Research on Interaction Strategies of Autonomous Vehicles Based on Social Preference

In the development of autonomous vehicles (AVs), ensuring safety and reliability is of paramount importance, especially when considering the presence of pedestrians and their behavior. This necessitates the implementation of robust motion control algorithms. Addressing the vehicle decision-making challenge in pedestrian-populated environments, we have devised an interactive pedestrian model that effectively combines a social force-based approach with pedestrian crossing intentions. By integrating these elements, our model can accurately determine pedestrian trajectories influenced by AVs. To further enhance the decision-making process, we introduce a framework rooted in social preference theory and deep reinforcement learning (DRL). This framework facilitates the generation of diverse decision strategies, accommodating various driving styles. To validate our proposed framework, we conduct rigorous network training and simulation experiments, systematically comparing the strategies derived from DRL algorithms with multiple control experiments. Our comprehensive evaluation demonstrates that the learning-based framework not only fosters smoother and more natural interactions between AVs and pedestrians but also enhances overall safety during such encounters.