Synapse
⌘+K
Synapse
PulseExploreClubsResearchersJournals
Instagram
HomeClubsExplore
October 3, 2025Open Access

Seeking to Collide: Online Safety-Critical Scenario Generation for Autonomous Driving with Retrieval Augmented Large Language Models

View Full Paper
Ask AI
Bookmark
Share

Authors

YMYongfeng MeiNankai UniversityTNTingting NieHenan University of Science and TechnologyJSJian SunXinjiang Agricultural University

Discussion

Loading...

Member takes

Implication

This framework generates safety-critical scenarios in real-time, reducing collision risks for autonomous vehicles, indicating enhanced safety measures.

Key Points

  • The model reduces the mean minimum time-to-collision from 1.62 to 1.08 seconds, improving safety outcomes.
  • Using the Waymo Open Motion Dataset, the framework achieves a 75% collision rate, demonstrating superior performance.
  • This approach leverages a dynamic retrieval bank of intent-planner pairs, facilitating real-time adaptation to novel scenarios.
  • Online retrieval-augmented scenario generation exposes autonomous vehicles to rare, dangerous driving situations effectively.

Cite This Study

Mei et al. (2025) studied this question.

synapsesocial.com/papers/68e03501f0e39f13e7fa3959https://doi.org/10.48550/arxiv.2505.00972
View Full Paper
Ask AI
Bookmark
Share

Also Consider

Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context:

  1. 1Large Language Model-Based Realistic Safety-Critical Driving Video Generation2026
  2. 2Automated Generation of Test Scenarios for Autonomous Driving Using LLMs2025 · 10 citations
  3. 3Enhancing Autonomous Vehicle Training with Language Model Integration and Critical Scenario Generation2024 · 3 citations
  4. 4ChatScene: Knowledge-Enabled Safety-Critical Scenario Generation for Autonomous Vehicles2024 · 1 citations
  5. 5Open-World Critical Scenario Recognition and Maneuver-Level Generation for Autonomous Driving Simulation Testing2026