Toward an integrative approach to machine learning for traffic management

This NSF project will develop and experiment with an integrative approach to applying machine learning (ML) method in traffic management. An integrative approach tailors the principles of ML methods––rather than prepacked toolboxes––and immerse them with the domain knowledge to create new, hybrid methods.  The transportation sector is currently experiencing monumental disruptions with the introduction and constant evolution of new technologies and mobility services.  The SEA—Sharing, Electrification, and Automation—change is expected to transform how mobility is planned, provided, and used by people and companies in cities and beyond.   The importance of traffic management will grow as the broad trends toward automation and sharing continue in the next decade. Ridesharing has long been considered a low-hanging fruit for traffic management.  Thanks to emerging technologies (e.g., transportation network companies and autonomous driving), the popularity of ridesharing has grown markedly in recent years. Fueling this renewed interest is the promise to make ridesharing flexible, user friendly, efficient, and cheap. On a grander scale, ridesharing is seen as integral to Mobility-as-a-Service (MaaS), the idea that mobility will be increasingly consumed as service in a future filled with shared electric autonomous vehicles, rather than through vehicle ownership.  In the era of MaaS, ridesharing would likely become a major, if not dominating, mode of travel.  The results from this project will help policy makers understand how ridesharing can be leveraged to improve the efficiency, environmental sustainability, and equity of our transportation systems.

This project will create a novel methodology for a range of challenging transportation problems involving sequential hierarchical optimization.   By bringing together transportation science, game theory, optimization, and ML, the framework exemplifies an integrative approach to cross-disciplinary research.   It also opens new pathways toward harnessing the power of ML in the transportation domain on the one hand and brings stimulating new challenges to the attention of the ML community on the other.  The project introduces and analyzes the problem of managing traffic by ridesharing, which is born of recent technology advances but has yet to receive much attention.  This problem connects hierarchical optimization, a traditional topic in transportation, to ridesharing, a rapidly growing field of study in several disciplines (economics, computer science, and transportation), and promises to push forward the frontiers of both. The findings will deepen our understanding of ridesharing not only as a mode of cooperative travel but also as a tool for traffic management.