” in Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit., Long Beach, CA, USA, 2019, pp. 9443–9452. doi: 10.1109/CVPR.2019.00966. Namhee Kim received the B.S. degree in ...

  42. [2022]

    degree in industrial engineering at Seoul National University, Seoul, South Korea

    She is currently pursuing the M.S. degree in industrial engineering at Seoul National University, Seoul, South Korea. Her research interests include AI, user experience, and digital human modeling. Woojin Park received the B.S. and M.S. degrees in industrial engineering from t...

  43. [2024]

    doi: 10.1109/LRA.2024.1234567

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.