Summary

This repository curates academic papers and resources on Physical AI, focusing on AI systems that interact with the physical world via robotic embodiments. It targets researchers and engineers by providing a structured overview of Vision-Language-Action (VLA) models, world models, embodied AI, and robotic foundation models, serving as a centralized knowledge base to accelerate development in real-world perception, reasoning, and action.

How It Works

The project functions as a comprehensive, categorized bibliography, systematically organizing research across key sub-fields. It covers VLA architectures, world models, reasoning, learning paradigms, scaling, deployment, and safety. This structured approach allows users to navigate the rapidly evolving landscape of Physical AI, identifying foundational models, novel architectures, and emerging trends.

Quick Start & Requirements

This repository is a curated list of academic resources and does not involve direct software installation or execution. It serves as a knowledge base for researchers and developers.

Highlighted Details

• VLA Architectures: Encompasses monolithic end-to-end models (e.g., RT-1/2, OpenVLA, PaLM-E) and modular designs, alongside efforts in compact and efficient models (e.g., SmolVLA, TinyVLA).
• World Models: Explores predictive architectures (JEPA), generative simulators (Genie, Sora), and embodied agent-specific models.
• Action Representation: Details methods for discrete action tokens (FAST, GR-1/2) and continuous/diffusion policies (Diffusion Policy, Octo, π₀).
• Datasets & Benchmarks: Compiles crucial evaluation resources like Open X-Embodiment, LIBERO, and SIMPLER for training and benchmarking Physical AI systems.

Maintenance & Community

As a GitHub repository, maintenance relies on community contributions via pull requests. It encourages additions and corrections to its curated list, fostering a collaborative environment for tracking Physical AI advancements. Specific community channels are not listed.

Licensing & Compatibility

The repository itself, as a curated list, does not have a specific software license. Users must consult the licenses of individual cited academic papers, code repositories, and datasets. Commercial use or closed-source linking depends on the licensing of each referenced resource.

Limitations & Caveats

This resource is a snapshot of research and may not be exhaustive or perfectly up-to-date due to the rapid advancement of Physical AI. It guides users to research papers and projects rather than providing a deployable software framework. The utility of specific resources depends on direct engagement with the linked academic work.