A Survey of LLMs for Autonomous Driving

The rapidly evolving fields of autonomous driving (AD) and Large Language Models (LLMs) represent two of the most transformative technological advancements of our time. While autonomous vehicles promise to revolutionize transportation, they face formidable challenges related to complex reasoning, human interaction, and safety in highly dynamic environments. Concurrently, LLMs have demonstrated unprecedented capabilities in understanding, generating, and reasoning with human language, offering a powerful new paradigm for intelligent systems. This survey establishes the motivation for exploring the convergence of these two powerful technologies, emphasizing the potential for LLMs to address some of the most intricate challenges in AD. We provide a comprehensive overview of LLM fundamentals and the core components of autonomous driving systems, bridging these domains to illustrate their inherent complementarity. Furthermore, we systematically analyze various methodologies and architectural patterns for integrating LLMs into AD, categorizing their key applications across perception, prediction, planning, control, and human-vehicle interaction. Finally, we critically examine the current challenges, limitations, and ethical considerations impeding widespread deployment, offering promising future research directions to advance the field towards safer, more intelligent, and human-centric autonomous systems.