Chat2Scenic Generates Autonomous Driving Scenario Scripts
TUM-AVS · hf · 2026-07-18
This work introduces Chat2Scenic, an iterative RAG framework for autonomous driving simulation scenario generation.
Core Problem
Validating autonomous driving systems requires a massive amount of diverse and compliant test scenarios, but automatically translating regulatory descriptions into executable scenario scripts remains challenging. Existing methods face trade-offs:
- Retrieval-assemble: Decent compilation rates, but lacks scalability.
- Retrieval-based full-script generation: Can directly generate complete scripts, but suffers from low compilation success rates.
Method
Chat2Scenic combines:
- RAG: Retrieves regulatory knowledge to reduce hallucinations.
- Conversational interaction: Supports step-by-step scenario refinement.
- DSL syntax constraints: Makes generated outputs easier to compile.
Results
The authors also released a new benchmark containing 123 scenarios sourced from various regulations (including NHTSA, UN Vehicle Regulations, etc.).
In evaluations, Chat2Scenic achieved:
- 76.42% Compilation Success Rate
- 58.17% Framework Accuracy
In contrast:
- Retrieval Assemble: 30.08% CSR / 11.03% FA
- Retrieval full script generation: 16.26% CSR / 10.86% FA
Open Source
Code is open-sourced: github.com/TUM-AVS/chat2scenic
More from Research
- NeurIPS 2026 workshop will focus on on-device intelligence and local execution — YiMaTweets · 2026-07-21
- NeurIPS 2026 workshop calls papers on on-device intelligence — YiMaTweets · 2026-07-21
- AI Security Institute says every tested model tried to cheat in cyber evaluations — connoraxiotes · 2026-07-21
- AI companies are buying old books to avoid training on AI-generated slop — CackleRooster · 2026-07-21
- Sakana says multiple diffusion models plus MCTS beat test-time scaling on coding and math — SakanaAILabs · 2026-07-21
- Soofi S 30B-A3B releases a full pretraining report and claims open-model leads in English and German — abursuc · 2026-07-21