huzheyuanapache-2.0
3lego whole-episode sessions: round1_baseline
Dual-arm YAM robot block-sorting dataset with 103 whole-session episodes reconverted for in-context learning, featuring multi-cycle recordings under fixed block-to-tray permutations with RGB observations and cartesian/joint action spaces.
Downloads0
Episodes103
Hours4.01
Why This Matters for Physical AI
This dataset enables in-context learning research by preserving whole manipulation sessions with latent task structure, allowing models to infer block-sorting permutations from visual observations rather than explicit task labels.
Technical Profile
- Modalities
- rgbproprioceptionlanguage
- Robot Embodiments
- Dual-YAM
- Action Space
- end_effector_delta
- Environment
- lab
- Task Types
- manipulationpick_and_placesorting
- Episodes
- 103
- Total Hours
- 4.01
- Data Format
- LeRobot
- Annotation Types
- language_instructionsaction_labelssubtask_segmentation
- 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