A Game Theory–Based Lane-Change Decision Model for Autonomous Vehicles

Abstract

Lane changing is a challenging but critical task for autonomous vehicles (AVs), especially when interacting with human-driven vehicles. Numerous lane change (LC) models have been proposed; however most pay little attention to interactions with surrounding vehicles. Although a few studies have recognized the importance of vehicle interaction during LC, most have considered it only from the perspective of safety. In addition, most previous studies relied solely on numerical data to validate the effectiveness of LC models. Furthermore, deploying newly developed LC models directly in real vehicles is often impractical because of the associated cost and safety concerns. To address these gaps, this study proposes a game theory–based LC decision model for AVs that determines the optimal timing for lane changing based on payoff evaluation. The payoff function comprehensively incorporates three aspects—safety, efficiency, and comfort—to provide more reasonable decision-making during the LC process. In addition, the driving preferences of competing vehicles are estimated by the AV and incorporated into payoff weighting process, enabling better interaction with different types of human drivers. A Stackelberg game framework is further introduced to determine the optimal decision strategy. The proposed model was evaluated in Simulink using potential conflicting LC events from the Shanghai Naturalistic Driving Study. In the simulation, the original lane-changing vehicles were replaced by AVs controlled by the proposed model. The results show that the model is capable of successfully completing lane-changing maneuvers while alleviating traffic conflicts to some extent compared with the original trajectories.