hk2392025apache-2.0
Vi2Act
A small synthetic video-to-action dataset for the Fourier GR1 humanoid robot, generated using Cosmos-Predict2 14B Video2World model and converted to LeRobot format.
Downloads0
Episodes15
Why This Matters for Physical AI
Demonstrates synthetic video-to-action dataset generation for humanoid robots using world models, enabling scalable creation of diverse manipulation training data without real-world collection.
Technical Profile
- Modalities
- rgbproprioceptionlanguage
- Robot Embodiments
- Fourier GR1
- Action Space
- joint_positions
- Environment
- simulation
- Task Types
- manipulationpick_and_placepouringgrasping
- Episodes
- 15
- Data Format
- LeRobot
- Annotation Types
- language_instructionsaction_labels
- License
- apache-2.0
Access
Need custom rgb data?
Claru builds purpose-built datasets for simulation applications with dense human annotations and quality assurance.
Request a Sample Pack