Gao, Qi (Columbia University), Di, Xuan (Columbia University)

Learning Traffic States and Fundamental Diagram Via Physics-Informed Neural Operators

Scheduled for presentation during the Invited Session "S22a-Emerging Trends in AV Research" (TH-LM-T22), Thursday, November 20, 2025, 12:10−12:30, Coolangata 1

2025 IEEE 28th International Conference on Intelligent Transportation Systems (ITSC), November 18-21, 2025, Gold Coast, Australia

This information is tentative and subject to change. Compiled on October 18, 2025

Abstract

Traffic state estimation (TSE) reconstructs the spatio-temporal evolution of traffic states, such as density and velocity, from limited sensor observations, playing a pivotal role in traffic monitoring and control. Traditional TSE approaches mainly bifurcate into model-driven and data-driven methods, each facing distinct limitations in accuracy, generalizability, and data efficiency. Motivated by the recent success of hybrid learning paradigms, this paper introduces a novel framework that leverages Physics-Informed Neural Operators (PINO) to jointly learn the traffic state and underlying fundamental diagram parameters directly from partial observations. By embedding physical constraints into Neural Operators, the proposed method achieves superior generalization across varying traffic conditions without the need for retraining. Experiments on simulated traffic dynamics governed by Lighthill-Whitham-Richards (LWR) and Aw-Rascle-Zhang (ARZ) models, and real world Next Generation SIMulation (NGSIM) dataset demonstrate the performance of our approach compared against baseline methods in both traffic state reconstruction and model parameter estimation, highlighting its potential for real-time and scalable deployment in transportation systems.