Behavioral Dataset: Multimodal Video, Depth & Actions
Human-teleoperated demonstrations of everyday household tasks that synchronize RGB video with quantized depth and instance segmentation from head and wrist cameras, shipped in LeRobot format so they drop into existing robot-learning pipelines.
Embodied AITeleoperationSegmentation
Overview
- 1.1
Human activity video with task-level context
- 1.2
Metadata suited for reviewing action labels and capture structure
- 1.3
Signals for evaluating scene diversity and multimodal alignment
Specifications
- License
- Non-exclusive license
- File types
- mp4
- Video format
- H.264
- Frame rate
- 30fps
- Resolution
- 720x720
- Depth
- 8-bit (quantized)
- Segmentation
- Instance ID maps
- Multi view
- Head & wrist cameras
- Format
- LeRobot
Metadata coverage
sample identifierssample titlesmedia previewsdurationmedia typePII statustagsfeature flagsmedia metadatamodality
