Festivus
Open source robot policies,
datasets, and benchmarks.
“for the Rest of Us”
Festivus is the open catalog of robot policies, datasets, environments, and benchmarks for . Browse what works, compare across hardware, and contribute what you have.
The catalog
Seven dimensions. One catalog.
01
Robots
Hardware fingerprints. Manipulators, humanoids, quadrupeds, mobile bases.
02
Policies
Open source robot policies. ACT, diffusion, RL, vision-language-action.
03
Datasets
Robotics datasets in LeRobot and HuggingFace formats. Provenance attached.
04
Environments
Sim and real environments. MuJoCo, Isaac, real-world rigs.
05
Tasks
Pick-and-place, folding, locomotion, navigation. The vocabulary of physical AI.
06
Papers
arXiv-linked physical AI papers tied to the policies that ship from them.
07
Deploy notes
What it took to get a policy off the bench and onto a real robot.
Why a catalog
Physical AI is scattered.
Policies live on HuggingFace, GitHub, and lab websites that go offline.
Datasets ship in five different schemas. Half lack hardware metadata.
Benchmarks compare apples to oranges. Different rigs, different metrics.
Nobody knows which policy ran on which robot, on which task, with which result.
How to use it
From task to running policy.
- 01
Pick a task
Folding laundry, transferring a cube, walking a quadruped. Start from the goal.
- 02
See compatible policies
Filtered by hardware. Every policy lists the robots it's been validated on.
- 03
Inspect the dataset
Provenance, episode count, sensor coverage, license. No CSV soup.
- 04
Read the deploy notes
What broke. What worked. What the next team should try first.
- 05
Run it
LeRobot weights, HuggingFace links, evaluation environments. One click.
Who uses Festivus
Built for the people shipping physical AI.
Robotics engineers
Find a policy that runs on your hardware. Skip the LeRobot model hunt.
AI researchers
Compare benchmarks across policies. See what beats what, on which task.
Lab leads
Stand up a reproducible stack: dataset → policy → environment → deploy.
Open source contributors
Fix gaps in the catalog. Add a robot, a policy, a deploy note. Get attribution.
Open by default
Papers stay on arXiv. Code stays on GitHub. Models stay on HuggingFace.
Festivus links the catalog. It doesn't lock it up. Every policy, every dataset, every paper points back to the source. Credit and provenance stay clear.
Fork it. Clone it. Remix it. Then push your contribution back so the next team finds it.
Open source robot policies.
All in one catalog.
Browse the policies, datasets, and benchmarks the physical AI community actually ships. Then contribute what you have.