Optimal Decision-Making Strategies for Self-Driving Car Inspired by Game Theory

This paper presents an optimal decision-making strategy for a self-driving car using a game-theoretic approach. To ensure the safety of the decision, Stackelberg game's maximin reward strategy, which considers concurrency, is applied. The receding horizon is included to increase the accuracy of the decision, but the computational burden is high. We assume that the follower takes only one prediction time, not the receding horizon, to relieve the computational burden. For an accurate prediction of interacting vehicles, the intention estimation model is suggested. We demonstrate the efficiency of our approach in a simulation environment and various traffic conditions.