This research demonstrates a novel approach to safely generate autonomous driving software code using LLMs guided by simulation environments.

• Combines LLM capabilities with simulation-based verification to ensure safety
• Develops a specialized prompting framework for autonomous driving code generation
• Achieves significant improvement in functional correctness of generated code
• Establishes a methodology for safely automating safety-critical software development

This innovation addresses key engineering challenges in autonomous vehicle development by reducing development time while maintaining safety standards, potentially accelerating the deployment of reliable self-driving technologies.

On Simulation-Guided LLM-based Code Generation for Safe Autonomous Driving Software