Yield or Rush? Social-Preference-Aware Driving Interaction Modeling Using Game-Theoretic Framework

The newly formed hybrid traffic flow where Human-driven Vehicles (HV) share roads with Autonomous Vehicles (AV) is a foreseeable trend in the modern transportation system, making HV-AV interactions inevitable. This paper proposes a Parallel-Game-based Interaction Model (PGIM) to describe social-preference-aware driving interactions. Firstly, quantitively defined social preference is integrated into a game-theoretic model to capture the social patterns in interactions between heterogeneous drivers. Secondly, an active semantic decision-making method is developed within PGIM via a counterfactual reasoning process. This method enables agents to gradually understand the social preference of an unacquainted interacting agent during the interaction and make semantic decisions social-compliantly. We verify the interactive driving performance of PGIM in simulations by showcasing in the unprotected left-turn scenario. The results show that the proposed PGIM could trigger distinct interaction evolvement patterns by varying social preferences of interacting agents. Therefore, the PGIM brings a potential way from the perspective of the social property to explicitly reveal the mechanism in human-involved driving interactions.