← Back to all news
arXiv Machine Learning By Federica Bragone, Matthieu Barreau

Learning Traffic Flow Dynamics with Stochastic Physics-Informed Neural Cellular Automata

Read the original on arXiv Machine Learning →

The paper introduces a physics-informed neural cellular automaton (PI‑NCA) designed to learn traffic flow dynamics from data while respecting physical constraints such as vehicle conservation and road topology. By extending the model to stochastic dynamics, it can capture probabilistic transition rules without violating these constraints. Experiments on Nagel‑Schreckenberg and Kerner‑Klenov‑Wolf cellular automata show that PI‑NCA outperforms standard neural cellular automata in learning both deterministic and stochastic traffic models.

Machine-generated by The Flow from the publisher's headline and feed description — not written or checked by a human. The full article lives at arXiv Machine Learning.

Related stories

arXiv Machine Learning
Aug 27

Quantum-Inspired Modeling of Driving Behavior

The paper introduces a quantum-inspired representation of driver behavior that models drivers as evolving density matrices, capturing continuous, probabilistic, context-dependent, and history-dependent interactions among behavioral variables. Trained unsupervised on the I‑24 MOTION dataset, the framework identifies three interpretable driving regimes—free flow, transition, and congestion—and reproduces macroscopic traffic phenomena such as the fundamental diagram and hysteresis loops. The representation also enhances practical applications by providing context-dependent parameters for classical car‑following models and enabling autonomous vehicles to forecast nearby drivers’ motion in real time.

By Mohammad Elayan, Omid Armantalab, Wissam Kontar
agentssafety
More like this →
arXiv Machine Learning
Sep 14

An Empirical Markov Chain Car-Following (MC-CF) Model

The paper introduces the Markov Chain Car‑Following (MC‑CF) model, an empirical probabilistic approach that represents car‑following as a Markov process and samples accelerations from empirical distributions within discretized state bins. Evaluation on the Waymo Open Motion Dataset shows that MC‑CF variants outperform physics‑based baselines and compete with modern data‑driven methods in both one‑step and open‑loop trajectory prediction. Zero‑shot transfer to the Naturalistic Phoenix dataset and microscopic ring‑road simulations demonstrate cross‑domain generalization and scalability, with the model reducing collisions and reproducing naturalistic shockwave propagation.

By Sungyong Chung, Yanlin Zhang, Nachuan Li, Dana Monzer, Alireza Talebpour
More like this →
arXiv AI
Jun 2

TrafficClaw: A Generalizable LLM Agent in the Unified Physical Environment for Urban Traffic Control

arXiv:2604. 17456v2 Announce Type: replace Abstract: Large language model (LLM) agents have shown strong capabilities in long-horizon reasoning, tool use, and decision-making in digital environments, yet extending them to physically grounded systems remains challenging.

By Siqi Lai, Pan Zhang, Yuping Zhou, Jindong Han, Yansong Ning, Hao Liu
llmsagentsreinforcement-learning
More like this →