Chat2Scenic Boosts Autonomous Driving Scenario Generation with RAG.

Autonomous driving systems require extensive testing with diverse, regulation-compliant scenarios. Generating these scenarios automatically, especially as executable scripts in a Domain Specific Language (DSL), has been a persistent challenge. Existing methods struggle with either scalability or success rates. Researchers have introduced Chat2Scenic, an innovative iterative framework leveraging Retrieval-Augmented Generation (RAG). This system features a chatbot interface for interactive scenario refinement and integrates RAG to ensure scenario generation adheres to regulatory knowledge and DSL syntax. Chat2Scenic demonstrates superior performance compared to current state-of-the-art methods, achieving a 76.42% compilation success rate and 58.17% framework accuracy. The project also includes an open benchmark of 123 scenarios from various regulations, and its code is open-sourced to foster further research.