Question 1

What is FoodforThought?

Accepted Answer

FoodforThought is an open-source robotics data infrastructure for organizing and sharing robot learning datasets. It supports VLA (Vision-Language-Action) training, diffusion policies, and behavioral cloning with full data lineage tracking from raw demonstrations to trained models.

Question 2

Is FoodforThought compatible with LeRobot?

Accepted Answer

Yes! FoodforThought natively supports LeRobot format import and export. You can import datasets directly from HuggingFace Hub and export your datasets in LeRobot-compatible format for training with ACT, Diffusion Policy, and other methods.

Question 3

What robots does FoodforThought support?

Accepted Answer

FoodforThought supports a wide range of robots including humanoid platforms (Figure, Unitree H1/G1, Boston Dynamics Atlas), manipulators (Franka, UR5, WidowX, ALOHA), and mobile robots (Spot, Fetch, TurtleBot). We also support egocentric and factory data.

Question 4

What training methods does FoodforThought support?

Accepted Answer

FoodforThought is training-framework-agnostic: it exports datasets in RLDS and LeRobot-compatible formats with the metadata trainers need (action space, observation types, episode statistics), so you can feed them into your own training stack for methods like OpenVLA, Octo, RT-X, or Diffusion Policy. It does not ship training-recipe templates for those methods.

Question 5

How do I contribute datasets to FoodforThought?

Accepted Answer

You can contribute datasets by signing up for a free account and using our upload page. We support multiple formats including HDF5, RLDS, ROS bags, Zarr, and LeRobot format. All contributions are credited to you.

mir500

mir500

Physical Specifications

Dimensions

Performance

Environment