Interactive Game-Theoretic Model for Decision Making in Traffic

Abstract

Decision making in traffic is a current hot research topic in the driving automation scientific community. The necessity to define interactive-aware algorithms, flexible and able to cooperate with other traffic participants without explicit communication, has increased the interest on Game Theory, powerful mathematical tool suitable for multi-agent environments as traffic. In this framework, a Nash Equilibrium (NE) describes the optimal point in which each player´s strategy is the best response to the strategies of all the other players. The computation of the NE for dynamic games can be however unfeasible for real time online application. This article presents a Bayesian approach for decision making in traffic, proposing as prior estimate the NE in mixed strategies (MSNE) of a static nonzero-sum polymatrix game with individual preferences, the likelihood is then computed based on the Kullback-Leibler divergence. The approach shows good results in simulation, with its computational complexity allowing for online applications