EMBODIEDDATA DOCUMENTATION
Open standards for embodied data.
Practical documentation for indexing, publishing, and consuming structured robotics datasets.
What is EmbodiedData?
EmbodiedData is an open catalogue and visualization layer for robotics demonstration datasets. It makes embodiment, observations, actions, tasks, quality signals, and licensing legible before a researcher downloads hundreds of gigabytes.
Core principle
A video can show a task. A robot-learning dataset must also explain what the robot observed, what it did, when it did it, and whether it succeeded.
Quick start
- Explore datasets by task, robot, sensor, or format.
- Open a dataset page and review its source type and action representation.
- Inspect sample episodes and validation results.
- Confirm the licence and cite the release.
- Download from the contributor’s approved data source.
Metadata schema
The schema organizes general information, robot configuration, task definition, dataset statistics, sensor streams, action representation, quality, limitations, and permissions.
{
"source_type": "robot_teleoperation",
"task": { "primary": "pick_and_place" },
"robot": { "model": "Franka Panda", "dof": 7 },
"actions": { "type": "cartesian_delta", "frequency_hz": 20 },
"observations": ["rgb", "depth", "joint_state"]
}