Collaborative Research: Learning-Based Scalable Predictive Control Strategies for Heterogeneous Traffic Networks

The widespread adoption of connected and automated vehicle technology is likely to take place over a number of years as the technology becomes more commonly accepted by the public and approved by regulatory authorities. Until then, it is essential to develop traffic management strategies that consider the uncertainty associated with heterogeneities in traffic networks and understand the extent to which these strategies improve the performance of traffic networks. This research project aims to develop and validate infrastructure- and vehicle-based control strategies to enhance heterogeneous traffic networks, addressing human-driven and automated vehicles, mobility, and energy efficiency. The project outcomes will be of interest to municipalities and transportation agencies, the automotive industry, and equipment manufacturers. Specifically, the control approaches will be of value to transportation agencies in understanding how infrastructure-based strategies can be exploited to improve energy efficiency and mobility in mixed traffic environments. Real-time control algorithms developed for autonomous vehicles can help the automotive industry determine a set of protocols that address the needs for safe and effective navigation in a mixed traffic network. Further, the models and techniques developed in this research are expected to have implications for a wide range of applications where the system's behavior can be modeled as an uncertain heterogeneous system, such as aerial and ground mobile robots operating in search and rescue missions. The educational plan is designed to impact graduate and undergraduate students, K-12 students, and minority students to prepare and engage a diverse STEM workforce.

This collaborative research aims to develop a framework for tractable modeling and optimal control of a heterogeneous traffic network consisting of autonomous and human-driven vehicles. This goal will be realized by combining data-driven modeling of uncertain systems, stochastic model predictive control, and distributed optimization. The project defines three research objectives: (1) development of distributed learning- and scenario-based model predictive control methods at the upper (macroscopic) level wherein functional variational Bayesian neural networks will be used to model the state- and input-dependent uncertainty associated with the heterogeneity in the traffic network, and distributed optimization algorithms will be used to enhance the computational efficiencies of the proposed control approach; (2) development of distributed cautious model predictive control-based approaches for heterogeneous multi-agent systems at the lower (microscopic) level to ensure the safety of individual vehicles while tracking the desired reference command set by the macroscopic-level controller; (3) test the effectiveness of the hierarchical learning-based control paradigm for both urban and highway traffic networks using the PTV-VISSIM traffic simulation software.

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.