JeffrinSamapache-2.0
G1_Brainco_PickDrink_Dataset (2x speed, 24fps, Cosmos Policy-ready)
Resampled teleoperated demonstration dataset for the Unitree G1 humanoid robot with BrainCo dexterous hands performing pick-and-drink tasks, converted to 2x playback speed and 24fps for Cosmos Policy world model training.
Downloads5
Episodes201
Why This Matters for Physical AI
This dataset provides high-quality teleoperated demonstrations of complex dexterous manipulation tasks with a humanoid robot, enabling training of world models for embodied AI that can predict and reason about fine-grained hand control and manipulation.
Technical Profile
- Modalities
- rgbproprioception
- Robot Embodiments
- Unitree G1BrainCo dexterous hand
- Action Space
- joint_positions
- Environment
- lab
- Task Types
- manipulationpick_and_placedexterous_manipulation
- Episodes
- 201
- Data Format
- LeRobot
- Annotation Types
- action_labels
- License
- apache-2.0
Access
Need custom rgb data?
Claru builds purpose-built datasets for lab applications with dense human annotations and quality assurance.
Request a Sample Pack