Minimax Strategy for Lane Choice Prediction in Markovian Traffic Modeling

Daily, vehicles overload the road networks. Starting from this worldwide issue, we can see the necessity of finding solutions for travel times improvements. Many solutions arise if we look at the lowest possible level in traffic modeling. At the microscopic level, we can have the best overview of the vehicle's interactions, about their velocity, acceleration or lane change behavior. This paper proposes a new approach of lane choice prediction using an AI (Artificial Intelligence) based gaming strategy. Having a piece of road with the lane change action modeled as Markovian process, for a chosen vehicle, this paper aims to find the best decision considering the desired destination and the involved costs. The lane change action will be described as a game where is involved the vehicle that initiates the action, together with parameters as traffic destination volumes, intermediary traffic volumes associated with distinct lanes and the placement of lanes splitting points. The comparison with the Dijkstra's algorithm, known as the optimal for route choice problem solving, proves the efficiency of the proposed approach. The similar results obtained as destination volumes by comparing with the mentioned optimal algorithm consists of a validation step in this paper proposal.