organize_table

LeRobot v3 dataset for Unitree G1 organize-table manipulation.

• 250 episodes, 87,840 frames, 30 FPS
• robot type: Unitree_G1_Inspire_Head
• visual input: full color_0 scene with the tracked target rendered as a black/white outline overlay
• control input: separate observation.mask_pick
• state/action dimension: 26
• no place mask and no gaze point from fine_init.json

The outlined overlay is the main visual observation. The binary pick mask is a separate feature used by the Gaze2Act DINOv3/control branch.