Pith. sign in

REVIEW 4 major objections 5 minor 30 references

ROAD: Responsibility-Oriented Reward Design for Reinforcement Learning in Autonomous Driving

T0 review · 4 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read Penalizing a simulated driving agent's crashes in proportion to the legal responsibility it would bear, instead of applying a fixed penalty to every collision, reduces the agent's primary-fault accidents and increases its success rate.

desk verdict Genuinely new pipeline, but the liability and accuracy claims are circular; the success-rate improvement is independent and directionally positive but under-powered. read the letter →

arxiv 2505.24317 v1 pith:FIRG5Y5G submitted 2025-05-30 cs.LG

Yongming Chen , Miner Chen , Liewen Liao , Mingyang Jiang , Xiang Zuo , Hengrui Zhang , Yuchen Xi , Songan Zhang This is my paper
classification cs.LG
keywords autonomousdrivingreinforcementlearningrewardfunctiondesignaccidentliabilitytrafficregulationsknowledgegraphretrieval-augmentedgenerationvision-languagemodel
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

Reinforcement learning agents that drive in simulation usually receive the same penalty for every collision. This paper argues that a crash should instead be punished in proportion to how much legal responsibility the agent would bear for it, and that responsibility should be predicted automatically from a single post-collision image. The authors build a knowledge graph from traffic-safety law, use a vision-language model with retrieval-augmented generation to label simulated accidents by liability, and train a small image classifier to supply those labels cheaply during policy learning. In intersection and roundabout scenarios, the responsibility-scaled crash penalty yields fewer collisions in which the agent is primarily at fault and higher overall success rates than a fixed crash penalty.

What carries the argument

The load-bearing object is a Traffic Regulation Knowledge Graph (TRKG) -- a graph whose nodes are driving scenes, applicable legal standards, and the accident-scene facts needed to apply those standards -- distilled from a traffic-safety law using a large language model. A vision-language model first identifies the driving scene from the collision image; retrieval-augmented generation pulls the relevant legal standard and required facts from the graph; the same VLM then extracts those facts from the image; and a language model assigns a responsibility level on a 0--10 scale. Those labels train a vision transformer that works offline, and its output is converted into the reward multiplier $\text{Accident\_liability\_ratio}$ in the crash-penalty term. Everything downstream -- the improved success rates, the lower primary-liability share, and the claimed behavioral changes such as yielding in left turns and roundabouts -- depends on this classifier's responsibility predictions being accurate.

What would settle it

Have licensed traffic-accident investigators independently assign responsibility levels to the same post-collision images and compare their labels with the paper's vision-language-model labels; if agreement is near chance, especially on cases requiring speed and direction facts that a still frame does not show, then the liability signal feeding the reward is not grounded.

Watch

Extended reading notes

Core claim

The paper's central claim is that replacing the fixed crash penalty with a liability-scaled penalty improves both safety and task performance: $r_{\text{crash}} = \lambda \times P_{\text{baseline}} \times \text{Accident\_liability\_ratio}$, where the liability ratio is $1$ for primary responsibility, $0.5$ for shared, and $0$ for secondary. Here primary means the ego vehicle bears the main fault, shared means roughly equal fault, and secondary means little or no fault. The agent is trained with a responsibility classifier that maps a post-collision image to one of these three levels, so a collision the agent did not cause costs nothing while one it caused costs the full baseline penalty. On the paper's experiments, the responsibility-scaled agent reaches a 73.2% success rate in the intersection scenario versus 65.0% for the fixed-penalty baseline, and its share of primary-responsibility crashes drops from 57.0% to 43.5%; in the roundabout scenario success rises from 42.8% to 54.0% and primary responsibility falls from 56.5% to 50.8%. The authors further claim that the retrieval-augmented, knowledge-graph-grounded labeling pipeline assigns responsibility more accurately than a plain vision-language model alone (about 79% versus 63--69% accuracy on the two test sets).

Load-bearing premise

The load-bearing premise is that the legal responsibility for a simulated crash can be correctly inferred from one post-collision image by a vision-language model, even though the image alone may not show facts such as each vehicle's speed or direction of travel, and no human legal expert labels are used to verify the inferred labels.

Editorial extensions

If this is right

  • A driving agent whose crash penalty is scaled by predicted responsibility should collide less often as the primary at-fault party, and should complete more trips, under the same simulation conditions.
  • Because the responsibility signal comes from an offline image classifier, the reward scheme adds no live API cost during policy training.
  • Since the penalty is zero for collisions in which the agent is only secondarily liable, the agent is not trained to avoid all contact, only contact it causes; in mixed traffic this could shift behavior toward confident, right-of-way-respecting maneuvers.
  • The same trained pipeline can label new accident images for new maps or traffic densities without re-running the language-model prompts, as long as the legal knowledge graph covers the scenario.

Reading between the lines

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

  • A direct comparison between the liability-scaled penalty and a uniformly larger fixed penalty would be needed to separate the effect of 'penalize crashes harder' from 'penalize in proportion to fault'; the paper only compares against one fixed baseline level.
  • The chosen numeric mapping (primary=1, shared=0.5, secondary=0) is an assumption; calibrating these weights to actual crash costs or treating them as hyperparameters might change the learned behaviors.
  • If the responsibility labels are noisy because scene details like speed and direction cannot be recovered from a single still image, training a classifier on those labels could bake in the same noise; human legal-expert labels on the same images would test this.
  • The mechanism could generalize to other social-norm domains, such as robot navigation in pedestrian crowds, where a responsibility-like attribution can be generated from observations.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The authors propose ROAD, a reward-shaping scheme for reinforcement learning in the MetaDrive simulator. A Traffic Regulation Knowledge Graph is constructed by prompting an LLM; a VLM with retrieval-augmented generation annotates post-collision images with primary/shared/secondary liability for the ego vehicle; a ViT classifier is trained on these labels; and the predicted liability ratio is inserted into the terminal crash penalty (Eq. 1). PPO agents are trained on intersection and roundabout maps and compared with a fixed crash-penalty baseline. The paper reports higher success rates and lower primary liability in Table 5, and reports that the RAG+VLM annotator beats plain GPT-4o in Table 6.

Significance. The central idea of anchoring crash penalties in traffic-law-based responsibility is timely, and the directionally positive success-rate differences are interesting. The authors also perform five repeated training runs per condition and report convergence curves, and the success-rate comparison itself is not circular. However, the two headline claims about responsibility-assignment accuracy and liability reduction are evaluated against labels produced by the same pipeline that generates the training signal, so the evidence is self-referential. If independently obtained legal-expert labels were introduced, the contribution could be meaningful, but as presented the main scientific claims are not supported.

major comments (4)
  1. [Section 4.3, Table 6] The responsibility labels are generated by the TRKG+VLM+RAG pipeline described in Section 3.2, and the same labels are used to train the ViT classifier, to score the accuracy in Table 6, and to produce the PR/SH/SE columns in Table 5. The classifier output also enters the reward through Eq. (1). This makes the accuracy and liability claims self-referential: any systematic bias in the VLM flows simultaneously into the training data, the reward signal, and the evaluation labels. Independent ground truth, e.g., labels from legal experts or from simulator state combined with coded law, is required before Table 6 and Table 5's responsibility distribution can be interpreted.
  2. [Table 5] Table 5 reports only aggregate percentages from the five runs per condition; no standard deviations, per-run values, confidence intervals, or significance tests are provided. The observed success-rate gaps (73.2 vs 65.0 for intersection, 54.0 vs 42.8 for roundabout) could be driven by one or two seeds. The authors should report the full distribution across the five runs and test whether the differences are statistically significant.
  3. [Table 2, Section 3.2] The 'Scene info extraction' prompt in Table 2 asks for the speed of each party at the time of the accident and the direction of travel, but a post-collision still image from MetaDrive does not contain pre-collision speeds, and direction may be ambiguous. Since responsibility assignment inherits these extracted facts, the single-image assumption is a concrete source of bias. This can be tested by comparing the extracted facts against MetaDrive's simulator ground truth (which knows true speeds and headings) and by adding a fact-extraction accuracy table.
  4. [Section 4.3, Table 4, Fig. 6] The classifier is the sole source of the liability ratio in Eq. (1), but its reliability is not established: the intersection validation set contains only 9 shared-liability and 32 secondary-liability samples, and the paper itself notes that validation loss fluctuates and even increases in later epochs, consistent with overfitting to the dominant primary-liability class. A balanced validation set, per-class precision/recall, and an analysis of how classifier errors propagate into the reward signal are needed.
minor comments (5)
  1. [Table 3] Values such as '1.8M\2.4M' and 'X\O' use backslashes where a slash or range notation would be clearer; also, the table header has a typo ('V alue').
  2. [Eq. (1), Section 4.4] The text says λ is 'a scaling factor ensuring consistency with the baseline,' but Table 3 sets λ=1 without any calibration or sensitivity analysis; please clarify how λ was chosen and whether the results are robust to it.
  3. [Table 5 caption] The abbreviations CR, TO, SU, PR, SH, SE are used without definitions in the caption; please define them and state how many episodes were evaluated per run.
  4. [References] Reference [29] conflates the PPO algorithm with Stable-Baselines3; please cite Stable-Baselines3 as its own reference and cite the original PPO paper accurately.
  5. [General] There are minor typographical issues throughout, e.g., 'AutonomousDriving' in the title, 'dist from the road center' in Eq. (3), and inconsistent spacing around equations.

Circularity Check

2 steps flagged · score 7.0 of 10

Responsibility-accuracy and liability-reduction claims are evaluated with labels produced by the same TRKG+VLM+RAG pipeline under test; no independent legal ground truth appears anywhere in the loop.

  1. self definitional [Section 3.2, Section 4.3, Section 4.5.3 (Table 6)]
    "Each incident was then annotated with responsibility labels following the procedure outlined in Section 3.2, resulting in a dataset of image–label pairs... We compared RAG-based responsibility assignment methods, as proposed in Section 3.2, and the baseline. For the baseline method, we use GPT-4o from Open AI [28]. Through experimentation, we obtained the accuracy, precision, recall and F1 score of these two methods, as shown in Table 6."

    Table 6 reports the assignment accuracy of the proposed RAG-based method against plain GPT-4o, but the ground-truth labels used for that accuracy computation were created by the same 'procedure outlined in Section 3.2' — i.e., by the proposed method itself. 'Ours' is the method, and its output is also the reference. Agreement between the method and itself (or between an augmented and unaugmented version of the same VLM family) is not an external accuracy measure; any systematic hallucination or bias in the TRKG+VLM+RAG pipeline is shared by the labels and the predictions.

  2. fitted input called prediction [Section 4.3, Section 3.3 Eq. (1), Section 4.5.1 (Table 5)]
    "This annotated data set was later used to train a responsibility classification neural network employing supervised learning... rcrash =λ×P baseline ×Accident_liability_ratio (1)... As shown, our model achieves higher success rates and bears significantly less primary responsibility in both scenarios compared to the baseline."

    The responsibility distribution in Table 5 (PR/SH/SE) is measured with the same ViT classifier that supplies Accident_liability_ratio to Eq. (1) during training. The classifier was fitted to labels generated by the same Section 3.2 pipeline that is being evaluated. Reporting 'less primary responsibility' under this setup is partly tautological: the policy is rewarded for minimizing this classifier's liability scores and then scored by that same classifier. This does not establish reduction in independently adjudicated legal liability. Only the CR/TO/SU metrics are genuinely independent; the paper's headline 'lower degree of accident liability' is not independently grounded.

full rationale

The two headline claims — improved accident-responsibility assignment accuracy and reduced agent liability — are both supported by measurements whose reference values originate inside the proposed method. Section 4.3 labels all collision images with the Section 3.2 TRKG+VLM+RAG procedure; Table 6 then compares that same procedure to plain GPT-4o against those self-generated labels, making the accuracy numbers self-referential. Table 5's liability distribution comes from a ViT classifier trained on the same labels and used as the Eq. (1) reward signal, so the policy is being evaluated with the very model it was trained to satisfy. There is no human legal-expert validation and no external accident-liability dataset; the single-image prompt also asks for non-visual facts (speed, direction), a concrete source of shared bias. The success/crash rates in Table 5 are independent and the paper candidly reports the fragile 9-sample shared-liability validation set, which is why the score is 7 rather than 8-10. No load-bearing self-citation or imported uniqueness theorem was found; the circularity is in the evaluation loop, not in the citation chain.

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

The central claim rests on the VLM+RAG labels being truthful and on the hand-chosen mapping from blame category to reward multiplier. The TRKG is a software artifact, not a new physical or ontological entity, so no invented entities are listed.

free parameters (4)
  • Accident_liability_ratio mapping = primary=1, shared=0.5, secondary=0
    Hand-chosen discretization of the LLM output into a crash-penalty multiplier; this mapping directly sets r_crash in Eq. (1) and is not derived from data or law.
  • lambda crash penalty scaling factor = 1
    Chosen to ensure consistency with the baseline (Section 3.3); no sensitivity analysis is provided.
  • P_baseline crash penalty = -20
    Fixed crash penalty inherited from the MetaDrive baseline; central to the scale of r_crash.
  • Reward coefficients alpha_d, alpha_c, alpha_s, Rarrival, early termination = 0.1, 0.01, 0.1, 40, -20
    Taken from the baseline reward function, not fitted; included for completeness.
assumptions (5)
  • domain assumption The TRKG extracted by an LLM from the Road Traffic Safety Law faithfully represents applicable legal rules.
    Section 4.2 builds the graph from one law using prompt-based LLM extraction; no legal expert validation is reported.
  • domain assumption A VLM can recover legal responsibility facts, including speed and direction of travel, from a single simulated post-collision image.
    Table 2 prompts ask for speed and direction, which are not physically visible in a still image; the labeled dataset in Section 4.3 inherits this.
  • ad hoc to paper Labels produced by the TRKG+VLM pipeline are treated as ground truth for training and evaluation.
    Section 4.3 uses these labels to train the ViT and Section 4.5.3 computes accuracy against them; no independent human labels are used.
  • domain assumption MetaDrive simulation scenarios are representative proxies for real driving conditions.
    All RL training and testing occur in the simulator; generalization to real roads is asserted but not demonstrated.
  • standard math PPO with the stated hyperparameters is a reliable optimizer for MetaDrive.
    Standard algorithm from Stable-Baselines3; acceptable.

how reviews work

0 comments
Cite this review

Pith. "Pith review of ROAD: Responsibility-Oriented Reward Design for Reinforcement Learning in Autonomous Driving." pith.science (2026). https://pith.science/paper/FIRG5Y5G

@misc{pith2026250524317,
  author       = {Pith},
  title        = {Pith review of: ROAD: Responsibility-Oriented Reward Design for Reinforcement Learning in Autonomous Driving},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/FIRG5Y5G}},
  note         = {Machine review of arXiv:2505.24317}
}
read the original abstract

Reinforcement learning (RL) in autonomous driving employs a trial-and-error mechanism, enhancing robustness in unpredictable environments. However, crafting effective reward functions remains challenging, as conventional approaches rely heavily on manual design and demonstrate limited efficacy in complex scenarios. To address this issue, this study introduces a responsibility-oriented reward function that explicitly incorporates traffic regulations into the RL framework. Specifically, we introduced a Traffic Regulation Knowledge Graph and leveraged Vision-Language Models alongside Retrieval-Augmented Generation techniques to automate reward assignment. This integration guides agents to adhere strictly to traffic laws, thus minimizing rule violations and optimizing decision-making performance in diverse driving conditions. Experimental validations demonstrate that the proposed methodology significantly improves the accuracy of assigning accident responsibilities and effectively reduces the agent's liability in traffic incidents.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

30 extracted references · 17 canonical work pages

  1. [1]

    In: 2024 27th International Conference on Computer Supported Cooperative Work in Design (CSCWD), pp

    Huang, X., Su, J.: Intermediate tasks enhanced end-to-end autonomous driving with uncertainty estimation. In: 2024 27th International Conference on Computer Supported Cooperative Work in Design (CSCWD), pp. 133–138 (2024). IEEE 15

    work page 2024

  2. [2]

    IEEE/ASME Transactions on Mechatronics29(4), 2983–2990 (2024) https://doi.org/10.1109/ TMECH.2024.3402126

    Du, J., Bai, Y., Li, Y., Geng, J., Huang, Y., Chen, H.: Evolutionary end-to-end autonomous driving model with continuous-time neural networks. IEEE/ASME Transactions on Mechatronics29(4), 2983–2990 (2024) https://doi.org/10.1109/ TMECH.2024.3402126

    arXiv 2024

  3. [3]

    In: Leonardis, A., Ricci, E., Roth, S., Russakovsky, O., Sattler, T., Varol, G

    Qiao, G., Quan, G., Qu, R., Liu, G.: Modelling competitive behaviors in autonomous driving under generative world model. In: Leonardis, A., Ricci, E., Roth, S., Russakovsky, O., Sattler, T., Varol, G. (eds.) Computer Vision – ECCV 2024, pp. 19–36. Springer, Cham (2025)

    work page 2025

  4. [4]

    IEEE Transactions on Vehicular Technology69(10), 10581–10595 (2020)

    Shan, Y., Zheng, B., Chen, L., Chen, L., Chen, D.: A reinforcement learning-based adaptive path tracking approach for autonomous driving. IEEE Transactions on Vehicular Technology69(10), 10581–10595 (2020)

    work page 2020

  5. [5]

    IEEE Transactions on Vehicular Technology (2024)

    Xu, Y., Shi, Y., Tong, X., Chen, S., Ge, Y.: A multi-agent reinforcement learning based control method for connected and autonomous vehicles in a mixed platoon. IEEE Transactions on Vehicular Technology (2024)

    work page 2024

  6. [6]

    CAD Computer-Aided Design 21(S23), 175–190 (2024)

    Chen, X., Chen, L.: Exploration of adaptive environment design strategy based on reinforcement learning in cad environment. CAD Computer-Aided Design 21(S23), 175–190 (2024)

    work page 2024

  7. [7]

    In: Pro- ceedings of the 2024 4th International Conference on Internet of Things and Machine Learning

    Wang, Z., Yan, H., Wei, C., Wang, J., Bo, S., Xiao, M.: Research on autonomous driving decision-making strategies based deep reinforcement learning. In: Pro- ceedings of the 2024 4th International Conference on Internet of Things and Machine Learning. IoTML ’24, pp. 211–215. Association for Computing Machin- ery, New York, NY, USA (2024). https://doi.org/...

    work page arXiv 2024

  8. [8]

    https://arxiv.org/abs/2104.13906

    Knox, W.B., Allievi, A., Banzhaf, H., Schmitt, F., Stone, P.: Reward (Mis)design for Autonomous Driving (2022). https://arxiv.org/abs/2104.13906

    arXiv 2022

Show all 30 references
  1. [9]

    arXiv preprint arXiv:2402.04764 (2024)

    Venuto, D., Islam, S.N., Klissarov, M., Precup, D., Yang, S., Anand, A.: Code as reward: Empowering reinforcement learning with vlms. arXiv preprint arXiv:2402.04764 (2024)

    2024 arXiv

  2. [10]

    arXiv preprint arXiv:2402.03681 (2024)

    Wang, Y., Sun, Z., Zhang, J., Xian, Z., Biyik, E., Held, D., Erickson, Z.: Rl-vlm-f: Reinforcement learning from vision language foundation model feedback. arXiv preprint arXiv:2402.03681 (2024)

    2024 arXiv

  3. [11]

    arXiv preprint arXiv:2412.15544 (2024)

    Huang, Z., Sheng, Z., Qu, Y., You, J., Chen, S.: Vlm-rl: A unified vision language models and reinforcement learning framework for safe autonomous driving. arXiv preprint arXiv:2412.15544 (2024)

    2024 arXiv

  4. [12]

    arXiv preprint arXiv:2402.18476 (2024) 16

    Zhu, L., Ji, D., Chen, T., Xu, P., Ye, J., Liu, J.: Ibd: Alleviating hallucina- tions in large vision-language models via image-biased decoding. arXiv preprint arXiv:2402.18476 (2024) 16

    2024 arXiv

  5. [13]

    https://arxiv.org/abs/2408.02032

    Huo, F., Xu, W., Zhang, Z., Wang, H., Chen, Z., Zhao, P.: Self-Introspective Decoding: Alleviating Hallucinations for Large Vision-Language Models (2024). https://arxiv.org/abs/2408.02032

    2024 arXiv

  6. [14]

    arXiv preprint arXiv:2402.00253 (2024)

    Liu, H., Xue, W., Chen, Y., Chen, D., Zhao, X., Wang, K., Hou, L., Li, R., Peng, W.: A survey on hallucination in large vision-language models. arXiv preprint arXiv:2402.00253 (2024)

    2024 arXiv

  7. [15]

    Transport Reviews41(5), 556–577 (2021) https://doi.org/10.1080/01441647.2020.1862355

    Martinho, A., Herber, N., Kroesen, M., Chorus, C.: Ethical issues in focus by the autonomous vehicles industry. Transport Reviews41(5), 556–577 (2021) https://doi.org/10.1080/01441647.2020.1862355

    2021

  8. [16]

    Ethics and Information Technology23(4), 657–673 (2021) https://doi

    Gill,T.:Ethicaldilemmasarereallyimportanttopotentialadoptersofautonomous vehicles. Ethics and Information Technology23(4), 657–673 (2021) https://doi. org/10.1007/s10676-021-09605-y

    2021 doi

  9. [17]

    American Journal of Roentgenology219(1), 152–156 (2022) https://doi.org/10.2214/AJR.21.27224 https://doi.org/10.2214/AJR.21.27224

    Mezrich, J.L.: Is artificial intelligence (ai) a pipe dream? why legal issues present significant hurdles to ai autonomy. American Journal of Roentgenology219(1), 152–156 (2022) https://doi.org/10.2214/AJR.21.27224 https://doi.org/10.2214/AJR.21.27224. PMID: 35138133

    2022 doi

  10. [18]

    arXiv preprint arXiv:2010.11929 (2020)

    Dosovitskiy, A., Beyer, L., Kolesnikov, A., Weissenborn, D., Zhai, X., Unterthiner, T., Dehghani, M., Minderer, M., Heigold, G., Gelly, S., et al.: An image is worth 16x16 words: Transformers for image recognition at scale. arXiv preprint arXiv:2010.11929 (2020)

    2020 arXiv

  11. [19]

    Accessed: 2025-01-21 (2019)

    China, G.: Regulations for the Implementation of the Road Traffic Safety Law of the People’s Republic of China. Accessed: 2025-01-21 (2019). https://www.gov. cn/gongbao/content/2019/content_5468932.htm

    2019

  12. [20]

    IEEE transactions on pattern analysis and machine intelligence45(3), 3461–3475 (2022)

    Li, Q., Peng, Z., Feng, L., Zhang, Q., Xue, Z., Zhou, B.: Metadrive: Com- posing diverse driving scenarios for generalizable reinforcement learning. IEEE transactions on pattern analysis and machine intelligence45(3), 3461–3475 (2022)

    2022

  13. [21]

    In: Leonardis, A., Ricci, E., Roth, S., Russakovsky, O., Sattler, T., Varol, G

    Marcu, A.-M., Chen, L., Hünermann, J., Karnsund, A., Hanotte, B., Chidananda, P., Nair, S., Badrinarayanan, V., Kendall, A., Shotton, J., Arani, E., Sinavski, O.: Lingoqa: Visual question answering for autonomous driving. In: Leonardis, A., Ricci, E., Roth, S., Russakovsky, O....

    2024

  14. [22]

    https: //arxiv.org/abs/2307.07162

    Fu, D., Li, X., Wen, L., Dou, M., Cai, P., Shi, B., Qiao, Y.: Drive Like a Human: Rethinking Autonomous Driving with Large Language Models (2023). https: //arxiv.org/abs/2307.07162

    2023 arXiv

  15. [23]

    https://arxiv.org/abs/2503

    Delavari, E., Khalil, A., Kwon, J.: CARIL: Confidence-Aware Regression in Imitation Learning for Autonomous Driving (2025). https://arxiv.org/abs/2503. 00783 17

    2025

  16. [24]

    https://arxiv.org/abs/ 2403.18965

    Ye, X., Tao, F., Mallik, A., Yaman, B., Ren, L.: LORD: Large Models based Opposite Reward Design for Autonomous Driving (2024). https://arxiv.org/abs/ 2403.18965

    2024 arXiv

  17. [25]

    https://arxiv.org/abs/2212.11419

    Lu, Y., Fu, J., Tucker, G., Pan, X., Bronstein, E., Roelofs, R., Sapp, B., White, B., Faust, A., Whiteson, S., Anguelov, D., Levine, S.: Imitation Is Not Enough: Robustifying Imitation with Reinforcement Learning for Challenging Driving Scenarios (2023). https://arxiv.org/abs/...

    2023 arXiv

  18. [26]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp

    Hu, Z., Iscen, A., Sun, C., Wang, Z., Chang, K.-W., Sun, Y., Schmid, C., Ross, D.A., Fathi, A.: Reveal: Retrieval-augmented visual-language pre-training with multi-source multimodal knowledge memory. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Rec...

    2023

  19. [27]

    https://arxiv.org/abs/2201.11903

    Wei, J., Wang, X., Schuurmans, D., Bosma, M., Ichter, B., Xia, F., Chi, E., Le, Q., Zhou, D.: Chain-of-Thought Prompting Elicits Reasoning in Large Language Models (2023). https://arxiv.org/abs/2201.11903

    2023 arXiv

  20. [28]

    https://openai.com/index/gpt-4o

    OpenAI: GPT-4o. https://openai.com/index/gpt-4o. Accessed: 2025-05-10 (2024)

    2024

  21. [29]

    arXiv preprint arXiv:1707.06347 (2017)

    Schulman, J., Wolski, F., Dhariwal, P., Radford, A., Klimov, O.: Proximal policy optimization algorithms. arXiv preprint arXiv:1707.06347 (2017)

    2017 arXiv

  22. [30]

    https://neo4j.com (2023) 18

    Neo4j: The Neo4j Graph Database Platform. https://neo4j.com (2023) 18

    2023

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.