Decision Making via Game Theory for Autonomous Vehicles in the Presence of a Moving Obstacle

Abstract

We consider an emergency maneuver scenario involving two autonomous vehicles interacting with a road obstacle characterized by a random behavior. We employ game theory to solve the resulting problems, first framing a static game of compete information, and further adding incomplete information about the obstacle so as to transform it into a Bayesian game. Depending on the considered scenario, the autonomous vehicles can have multiple available actions, such as to stay at the same lane and swerve and move to another one. These actions can lead to different outcomes, such as keep driving on an empty lane, hit the obstacle, or hit another car. We analyse the Nash equilibria of the game and test the hypothesis that the knowledge of one vehicle about an obstacle can be advantageous to other road participants, which is key in the context of connected vehicles.