CAREER: Human Mobility Prediction and Intervention based on Cross-Domain Infrastructure-Human Interactions

This project aims to model and support human mobility in real time at an urban scale. Better models of human mobility can help meet both sustainability challenges (through predicting traffic congestion, air quality, and energy consumption) and improve urban resilience to disruptive events (such as infrastructure failures, natural disasters, or pandemics). The key idea of the project is that people's increasingly frequent interaction with transportation, communication, financial, and other infrastructure can be used to understand mobility patterns. However, collecting and integrating this information into mobility models is still an open challenge. Further, people's collective decisions can negatively impact infrastructure, increasing wait times and reducing capacity in both transit and information infrastructures. Through collecting and integrating mobility-related behavior across multiple sources, the project team will advance the state of the art around mobility modeling methods and develop interventions that encourage people to make choices that improve mobility outcomes at scale, especially during crisis events. The team will also develop educational materials to train students to be both future researchers and workers who possess data collection and modeling expertise, particularly around questions of human mobility.

The project is structured as two main thrusts that align with the goals of mobility modeling and mobility interventions. The first thrust for mobility prediction will explore the correlation and interdependency of interactions across multiple types of infrastructure, notably transportation, communication, and financial interactions. This will be done through advancing techniques based on multi-view learning to integrate cross-domain interactions, which will be integrated into a prediction framework using a correlation-driven multi-task recurrent neural network architecture. The second thrust aims to improve urban resilience by developing interventions to improve mobility under disruptive events. The team will use a novel dynamic Markov decision process formulation solved with distributed deep reinforcement learning to develop recommendations that enhance collective mobility, such as new departure times or routes for individuals, or road closures and transit capacity allocations for city planners. These models will leverage and be evaluated using existing datasets of infrastructure interaction and disrupted mobility collected before, during, and after the COVID-19 pandemic. Together, this work will lead to general principles, design methodologies, and a long-term research trajectory for cross-domain infrastructure-human interaction.

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.