Interactive adversarial driving with quantified psychological traits

Autonomous driving research increasingly focuses on adversarial driving behaviour (ADB) generation. However, existing methods mainly rely on preset trajectories or simple intervention strategies, limiting real-time interaction capability. In addition, most generative approaches emphasise macroscopic trajectory distributions while insufficiently modelling the intrinsic cognitive mechanisms underlying adversarial interactions. To address these limitations, an interactive ADB framework incorporating multiple human psychological traits is proposed. In this study, psychological characteristics including distracted driving, response delay, patience and politeness were quantitatively characterised using kinematic indicators. A dual-input ADB framework was then developed by integrating psychological factors with traffic game theory. After this, a dual-network switching trajectory generation method was constructed using external traffic networks and internal psychological networks. Finally, the proposed model was validated using a naturalistic driving dataset from the perspectives of decision consistency and trajectory similarity. The proposed model achieved over 85% accuracy for both metrics across multiple scenarios, demonstrating the effectiveness of psychological trait modelling and interactive ADB generation.