keon/awesome-physical-ai

What it solves

Physical AI is the study of AI systems that interact with the physical world through robots. The field is evolving quickly, with many new papers, models, and resources appearing constantly. This project solves the problem of keeping track of all that work by curating a structured, categorized list of the most important academic papers, models, datasets, and tools.

How it works

The repository is a classic "Awesome List" — a curated Markdown file that organizes entries under clear headings. It covers areas like Vision-Language Backbones, VLA architectures, world models, learning paradigms, scaling and generalization, deployment, safety, and applications. Each entry typically includes the paper title, venue, links to the paper, code, or project page, and sometimes a short note. The list is free to browse and is meant to serve as a quick reference.

Who it’s for

This is for anyone working in or entering the field of Physical AI, including researchers, graduate students, engineers, and hobbyists who want a high-level map of the landscape. It’s also useful for planning research directions, finding relevant prior work, and discovering open-source implementations.

Highlights

• Covers the full spectrum: from foundational vision-language models to end-to-end VLA architectures, world models, and deployment considerations.
• Each entry often links to the paper, code, and project, making it easy to dive deeper.
• Highlights both established works (e.g., RT-1, RT-2, PaLM-E) and very recent breakthroughs (e.g., π0.5, Gemini Robotics 1.5).
• Includes resources like datasets, benchmarks, simulation platforms, and companies, giving a practical starting point.