As urban areas experience rapid population growth and development, the pervasive issue of traffic congestion has become increasingly challenging, particularly in large cities. Various strategies aimed at mitigating congestion have been proposed, spanning from the expansion of roadways and the optimization of transportation systems to the implementation of demand management. Congestion pricing and incentive schemes, integral to demand management, have emerged as tools to reinforce traffic control measures. While extensive research has focused on congestion pricing in conventional mobility systems, there's a notable gap in understanding its implications for future mobility services. In the traditional paradigm, the driver, who is also the car owner, holds absolute decision-making authority over aspects such as origin, destination, route, and travel time. In contrast, future mobility systems involve a diverse array of organizations and companies that exert varying degrees of influence over the behavior of individual drivers, whether human or AI-based. These entities encompass car-sharing services (e.g., Zipcar), ride-hailing services (e.g., Waymo, Uber, Lyft), crowdsourced delivery systems (e.g., Amazon Flex, Instacart, DoorDash), navigation applications (e.g., Google Maps and Waze), and even companies producing autonomous vehicles with integrated navigation systems (e.g., Tesla), among others.
In future mobility systems, (real-time) incentivization of organizations can be more effective than targeting individual drivers. In addition, having insights into organizations' anticipated road network demand becomes instrumental for the central planner to implement impactful incentives, steering the entire network towards the "system optimal" point. Consider, for instance, the knowledge of impending delivery deadlines for Amazon packages or the anticipated demand for road segments by Waymo/Uber drivers in the next few minutes—such information empowers the central planner to offer targeted and effective incentives. However, the challenge lies in acquiring this knowledge, as organizations are often reluctant to share their data due to data protection regulations or concerns about divulging information to competitors.
The objective of this project is to develop "private collaborative" mechanisms facilitating data sharing among organizations and a central planner. Subsequently, the central planner utilizes this shared data to offer incentives, encouraging organizations to influence the behavior of individual drivers within their ranks or those utilizing their services. Drawing from the principles of "differential privacy" in machine learning, we aim to ensure a rigorous guarantee of privacy protection for the organizations' data. Our proposed framework integrates real-time traffic data, alongside the anticipated future demands of organizations and their tasks, to continually forecast network traffic flow and incentivize organizations to alleviate congestion. Our methodology's performance will be assessed through an evaluation utilizing data from the highly congested urban landscapes of Los Angeles and New York. These locations are chosen as ideal validation areas due to being among the most traffic-congested cities in the United States. Specifically, our assessment will rely on the Archived Data Management System (ADMS) in Los Angeles, The New York City Taxi and Limousine Commission (TLC) data in New York, and The City of New York Department of Transportation's real-time traffic data.