This repository serves as a curated list of Vision-Language-Action (VLA) models for Autonomous Driving (AD), complementing a survey paper on the topic. It targets researchers and developers in autonomous driving and multimodal AI, providing a structured overview of the evolving VLA4AD landscape, from explanatory perception to end-to-end control.

How It Works

The repository categorizes VLA4AD models into four paradigms: VLM as Explainers, Modular VLA4AD, End-to-End VLA4AD, and Reasoning-Augmented VLA4AD. It details the progression from simple language explanations to complex systems that integrate vision, language, and action for instruction understanding, reasoning, and vehicle control, often leveraging large language models (LLMs) and diffusion models.

Quick Start & Requirements

• Primary install / run command: git clone https://github.com/JohnsonJiang1996/Awesome-VLA4AD.git followed by cd Awesome-VLA4AD.
• Prerequisites: None explicitly listed beyond standard Git. The repository itself is a collection of links and information, not executable code.
• Setup time: Minimal, as it's a reference list.
• Links: arXiv, GitHub Stars, GitHub Forks.

Highlighted Details

• Comprehensive categorization of VLA4AD models into four distinct paradigms.
• Links to numerous research papers (many with code) published between 2023-2025.
• Curated list of relevant datasets and benchmarks for VLA4AD tasks.
• Includes citation details for the accompanying survey paper.

Maintenance & Community

• The repository is actively maintained, indicated by recent paper additions (2025).
• Contributions are welcomed via issues or pull requests.
• Contact emails are provided for questions and suggestions.

Licensing & Compatibility

• The repository itself is likely under a permissive license (e.g., MIT, Apache 2.0, common for "awesome" lists), but the licenses of linked projects vary.
• Compatibility for commercial use depends on the licenses of the individual projects linked within the list.

Limitations & Caveats

This repository is a curated list of resources and does not contain executable code for VLA4AD models. Users must individually locate, install, and configure the specific models and datasets they wish to use, each with its own set of dependencies and requirements.