Collaborative Research: Towards Automatic Learning of Traffic Dynamics

The objective of this project is to support research on deep learning (DL)-based methodologies for discovering the governing equations of traffic dynamics and probing how connected and automated vehicles (CAVs) behave and interact with other road users. With rapid development of artificial intelligence and availability of ubiquitous traffic data, the project aims to transform the methods of learning traffic dynamics from conventional studies to a DL-based automatic paradigm. New traffic dynamics models with CAVs are essential for achieving safety, mobility, and other goals related to future transportation systems. The project team adopts an “open science” approach to encourage collaborations, stimulate interests, and grow research capacity for this important topic. Results are integrated into existing and new courses and provide opportunities for graduate and undergraduate students to participate in cutting-edge research. Findings are broadly shared with transportation agencies, academic communities, and the industry via publications, meetings, and presentations/webinars. This project develops specialized, effective methods for learning traffic dynamics, especially for traffic flow with CAVs, from data directly. This is accomplished by designing new DL structures to address data noises, a coordinated learning framework to deal with the unique features of traffic dynamics due to diverse vehicle classes and/or driving behaviors. Equally important, it formulates new metrics and methods for four essential objectives: accuracy, parsimony, interpretability, and generalizability. Understanding of governing equations of traffic dynamics is fundamental to traffic prediction, transportation planning, traffic management and control. The project thus advances the scientific discovery of new traffic dynamics with CAVs and informs society to better prepare for the wide deployment of emerging technologies. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.