Research map / Large language models / Instruction and fine-tuning
TrafficAlign: Aligning Large Language Models for Traffic Scenario Generation
Zhi Tu, Liangkun Niu, Tianyi Zhang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add LLM alignment on video-derived scenarios (vs GPT-5 prompted directly) (ablation) | TrafficAlign scenario generator · SafeBench testing, Los Angeles | collision rate: 0.889 -> 0.933 | Improved |
| add LLM alignment (vs Claude Sonnet 4 prompted directly) (ablation) | TrafficAlign scenario generator · SafeBench testing, Los Angeles | collision rate: 0.894 -> 0.933 | Improved |
| add LLM alignment (vs Qwen3 prompted directly) (ablation) | TrafficAlign scenario generator · SafeBench testing, Los Angeles | overall score: 0.450 -> 0.405 | Improved |
| use TrafficAlign scenarios instead of ChatScene for testing | driving models · SafeBench testing, Los Angeles | collision rate: 0.825 -> 0.933 | Improved |
| use TrafficAlign scenarios instead of ChatScene for testing | driving models · SafeBench testing, Los Angeles | overall score: 0.481 -> 0.405 | Improved |
| use TrafficAlign scenarios instead of ChatScene for testing | driving models · SafeBench testing, Small Towns | running red lights: 0.191 -> 0.186 | No clear effect |
| fine-tune driving models on TrafficAlign scenarios vs no fine-tuning | PPO/SAC/TD3 driving models · Held-out set, Los Angeles | collision rate: 0.968 -> 0.701 | Improved |
| fine-tune on TrafficAlign scenarios instead of ChatScene scenarios | PPO/SAC/TD3 driving models · Held-out set, Los Angeles | collision rate: 0.733 -> 0.701 | Improved |
| fine-tune on TrafficAlign scenarios instead of ChatScene scenarios | PPO/SAC/TD3 driving models · Held-out set, Yosemite | collision rate: 0.641 -> 0.604 | Improved |
| fine-tune on TrafficAlign scenarios instead of ChatScene scenarios | PPO/SAC/TD3 driving models · Held-out set, Los Angeles | overall score: 0.301 -> 0.385 | Improved |
| use 5s instead of 15s frame sampling interval on intense driving videos (ablation) | TrafficAlign · SafeBench testing, 10 intense videos | collision rate: 0.899 -> 0.907 | Improved |
| use 5s instead of 15s frame sampling interval on intense driving videos (ablation) | TrafficAlign · Fine-tuning driving models, 10 intense videos | collision rate: 0.768 -> 0.766 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- ChatScene — ChatScene: Knowledge-Enabled Safety-Critical Scenario Generation for Autonomous Vehicles (CVPR 2024) · better on 3 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Driving Everywhere with Large Language Model Policy Adaptation (CVPR 2024 · similarity 0.88)
- Generating Traffic Scenarios via In-Context Learning to Learn Better Motion Planner (AAAI 2025 · similarity 0.87)
- Trajeglish: Traffic Modeling as Next-Token Prediction (ICLR 2024 · similarity 0.87)
- SceneDiffuser++: City-Scale Traffic Simulation via a Generative World Model (CVPR 2025 · similarity 0.87)
- STSBench: A Spatio-temporal Scenario Benchmark for Multi-modal Large Language Models in Autonomous Driving (NeurIPS 2025 · similarity 0.87)
- Language-Driven Interactive Traffic Trajectory Generation (NeurIPS 2024 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
