Transportation Engineering Seminar – Collective rationality from self-interested agents in mixed autonomy traffic

About the event

Abstract: The rapid development of self-driving technologies makes it imperative to characterize and model the behaviors of mixed autonomy traffic. The problem is especially important, yet more challenging, when automated vehicles are self-interested, instead of being centrally controlled or coordinated. In this talk, I will present my recent research towards solving this problem. I will first introduce a game theory model of mixed autonomy traffic. An intriguing finding is that, under mild conditions, agents in mixed autonomy traffic can always attain Pareto-efficient Nash equilibria, also known as “collective rationality”, even if they are all self-interested and behave differently. Then I will present a data-driven numerical approach to model mixed traffic flow and show how the model achieves good accuracy through capturing the complicated inter-class interactions in real-world traffic parsimoniously. I will conclude the talk with a discussion of my research roadmap and ongoing work on automated vehicle behavior design.