Research papers · Posted 2025.05.02 · Publication / acceptance: author-reported
Seeking to Collide: Online Safety-Critical Scenario Generation for Autonomous Driving with Retrieval Augmented Large Language Models
Seeking to Collide: Online Safety-Critical Scenario Generation for Autonomous Driving with Retrieval Augmented Large Language Models
Overview
The framework infers hazardous background-vehicle intentions with LLMs and generates feasible adversarial trajectories online. A growing retrieval bank of intent-planner pairs supports adaptation, evaluated using motion data and collision-related metrics.
Publication and version information
Publication, acceptance or venue information is author-reported on arXiv. The exact source wording is preserved below.
Author-supplied source information
Author-supplied information on arXiv.
IEEE International Conference on Intelligent Transportation Systems, 2025
Accepted at IEEE ITSC 2025
Version: arXiv v2 — manuscript revision