LLM-powered Driving Intelligence Evaluation Framework

Official implementation of the paper:
"A Comprehensive LLM-powered Framework for Driving Intelligence Evaluation"

Work in Progress: Full implementation will be uploaded shortly

Framework Overview

This repository contains the core components for automating driving intelligence evaluation using LLMs: ''' . ├── driving_context_to_des.py # Preprocessing: Driving context summarisation ├── evaluation_rag_auto.py # Main evaluation pipeline └── (Additional components pending upload) '''

Key Components

1. Driving Context Processor (driving_context_to_des.py)

• Function: Ingests raw driving context data and generates structured summaries
• Input: Time-series driving signals (steering, acceleration, etc.)
• Output: Natural language descriptions for LLM processing
• Features:
  • Event detection
  • Temporal pattern extraction

2. Automated Evaluation Engine (evaluation_rag_auto.py)

• Function: Performs end-to-end driving intelligence assessment
• Core Methods:
  • RAG-based knowledge retrieval from driving manuals
  • Multi-aspect evaluation (safety, intelligence, comfort)
  • Explainable scoring with LLM-generated feedback
• Output Formats:
  • Quantitative scores (0-10 scale)
  • Qualitative improvement suggestions
  • Evaluation reports