There are several works in which game theoretic learning is used as a model of the evolution of traffic patterns. Self-interested users adjust their traffic decisions to minimize their own experienced congestion. Decisions are updated based on prior experiences and limited observations of traffic data. The resulting game can exhibit a special underlying structure (potential game) so that various update rules asymptotically settle into a (Nash or Wardrop) equilibrium. Given such a model, the question of social influence is how a social planner with limited authority can affect the resulting patterns through the introduction of incentives or penalties. This talk reviews selected work on game theoretic learning with the focus on settings relevant to traffic decisions and discusses some challenges and initial results on using these models for social influence.
Copyright. All Rights Reserved.