ajaysri
Episodic box search: robot demonstrations with SAM3 home points
45 bimanual ARX robot sessions with 638 primitive episodes and 118,299 frames at 30 Hz for visual-goal-conditioned low-level policy learning. The robot opens boxes, displays contents, returns boxes to home positions, and empties requested objects into a tray.
Downloads73
Episodes638
Why This Matters for Physical AI
This dataset enables training of visual-goal-conditioned robotic policies with automatic geometric annotation and primitive-level task decomposition, advancing embodied AI's ability to perform complex multi-step manipulation tasks with visual feedback.
Technical Profile
- Modalities
- rgbproprioception
- Robot Embodiments
- ARX
- Action Space
- joint_positions
- Environment
- lab
- Task Types
- manipulationgraspingobject_retrieval
- Episodes
- 638
- Data Format
- LeRobot
- Annotation Types
- language_instructionsaction_labelsbounding_boxes
Access
Need custom rgb data?
Claru builds purpose-built datasets for lab applications with dense human annotations and quality assurance.
Request a Sample Pack