Collaborative Research: Scalable Data-Enabled Predictive Control for Heterogeneous Mixed Traffic Systems

Michigan State University, East Lansing MI

Investigators

Abstract

This grant will fund research that enables advancements in transportation efficiency and safety through the deployment of virtually connected and automated vehicles among human-driven vehicles, thereby promoting the progress of science and advancing the national prosperity. While the potential benefits for fuel efficiency and road safety from full vehicle automation and vehicle-to-vehicle communication peak in a traffic system without human drivers, mixed traffic scenarios with coexistence between human-driven vehicles and automated vehicles will be the norm in the intermediate term. A major challenge to the control of automated vehicles in such environments is the requirement that the behavior of the human drivers either be reliably described using explicit car-following models or accurately predicted using computationally efficient, data-driven techniques, neither of which is currently possible. This project aims to resolve this challenge by developing a new model-free, data-efficient control and optimization framework that will enable fast decision-making for efficient, robust, and safe coordination of multiple connected and automated vehicles in mixed traffic systems. The results will be disseminated to the research community and the automotive industry through sharing of open-source software code and organization of a workshop with speakers from both academia and industry. These efforts are closely integrated with educational and outreach activities that aim to increase the participation of undergraduate and high-school students in engineering research. This research aims to develop the foundations of efficient and scalable control designs for connected and automated vehicles that can meet real-time computational constraints and guarantee safe performance in mixed traffic, without explicit modeling of the behavior of human-driven vehicles. It accomplishes this outcome by building a data-driven predictive control framework in which system-level cost functions and constraints are synergistically designed to handle unknown and uncertain traffic dynamics directly from input/output data, and adaptive data library updates respond to time-varying traffic conditions. Additionally, it develops algorithms for scalable, online data compression and distributed optimization that exploit cascading system structures to decompose centralized predictive control problems into those of lower dimension without compromising control performance. Extensive simulations and field experiments conducted in collaboration with an industry partner will be used to evaluate the theoretical outcomes. 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.