This project explores rethinking autonomous driving by integrating Large Language Models (LLMs) to enable human-like decision-making in driving scenarios. It targets researchers and developers interested in AI-driven autonomous systems, offering a novel approach to closed-loop driving control.

How It Works

The system leverages LLMs to process driving context and make decisions. It utilizes a tool-use paradigm where the LLM can interact with simulated driving environments. The core advantage lies in enabling LLMs to understand and respond to complex driving situations with common-sense reasoning, mimicking human driving behavior.

Quick Start & Requirements

• Install dependencies: pip install highway-env and pip install -r requirements.txt.
• Configure LLM API keys and model details in config.yaml (supports Azure OpenAI and OpenAI).
• Run the simulation: python HELLM.py.
• Requires an LLM API key (GPT-3.5 with 8k+ tokens recommended).
• Demo available on Hugging Face Spaces.

Highlighted Details

• Demonstrates closed-loop interaction ability in driving scenarios.
• Focuses on LLM reasoning with common sense for driving decisions.
• Explores performance enhancement through memorization capabilities.
• Built upon highway-env and LangChain.

Maintenance & Community

The project is associated with PJLab-ADG. Contact is via GitHub issues.

Licensing & Compatibility

The README does not explicitly state a license. Compatibility for commercial use or closed-source linking is not specified.

Limitations & Caveats

The project is presented as research and may require significant configuration for specific LLM providers. The reliance on external LLM APIs introduces potential costs and latency.