Risk Preference-Based Decision-Making and Control Framework for Pedestrian Interaction

Safety is undoubtedly crucial in autonomous vehicle-pedestrian interaction; however, drivers’ individual risk preferences significantly influence user satisfaction, trust, and acceptance of AV systems. To address this, we propose an adaptive decision-making and control framework with personalized risk preferences, enabling both controlled safety and diverse driving styles during pedestrian interactions. Firstly, a conditional random field-based probabilistic module was employed to assess the risk level of surrounding traffic pedestrians. Based on this assessment, a risk avoidance decision-making mechanism with personalized risk preferences (e.g., aggressive or conservative) is designed to accommodate various driver needs. Furthermore, an adaptive dynamic control model is developed. All model parameters in this study are calibrated using real-world traffic data. Extensive validation confirms safety and risk preference adaptability. The results show that the proposed framework effectively reduces collision risks in diverse scenarios and outperforms other commonly-used methods. Additionally, driving simulation experiments reveal that participants report higher levels of perceived safety, comfort, usability, and overall acceptability compared to conventional automatic emergency braking systems and existing risk-aware deep reinforcement learning methods. This framework provides an effective pathway for achieving personalized and customizable safety-oriented decision-making and control in autonomous vehicles.