Context-aware decision making in autonomous vehicles: Integrating social behavior modeling with large language models

Integrating context-aware decision-making in autonomous vehicles (AVs) is critical for advancing operational efficiency, safety, and user experience. However, existing frameworks struggle to incorporate social context into real-time navigation, relying on deterministic algorithms or reinforcement learning models that overlook implicit social norms and face challenges in translating LLM-derived reasoning into safety-compliant control policies. This paper investigates the application of social behavior modeling fused with large language models (LLMs) to establish a comprehensive framework for context-aware understanding and decision-making processes by AVs. Through understanding of the scene and the deployment of LLMs, this framework enables AVs to interpret and respond to complex social interactions and contextual cues, enhancing adaptability in dynamic environments. We propose concepts and approaches to foster context-aware and socially responsible decision-making processes, including test cases for validation to some level. The results demonstrate substantial improvements in decision accuracy adopting virtual simulation, providing a foundation for addressing complex ethical dilemmas and real-time decision-making challenges that AVs encounter in diverse and dynamic settings.