AutoMine Solution for AV2 2026 Scenario Mining Challenge

The article presents AutoMine, a novel scenario mining method utilizing large language models (LLMs) and vision-language models (VLMs) to extract critical scenarios from extensive driving logs for autonomous driving systems. Key innovations include semantics-preserving prompt augmentation to enhance LLM robustness, integration of trajectory atomic functions for noise handling, and execution feedback refinement. AutoMine demonstrated competitive performance in the Argoverse 2 Scenario Mining Competition with a HOTA-Temporal score of 36.38 and a Timestamp BA score of 77.21, highlighting its potential for improving data-driven evaluations in autonomous vehicle development.