Autonomous discovery of traffic laws with AI traffic scientists

Chao Guo; Chao Xiang; Fei-Yue Wang; Junyou Shang; Qinghai Miao; Xiaoyan Gong; Xingyuan Dai; Yisheng Lv; Yizhang Chai; Yonglin Tian

arxiv: 2607.01639 · v1 · pith:JHBGIZLOnew · submitted 2026-07-02 · 💻 cs.AI

Autonomous discovery of traffic laws with AI traffic scientists

Xingyuan Dai , Yue Liu , Xiaoyan Gong , Qinghai Miao , Junyou Shang , Yutong Wang , Chao Guo , Yonglin Tian

show 4 more authors

Yizhang Chai Chao Xiang Yisheng Lv Fei-Yue Wang

This is my paper

Pith reviewed 2026-07-03 14:41 UTC · model grok-4.3

classification 💻 cs.AI

keywords traffic law discoveryautonomous scientific discoveryagentic AI workflowurban driving behaviortemporal memory scaleobservational validationtransportation planninghypothesis induction

0 comments

The pith

TrafficSci AI autonomously rediscovers three traffic laws and identifies a new consistent temporal memory scale in urban driving.

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

The paper presents TrafficSci as an agentic AI system that turns traffic-law discovery into an iterative workflow of evidence scoping, critic-judge hypothesis induction, and observational-interventional validation. In four case studies across population, network, control, and trajectory scales, the system recovers three known laws while also detecting an unreported intrinsic temporal memory scale that holds statistically across eight cities and two datasets. The work matters because traffic laws supply the recurring patterns needed for planning, management, and control, and shifting their discovery from expert-driven effort to automated workflows could scale the process to larger and messier data. A sympathetic reader would see value if the same steps reliably surface genuine regularities instead of workflow artifacts.

Core claim

TrafficSci formulates traffic-law discovery as an iterative, auditable workflow integrating evidence scoping, critic-judge hypothesis induction, and observational-interventional validation. Across four case studies spanning population, network, control and trajectory scales, TrafficSci autonomously rediscovers three established traffic laws and identifies an unreported intrinsic temporal memory scale in urban driving behavior, statistically consistent across eight cities and two trajectory datasets.

What carries the argument

The agentic AI workflow that integrates evidence scoping, critic-judge hypothesis induction, and observational-interventional validation to induce and test candidate traffic laws.

If this is right

Traffic laws can be rediscovered without requiring prior expert identification of candidate regularities.
An intrinsic temporal memory scale exists in urban driving behavior and remains consistent across cities and datasets.
The workflow extends AI-driven discovery methods from controlled laboratory settings to complex urban transportation systems.
Discovered laws supply a scientific basis for transportation planning, management, and control at multiple scales.

Where Pith is reading between the lines

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

The same workflow structure could be tested on other complex observational domains such as economic flows or ecological population dynamics to check transferability.
If the temporal memory scale proves robust, it could be inserted directly into existing traffic simulation models to improve short-term prediction accuracy.
Running the validation step on controlled interventions, such as temporary road closures, would provide a direct test of whether the induced laws support causal claims.
Scaling the evidence-scoping component to real-time streaming data might allow continuous updating of laws as cities evolve.

Load-bearing premise

The critic-judge hypothesis induction and observational-interventional validation steps produce laws that reflect genuine traffic regularities rather than artifacts of the AI workflow or data processing choices.

What would settle it

Applying the identical workflow to fresh trajectory data from additional independent cities and finding that the reported temporal memory scale loses statistical consistency or that the rediscovered laws diverge from separate expert validation on the same data.

read the original abstract

Universal traffic laws describe recurrent patterns in congestion, mobility and driving behavior across cities, providing a scientific basis for transportation planning, management and control. Their discovery, however, remains expert-driven, requiring candidate regularities to be identified from heterogeneous observational evidence or validated through intervention experiments. Although autonomous artificial intelligence (AI) systems have advanced scientific discovery in controlled laboratory settings, extending them to complex transportation domains remains a challenge. Here we present TrafficSci, an agentic AI system that formulates traffic-law discovery as an iterative, auditable workflow integrating evidence scoping, critic-judge hypothesis induction, and observational-interventional validation. Across four case studies spanning population, network, control and trajectory scales, TrafficSci autonomously rediscovers three established traffic laws and identifies an unreported intrinsic temporal memory scale in urban driving behavior, statistically consistent across eight cities and two trajectory datasets. TrafficSci provides a route for extending AI-driven scientific discovery from controlled domains to complex urban systems.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

TrafficSci outlines an agentic workflow that claims to rediscover three known traffic laws and surface a new consistent temporal memory scale across cities, but the abstract supplies no equations, metrics, or validation steps so the strength of those claims cannot be judged yet.

read the letter

The main takeaway is that this paper describes TrafficSci as an iterative AI system that scopes evidence, induces hypotheses via a critic-judge setup, and validates them observationally and interventionally. It reports rediscovering three established traffic laws plus one unreported temporal memory scale that appears statistically consistent across eight cities and two trajectory datasets.

The workflow itself is the clearest piece of new content. Framing traffic-law discovery as an auditable, multi-step agentic process and applying it across population, network, control, and trajectory scales is a concrete extension of earlier AI-for-science ideas into urban systems. The consistency claim for the new scale, if backed by independent checks, would be the sort of reproducible regularity that transportation researchers could actually use.

What the paper does reasonably well is keep the method description high-level but structured, and it positions the work as moving discovery from expert intuition to something more systematic. That framing is honest about the domain challenges.

The soft spots are mostly around missing detail. The abstract states outcomes but gives no equations, no data-processing rules, no error bars, and no ablation on the critic-judge step. Without those, it is impossible to assess whether the new scale reflects real behavior or choices in the induction and validation pipeline. The circularity concern the reader flagged is real on present evidence: if validation is not fully decoupled from how hypotheses were generated, the cross-city consistency could be an artifact.

This paper is for people working on AI-driven discovery in messy real-world domains or on data-driven traffic modeling. A reader who already follows agentic systems or urban informatics would get value from seeing the workflow applied at city scale. It is not yet ready for someone who needs the actual numbers or code to build on.

I would send it to peer review. The core idea is testable and the claims are specific enough that referees could demand the missing validation details and data.

Referee Report

1 major / 0 minor

Summary. The manuscript presents TrafficSci, an agentic AI system for autonomous traffic-law discovery formulated as an iterative workflow of evidence scoping, critic-judge hypothesis induction, and observational-interventional validation. Across four case studies at population, network, control, and trajectory scales, the system is reported to rediscover three established traffic laws and to identify a previously unreported intrinsic temporal memory scale in urban driving behavior that remains statistically consistent across eight cities and two trajectory datasets.

Significance. If the central claims are supported by the full methods and results, the work would constitute a meaningful extension of AI-driven discovery methods from controlled laboratory settings to complex urban systems. The cross-city and cross-dataset consistency of the new temporal scale, together with the auditable workflow, could provide a useful template for transportation science and planning.

major comments (1)

Abstract: the claim that TrafficSci 'autonomously rediscovers three established traffic laws' and identifies a new 'statistically consistent' temporal scale cannot be evaluated because the abstract supplies no equations, data details, exclusion rules, error bars, or validation metrics; without these elements it is impossible to determine whether the reported outcomes reflect genuine regularities or artifacts of the workflow or data processing.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their review. We address the single major comment below, noting that the full manuscript contains the technical details referenced.

read point-by-point responses

Referee: [—] Abstract: the claim that TrafficSci 'autonomously rediscovers three established traffic laws' and identifies a new 'statistically consistent' temporal scale cannot be evaluated because the abstract supplies no equations, data details, exclusion rules, error bars, or validation metrics; without these elements it is impossible to determine whether the reported outcomes reflect genuine regularities or artifacts of the workflow or data processing.

Authors: Abstracts are constrained by length and are designed to summarize scope and findings at a high level rather than to contain equations, exclusion rules, error bars or full validation metrics. The manuscript provides these elements in the Methods (workflow description, hypothesis induction procedure, observational and interventional validation protocols), Results (four case studies with rediscovered laws and the temporal memory scale), and supplementary materials (city-specific data, statistical tests, consistency metrics across eight cities and two datasets). The abstract's claims are therefore intended to be evaluated against the full text, which reports the specific regularities, statistical consistency measures, and controls for artifacts. revision: no

Circularity Check

0 steps flagged

No significant circularity

full rationale

The supplied manuscript text consists solely of the abstract. No equations, pseudocode, derivation steps, fitted parameters, or self-citation chains are present that could be inspected for any of the enumerated circularity patterns. The workflow is described at a high level only, with no opportunity to exhibit a reduction of a claimed prediction or law to its own inputs by construction. This is the expected honest non-finding when the derivation chain cannot be walked.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract supplies no information on free parameters, axioms, or invented entities; full manuscript required to populate the ledger.

pith-pipeline@v0.9.1-grok · 5723 in / 1018 out tokens · 32138 ms · 2026-07-03T14:41:32.589408+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

50 extracted references · 50 canonical work pages

[1]

Duan, J.et al.Spatiotemporal dynamics of traffic bottlenecks yields an early signal of heavy congestions.Nature Communications14, 8002 (2023)

work page 2023
[2]

Nature Communications11, 1616 (2020)

Saberi, M.et al.A simple contagion process describes spreading of traffic jams in urban networks. Nature Communications11, 1616 (2020)

work page 2020
[3]

Schl¨ apfer, M.et al.The universal visitation law of human mobility.Nature593, 522–527 (2021)

work page 2021
[4]

& Guimer` a, R

Cabanas-Tirapu, O., Dan´ us, L., Moro, E., Sales-Pardo, M. & Guimer` a, R. Human mobility is well described by closed-form gravity-like models learned automatically from data.Nature Communications 16, 1336 (2025)

work page 2025
[5]

M., Santi, P., Resta, G., Strogatz, S

Vazifeh, M. M., Santi, P., Resta, G., Strogatz, S. H. & Ratti, C. Addressing the minimum fleet problem in on-demand urban mobility.Nature557, 534–538 (2018)

work page 2018
[6]

Hamedmoghadam, H., Jalili, M., Vu, H. L. & Stone, L. Percolation of heterogeneous flows uncovers the bottlenecks of infrastructure networks.Nature Communications12, 1254 (2021)

work page 2021
[7]

& Duan, Y

Lv, Y., Zhang, X., Kang, W. & Duan, Y. Managing emergency traffic evacuation with a partially random destination allocation strategy: A computational-experiment-based optimization approach. IEEE Transactions on Intelligent Transportation Systems16, 2182–2191 (2015)

work page 2015
[8]

E., Colak, S., Shafiei, S., Saberi, M

Olmos, L. E., Colak, S., Shafiei, S., Saberi, M. & Gonzalez, M. C. Macroscopic dynamics and the collapse of urban traffic.Proceedings of the National Academy of Sciences of the United States of America115, 12654–12661 (2018)

work page 2018
[9]

Wang, F.-Y. Parallel control and management for intelligent transportation systems: Concepts, archi- tectures, and applications.IEEE Transactions on Intelligent Transportation Systems11, 630–638 (2010)

work page 2010
[10]

& Wang, F.-Y

Lv, Y., Duan, Y., Kang, W., Li, Z. & Wang, F.-Y. Traffic flow prediction with big data: A deep learning approach.IEEE Transactions on Intelligent Transportation Systems16, 865–873 (2014)

work page 2014
[11]

& Tegmark, M

Udrescu, S.-M. & Tegmark, M. AI Feynman: A physics-inspired method for symbolic regression. Science Advances6, eaay2631 (2020)

work page 2020
[12]

Wang, H.et al.Scientific discovery in the age of artificial intelligence.Nature620, 47–60 (2023)

work page 2023
[13]

J., Yigitcanlar, T

Fu, X., Li, C., Quan, S. J., Yigitcanlar, T. & Wasserman, D. Large language models in urban planning. Nature Cities2, 585–592 (2025)

work page 2025
[14]

Zhang, Y.et al.Exploring the role of large language models in the scientific method: from hypothesis to discovery.npj Artificial Intelligence1, 14 (2025)

work page 2025
[15]

Nature Synthesis3, 606–614 (2024)

Chen, J.et al.Navigating phase diagram complexity to guide robotic inorganic materials synthesis. Nature Synthesis3, 606–614 (2024)

work page 2024
[16]

Lu, C.et al.Towards end-to-end automation of AI research.Nature651, 914–919 (2026)

work page 2026
[17]

A., MacKnight, R., Kline, B

Boiko, D. A., MacKnight, R., Kline, B. & Gomes, G. Autonomous chemical research with large language models.Nature624, 570–578 (2023)

work page 2023
[18]

J.et al.An autonomous laboratory for the accelerated synthesis of novel materials

Szymanski, N. J.et al.An autonomous laboratory for the accelerated synthesis of novel materials. Nature624, 86–91 (2023)

work page 2023
[19]

Gottweis, J., Weng, W.-H., Daryin, A.et al.Accelerating scientific discovery with Co-Scientist.Nature (2026). 17

work page 2026
[20]

Nie, T., Sun, J. & Ma, W. Exploring the roles of large language models in reshaping transportation systems: A survey, framework, and roadmap.Artificial Intelligence for Transportation1, 100003 (2025)

work page 2025
[21]

Nature625, 468–475 (2024)

Romera-Paredes, B.et al.Mathematical discoveries from program search with large language models. Nature625, 468–475 (2024)

work page 2024
[22]

Tachet, R.et al.Scaling law of urban ride sharing.Scientific Reports7, 42868 (2017)

work page 2017
[23]

& Barab´ asi, A.-L

Song, C., Qu, Z., Blumm, N. & Barab´ asi, A.-L. Limits of predictability in human mobility.Science 327, 1018–1021 (2010)

work page 2010
[24]

R., Song, C

Zhong, L., Dong, L., Wang, Q. R., Song, C. & Gao, J. Universal expansion of human mobility across urban scales.Nature Cities2, 603–607 (2025)

work page 2025
[25]

& Gonz´ alez, M

C ¸ olak, S., Lima, A. & Gonz´ alez, M. C. Understanding congested travel in urban areas.Nature Communications7, 10793 (2016)

work page 2016
[26]

Zeng, G.et al.Unveiling city jam-prints of urban traffic based on jam patterns.Communications Physics8, 121 (2025)

work page 2025
[27]

& Barthelemy, M

Taillanter, E. & Barthelemy, M. Empirical evidence for a jamming transition in urban traffic.Journal of the Royal Society Interface18, 20210391 (2021)

work page 2021
[28]

& Barthelemy, M

Louf, R. & Barthelemy, M. How congestion shapes cities: from mobility patterns to scaling.Scientific Reports4, 5561 (2014)

work page 2014
[29]

Advance online publication

Arora, N.et al.Urban congestion relief experiments through routing-app interventions.Nature Cities (2026). Advance online publication

work page 2026
[30]

& Barab´ asi, A.-L

Gao, J., Barzel, B. & Barab´ asi, A.-L. Universal resilience patterns in complex networks.Nature530, 307–312 (2016)

work page 2016
[31]

Li, M., Pan, X., Liu, C. & Li, Z. Federated deep reinforcement learning-based urban traffic signal optimal control.Scientific Reports15, 11724 (2025)

work page 2025
[32]

& Zhang, T

Wang, K., Shen, Z., Lei, Z., Liu, X. & Zhang, T. Towards multi-agent reinforcement learning based traffic signal control through spatio-temporal hypergraphs.IEEE Transactions on Mobile Computing (2025)

work page 2025
[33]

R., Parvate, K., Vinitsky, E

Wu, C., Kreidieh, A. R., Parvate, K., Vinitsky, E. & Bayen, A. M. Flow: A modular learning framework for mixed autonomy traffic.IEEE Transactions on Robotics38, 1270–1286 (2021)

work page 2021
[34]

Wu, K.et al.Big-data empowered traffic signal control could reduce urban carbon emission.Nature Communications16, 2013 (2025)

work page 2013
[35]

& Zheng, Z

Saifuzzaman, M. & Zheng, Z. Incorporating human-factors in car-following models: A review of recent developments and research needs.Transportation Research Part C: Emerging Technologies48, 379–403 (2014)

work page 2014
[36]

& Sun, J

Huang, X., Sun, J. & Sun, J. A car-following model considering asymmetric driving behavior based on long short-term memory neural networks.Transportation Research Part C: Emerging Technologies 95, 346–362 (2018)

work page 2018
[37]

Ma, L. & Qu, S. A sequence to sequence learning based car-following model for multi-step predic- tions considering reaction delay.Transportation Research Part C: Emerging Technologies120, 102785 (2020)

work page 2020
[38]

& Schwarte, L

Schober, P., Boer, C. & Schwarte, L. A. Correlation coefficients: appropriate use and interpretation. Anesthesia & Analgesia126, 1763–1768 (2018). 18

work page 2018
[39]

Proceedings of the Neural Information Processing Systems Track on Datasets and Benchmarks(2021)

Wilson, B.et al.Argoverse 2: Next generation datasets for self-driving perception and forecasting. Proceedings of the Neural Information Processing Systems Track on Datasets and Benchmarks(2021)

work page 2021
[40]

Caesar, H.et al.nuScenes: A multimodal dataset for autonomous driving.Proceedings of the IEEE/CVF conference on Computer Vision and Pattern Recognition11621–11631 (2020)

work page 2020
[41]

& Wang, Y.-X

Zhou, A., Yan, K., Shlapentokh-Rothman, M., Wang, H. & Wang, Y.-X. Language Agent Tree Search unifies reasoning, acting, and planning in language models.Proceedings of the 41st International Conference on Machine Learning235, 56561–56584 (2024)

work page 2024
[42]

Computational Materials Science58, 218–226 (2012)

Curtarolo, S.et al.AFLOW: An automatic framework for high-throughput materials discovery. Computational Materials Science58, 218–226 (2012)

work page 2012
[43]

Yao, S.et al.Tree of thoughts: Deliberate problem solving with large language models.Advances in Neural Information Processing Systems36, 11809–11822 (2023)

work page 2023
[44]

& Huang, K

Zippelius, A. & Huang, K. Density-wave fronts on the brink of wet granular condensation.Scientific Reports7, 3613 (2017)

work page 2017
[45]

& Crockett, M

Messeri, L. & Crockett, M. J. Artificial intelligence and illusions of understanding in scientific research. Nature627, 49–58 (2024)

work page 2024
[46]

& Massaro, D

Marchetti, A., Manzi, F., Riva, G., Gaggioli, A. & Massaro, D. Artificial intelligence and the illusion of understanding: A systematic review of theory of mind and large language models.Cyberpsychology, Behavior, and Social Networking28, 505–514 (2025)

work page 2025
[47]

Li, J.et al.An astronomical question answering dataset for evaluating large language models.Scientific Data12, 447 (2025)

work page 2025
[48]

& Zhou, D

Cai, T., Wang, X., Ma, T., Chen, X. & Zhou, D. Large language models as tool makers.International Conference on Learning Representations(2024)

work page 2024
[49]

Advances in Neural Information Processing Systems37, 50528–50652 (2024)

Yang, J.et al.SWE-agent: Agent-computer interfaces enable automated software engineering. Advances in Neural Information Processing Systems37, 50528–50652 (2024)

work page 2024
[50]

Schick, T.et al.Toolformer: Language models can teach themselves to use tools.Advances in Neural Information Processing Systems36, 68539–68551 (2023). 19

work page 2023

[1] [1]

Duan, J.et al.Spatiotemporal dynamics of traffic bottlenecks yields an early signal of heavy congestions.Nature Communications14, 8002 (2023)

work page 2023

[2] [2]

Nature Communications11, 1616 (2020)

Saberi, M.et al.A simple contagion process describes spreading of traffic jams in urban networks. Nature Communications11, 1616 (2020)

work page 2020

[3] [3]

Schl¨ apfer, M.et al.The universal visitation law of human mobility.Nature593, 522–527 (2021)

work page 2021

[4] [4]

& Guimer` a, R

Cabanas-Tirapu, O., Dan´ us, L., Moro, E., Sales-Pardo, M. & Guimer` a, R. Human mobility is well described by closed-form gravity-like models learned automatically from data.Nature Communications 16, 1336 (2025)

work page 2025

[5] [5]

M., Santi, P., Resta, G., Strogatz, S

Vazifeh, M. M., Santi, P., Resta, G., Strogatz, S. H. & Ratti, C. Addressing the minimum fleet problem in on-demand urban mobility.Nature557, 534–538 (2018)

work page 2018

[6] [6]

Hamedmoghadam, H., Jalili, M., Vu, H. L. & Stone, L. Percolation of heterogeneous flows uncovers the bottlenecks of infrastructure networks.Nature Communications12, 1254 (2021)

work page 2021

[7] [7]

& Duan, Y

Lv, Y., Zhang, X., Kang, W. & Duan, Y. Managing emergency traffic evacuation with a partially random destination allocation strategy: A computational-experiment-based optimization approach. IEEE Transactions on Intelligent Transportation Systems16, 2182–2191 (2015)

work page 2015

[8] [8]

E., Colak, S., Shafiei, S., Saberi, M

Olmos, L. E., Colak, S., Shafiei, S., Saberi, M. & Gonzalez, M. C. Macroscopic dynamics and the collapse of urban traffic.Proceedings of the National Academy of Sciences of the United States of America115, 12654–12661 (2018)

work page 2018

[9] [9]

Wang, F.-Y. Parallel control and management for intelligent transportation systems: Concepts, archi- tectures, and applications.IEEE Transactions on Intelligent Transportation Systems11, 630–638 (2010)

work page 2010

[10] [10]

& Wang, F.-Y

Lv, Y., Duan, Y., Kang, W., Li, Z. & Wang, F.-Y. Traffic flow prediction with big data: A deep learning approach.IEEE Transactions on Intelligent Transportation Systems16, 865–873 (2014)

work page 2014

[11] [11]

& Tegmark, M

Udrescu, S.-M. & Tegmark, M. AI Feynman: A physics-inspired method for symbolic regression. Science Advances6, eaay2631 (2020)

work page 2020

[12] [12]

Wang, H.et al.Scientific discovery in the age of artificial intelligence.Nature620, 47–60 (2023)

work page 2023

[13] [13]

J., Yigitcanlar, T

Fu, X., Li, C., Quan, S. J., Yigitcanlar, T. & Wasserman, D. Large language models in urban planning. Nature Cities2, 585–592 (2025)

work page 2025

[14] [14]

Zhang, Y.et al.Exploring the role of large language models in the scientific method: from hypothesis to discovery.npj Artificial Intelligence1, 14 (2025)

work page 2025

[15] [15]

Nature Synthesis3, 606–614 (2024)

Chen, J.et al.Navigating phase diagram complexity to guide robotic inorganic materials synthesis. Nature Synthesis3, 606–614 (2024)

work page 2024

[16] [16]

Lu, C.et al.Towards end-to-end automation of AI research.Nature651, 914–919 (2026)

work page 2026

[17] [17]

A., MacKnight, R., Kline, B

Boiko, D. A., MacKnight, R., Kline, B. & Gomes, G. Autonomous chemical research with large language models.Nature624, 570–578 (2023)

work page 2023

[18] [18]

J.et al.An autonomous laboratory for the accelerated synthesis of novel materials

Szymanski, N. J.et al.An autonomous laboratory for the accelerated synthesis of novel materials. Nature624, 86–91 (2023)

work page 2023

[19] [19]

Gottweis, J., Weng, W.-H., Daryin, A.et al.Accelerating scientific discovery with Co-Scientist.Nature (2026). 17

work page 2026

[20] [20]

Nie, T., Sun, J. & Ma, W. Exploring the roles of large language models in reshaping transportation systems: A survey, framework, and roadmap.Artificial Intelligence for Transportation1, 100003 (2025)

work page 2025

[21] [21]

Nature625, 468–475 (2024)

Romera-Paredes, B.et al.Mathematical discoveries from program search with large language models. Nature625, 468–475 (2024)

work page 2024

[22] [22]

Tachet, R.et al.Scaling law of urban ride sharing.Scientific Reports7, 42868 (2017)

work page 2017

[23] [23]

& Barab´ asi, A.-L

Song, C., Qu, Z., Blumm, N. & Barab´ asi, A.-L. Limits of predictability in human mobility.Science 327, 1018–1021 (2010)

work page 2010

[24] [24]

R., Song, C

Zhong, L., Dong, L., Wang, Q. R., Song, C. & Gao, J. Universal expansion of human mobility across urban scales.Nature Cities2, 603–607 (2025)

work page 2025

[25] [25]

& Gonz´ alez, M

C ¸ olak, S., Lima, A. & Gonz´ alez, M. C. Understanding congested travel in urban areas.Nature Communications7, 10793 (2016)

work page 2016

[26] [26]

Zeng, G.et al.Unveiling city jam-prints of urban traffic based on jam patterns.Communications Physics8, 121 (2025)

work page 2025

[27] [27]

& Barthelemy, M

Taillanter, E. & Barthelemy, M. Empirical evidence for a jamming transition in urban traffic.Journal of the Royal Society Interface18, 20210391 (2021)

work page 2021

[28] [28]

& Barthelemy, M

Louf, R. & Barthelemy, M. How congestion shapes cities: from mobility patterns to scaling.Scientific Reports4, 5561 (2014)

work page 2014

[29] [29]

Advance online publication

Arora, N.et al.Urban congestion relief experiments through routing-app interventions.Nature Cities (2026). Advance online publication

work page 2026

[30] [30]

& Barab´ asi, A.-L

Gao, J., Barzel, B. & Barab´ asi, A.-L. Universal resilience patterns in complex networks.Nature530, 307–312 (2016)

work page 2016

[31] [31]

Li, M., Pan, X., Liu, C. & Li, Z. Federated deep reinforcement learning-based urban traffic signal optimal control.Scientific Reports15, 11724 (2025)

work page 2025

[32] [32]

& Zhang, T

Wang, K., Shen, Z., Lei, Z., Liu, X. & Zhang, T. Towards multi-agent reinforcement learning based traffic signal control through spatio-temporal hypergraphs.IEEE Transactions on Mobile Computing (2025)

work page 2025

[33] [33]

R., Parvate, K., Vinitsky, E

Wu, C., Kreidieh, A. R., Parvate, K., Vinitsky, E. & Bayen, A. M. Flow: A modular learning framework for mixed autonomy traffic.IEEE Transactions on Robotics38, 1270–1286 (2021)

work page 2021

[34] [34]

Wu, K.et al.Big-data empowered traffic signal control could reduce urban carbon emission.Nature Communications16, 2013 (2025)

work page 2013

[35] [35]

& Zheng, Z

Saifuzzaman, M. & Zheng, Z. Incorporating human-factors in car-following models: A review of recent developments and research needs.Transportation Research Part C: Emerging Technologies48, 379–403 (2014)

work page 2014

[36] [36]

& Sun, J

Huang, X., Sun, J. & Sun, J. A car-following model considering asymmetric driving behavior based on long short-term memory neural networks.Transportation Research Part C: Emerging Technologies 95, 346–362 (2018)

work page 2018

[37] [37]

Ma, L. & Qu, S. A sequence to sequence learning based car-following model for multi-step predic- tions considering reaction delay.Transportation Research Part C: Emerging Technologies120, 102785 (2020)

work page 2020

[38] [38]

& Schwarte, L

Schober, P., Boer, C. & Schwarte, L. A. Correlation coefficients: appropriate use and interpretation. Anesthesia & Analgesia126, 1763–1768 (2018). 18

work page 2018

[39] [39]

Proceedings of the Neural Information Processing Systems Track on Datasets and Benchmarks(2021)

Wilson, B.et al.Argoverse 2: Next generation datasets for self-driving perception and forecasting. Proceedings of the Neural Information Processing Systems Track on Datasets and Benchmarks(2021)

work page 2021

[40] [40]

Caesar, H.et al.nuScenes: A multimodal dataset for autonomous driving.Proceedings of the IEEE/CVF conference on Computer Vision and Pattern Recognition11621–11631 (2020)

work page 2020

[41] [41]

& Wang, Y.-X

Zhou, A., Yan, K., Shlapentokh-Rothman, M., Wang, H. & Wang, Y.-X. Language Agent Tree Search unifies reasoning, acting, and planning in language models.Proceedings of the 41st International Conference on Machine Learning235, 56561–56584 (2024)

work page 2024

[42] [42]

Computational Materials Science58, 218–226 (2012)

Curtarolo, S.et al.AFLOW: An automatic framework for high-throughput materials discovery. Computational Materials Science58, 218–226 (2012)

work page 2012

[43] [43]

Yao, S.et al.Tree of thoughts: Deliberate problem solving with large language models.Advances in Neural Information Processing Systems36, 11809–11822 (2023)

work page 2023

[44] [44]

& Huang, K

Zippelius, A. & Huang, K. Density-wave fronts on the brink of wet granular condensation.Scientific Reports7, 3613 (2017)

work page 2017

[45] [45]

& Crockett, M

Messeri, L. & Crockett, M. J. Artificial intelligence and illusions of understanding in scientific research. Nature627, 49–58 (2024)

work page 2024

[46] [46]

& Massaro, D

Marchetti, A., Manzi, F., Riva, G., Gaggioli, A. & Massaro, D. Artificial intelligence and the illusion of understanding: A systematic review of theory of mind and large language models.Cyberpsychology, Behavior, and Social Networking28, 505–514 (2025)

work page 2025

[47] [47]

Li, J.et al.An astronomical question answering dataset for evaluating large language models.Scientific Data12, 447 (2025)

work page 2025

[48] [48]

& Zhou, D

Cai, T., Wang, X., Ma, T., Chen, X. & Zhou, D. Large language models as tool makers.International Conference on Learning Representations(2024)

work page 2024

[49] [49]

Advances in Neural Information Processing Systems37, 50528–50652 (2024)

Yang, J.et al.SWE-agent: Agent-computer interfaces enable automated software engineering. Advances in Neural Information Processing Systems37, 50528–50652 (2024)

work page 2024

[50] [50]

Schick, T.et al.Toolformer: Language models can teach themselves to use tools.Advances in Neural Information Processing Systems36, 68539–68551 (2023). 19

work page 2023