Reproduce “Text2Scenario: Text-Driven Scenario Generation for Autonomous Driving Test” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Xuan Cai, Xuesong Bai, Zhiyong Cui, Danmu Xie, Daocheng Fu, Haiyang Yu, Yilong Ren

Text2Scenario (T2S) is a framework that leverages large language models to automatically convert natural language descriptions into executable ASAM OpenScenario DSL files for autonomous driving simulation testing. The system demonstrates significantly faster scenario generation compared to human experts and effectively identifies safety violations and technical vulnerabilities across various autonomous driving stacks.

Key insights

  • Large Language Models, particularly advanced ones like ChatGPT-4.0, can achieve high accuracy (up to 92.3%) in parsing nuanced elements from natural language descriptions for AD scenario generation.
  • A structured prompt engineering pipeline significantly enhances LLM performance, improving the interpretive capabilities for implicit expressions and ensuring structural integrity of outputs.
  • Automated DSL scenario generation dramatically reduces the time and human effort required for test case creation, enabling more efficient and scalable validation of autonomous driving systems.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.