02

The HuggingFace for Robotics Is HuggingFace

Robotics-First

Policies and datasets do well on HuggingFace. What's missing is everything around them.

This is Festivus →

Our Thesis

Many people talk about building "the HuggingFace for robotics." We believe that already exists. HuggingFace is great at hosting models and datasets. We don't compete with that. Festivus is the layer that covers what HF, arXiv, and GitHub don't: the robotics-native connective tissue.

What's Missing

These are the gaps not covered by arXiv, GitHub, or HuggingFace:

1. 1.Playing with simulations without installing MuJoCo or Isaac Sim. No download, no conda env, no GPU driver dance.
2. 2.Seeing what a policy actually does across different robot embodiments. Does ACT work on SO-100 AND ALOHA 2? Side by side.
3. 3.Comparing five manipulation policies on the same task. Which one actually transfers to real hardware?
4. 4.Validated URDFs. Model cards don’t specify action space or control frequency. 16,000+ robotics models on HF with no robotics-native metadata.
5. 5.Deploy readiness scores. Can I actually run this on my robot today, or is it research-only?
6. 6.Images, 3D viewers, simulation previews. Physical AI lives in visual media, not text READMEs.
7. 7.The 70% of a robot project that doesn’t live in code: BOMs, wiring diagrams, calibration notes, deployment logs.

What We Build