This research introduces a novel autonomous driving assistance system that integrates vision capabilities with LLM reasoning to improve decision-making in challenging road situations.

• Combines YOLOv4 and Vision Transformer (ViT) for comprehensive visual feature extraction
• Leverages GPT for advanced reasoning about spatial relationships in driving scenarios
• Outperforms traditional autonomous systems in complex, unexpected scenarios
• Addresses critical security concerns by enhancing decision quality in potentially dangerous driving situations

This approach represents a significant advancement for vehicle safety systems by enabling more human-like understanding of road conditions, improving trust in autonomous technology, and potentially reducing accidents in edge cases where traditional systems fail.

Vision-Integrated LLMs for Autonomous Driving Assistance: Human Performance Comparison and Trust Evaluation