Overview — Awesome Physical AI Resources & Roadmap
This site is a navigation layer around the root README.md. The README remains the source of truth for the curated catalog.
Awesome Physical AI is a curated, engineering-oriented map of Physical AI resources — a Physical AI roadmap for researchers, ML and robotics engineers, and technical leaders working on robot learning, embodied AI, embodied agents, robotics simulation, sim-to-real transfer, world models, vision-language-action (VLA) models, robotics foundation models, generalist robot policies, Physical AI benchmarks, and production-grade, safe embodied AI systems. The repository content analyzed for this site centered on README.md, CONTRIBUTING.md, LICENSE, and small GitHub workflow files under .github/workflows. That means the project architecture is editorial, not runtime-oriented.
Navigation layers
There are two views over the same catalog, and they intentionally do not match line-for-line:
- The README headings (~7 top-level sections) are optimised for scanning a single long Markdown file on GitHub. They group related categories together (for example, all simulation work under one heading).
- The 14 docs-site categories (the sidebar on the left under Categories) are the authoritative taxonomy for curation, contribution, and the entry-count CI check. Each docs-site category corresponds to one or more sections in the README, and
CONTRIBUTING.mdis written against this 14-category structure.
When the two diverge — for example, when the README places "Manipulation" and "Locomotion" under a single umbrella while the docs site splits them — the docs-site taxonomy wins, and the README will be reorganised on the next curation pass. The task table below adds a third, non-authoritative overlay (Learn / Build / Deploy / Measure / Track / Practice / Related) to help newcomers route into the 14 canonical categories quickly.
Top-level map
The README is organised into 14 canonical categories plus a set of supporting appendices (reading material, hardware, community, project ladder). The table below groups those canonical categories by task so you can route to the right section quickly. The grouping is a docs-site overlay — the README itself does not contain "Learn", "Build", or "Deploy" headings, so each link below points to a real category anchor. The 14 categories themselves remain the authoritative taxonomy described above.
| Task | Canonical categories | Use it when |
|---|---|---|
| Learn | Courses, and the appendices Books, Tutorials & Guides, Key Papers, Survey Papers | You need structured learning material or a paper-reading path. |
| Build | Robotics Foundation Models, World Models, Simulators, Manipulation, Locomotion | You are comparing model families, simulators, or task-specific methods. |
| Deploy | Sim-to-Real, Safety & Robustness, Evaluation Methodology, Production Patterns / Reference Architectures, Governance & Policy, and the appendix Hardware Platforms | You are moving from research tooling toward a real system and need transfer, evaluation, safety, or operational references. |
| Measure | Benchmarks, Datasets | You need a comparable evaluation harness or training data. |
| Track the field | Companies, and the appendices Conferences, Community, Newsletters & Blogs, People to Follow | You want a continuing information feed instead of a one-time reading list. |
| Practice | Getting Started Projects — Beginner, Intermediate, Advanced, Hardware Projects | You want a staged practice plan rather than a flat list of links. |
| Related | Related Awesome Lists | You need adjacent catalogs beyond this repository's scope. |
If you would rather browse the canonical categories directly without the task overlay, see the Categories index.
Why a standalone site exists
The root README is already dense and useful. It also carries the entire catalog in one file. The job of this site is to reduce time-to-orientation:
- route readers to the right canonical category quickly, with a task-based overlay where it helps
- surface the contribution and curation policy without forcing a separate file read
- make the repository structure legible without pretending there is an application runtime to document
What this site will not do
Because the source repository is not a software package, this site deliberately avoids invented sections for installation, runtime configuration, integrations, or production architecture that do not exist in the tracked files.