AutoMine Solution for AV2 2026 Scenario Mining Challenge

2026-06-10 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems, Data Science & Analytics · Depth: Expert, quick

Summary

AutoMine is a robust, self-refining scenario mining method designed for autonomous driving systems, leveraging Large Language Models (LLMs) and Vision-Language Models (VLMs). Its purpose is to identify high-value, safety-critical, and planning-relevant scenarios from extensive driving logs, crucial for data-driven evaluation. The system incorporates semantics-preserving prompt augmentation to mitigate LLM prompt sensitivity and integrates robust trajectory atomic functions with VLM-based functions to address perception noise and open-world visual cues. Furthermore, AutoMine refines its generated code through execution feedback derived from real driving logs. In the Argoverse 2 Scenario Mining Competition at CVPR 2026, AutoMine achieved a HOTA-Temporal score of 36.38 and a Timestamp BA score of 77.21, demonstrating its effectiveness.

Key takeaway

For autonomous driving engineers evaluating system safety and planning, AutoMine's approach offers a robust method for scenario mining. You should consider integrating LLM prompt augmentation and VLM-based functions to enhance scenario extraction from large driving logs. Implement execution feedback loops to continuously refine your scenario generation code, improving the relevance and criticality of identified situations.

Key insights

AutoMine uses LLMs and VLMs with self-refinement to extract critical autonomous driving scenarios from logs.

Principles

• Reduce LLM prompt sensitivity.
• Combine trajectory functions with VLM.
• Refine code via execution feedback.

Method

AutoMine employs semantics-preserving prompt augmentation for LLMs, integrates robust trajectory atomic functions with VLM-based functions, and refines generated code using execution feedback from real driving logs.

In practice

• Apply prompt augmentation for LLM robustness.
• Integrate VLM for perception noise handling.
• Use execution feedback for code refinement.

Topics

• Autonomous Driving
• Scenario Mining
• Large Language Models
• Vision-Language Models
• Argoverse 2
• CVPR 2026

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, Computer Vision Engineer

Related on AIssential

• From Data to Theory: Autonomous Large Language Model Agents for Materials Science — Autonomous LLM agents can perform end-to-end scientific theory development, from equation selection to code execution and validation.
• Improving 3D Labeling in Self-Driving by Inferring Vehicle Information using Vision Language Models — VLMs can significantly enhance 3D vehicle auto-labeling in self-driving by inferring precise dimensions and VMMGR, even correcting human errors.
• ScenePilot: Controllable Boundary-Driven Critical Scenario Generation for Autonomous Driving — ScenePilot generates critical, physically solvable autonomous driving scenarios that expose autonomy stack failures.
• CLEAR: Cognition and Latent Evaluation for Adaptive Routing in End-to-End Autonomous Driving — CLEAR achieves efficient, multi-modal autonomous driving by combining single-step generative planning with LLM-driven cognitive reasoning.
• Planning with the Views via Scene Self-Exploration — Self-exploration and view graph distillation enable VLMs to plan multi-turn camera movements in 3D environments.
• Xiaomi EV’s Next Self-Driving AI Explained — Autonomous driving models can achieve superior performance by shifting from linguistic reasoning to a latent spatial-temporal mathematical representation.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.