Sssapollo/data_lerobot_blue_cube_165_trials

Blue Cube To Green Plate LeRobot Dataset

This dataset combines the previous blue-cube training set with the July 9, 2026 no-obstacle and obstacle recordings for pi05_droid_jointpos fine-tuning.

• Prompt: pick up the blue cube and place it onto the green plate
• FPS: 30
• Observation state: [q1..q7, gripper_width]
• Action: next-frame joint-position target [q1..q7, gripper_width]
• Cameras: observation.images.exterior_image_1_left, observation.images.wrist_image_left, observation.images.exterior_image_2_right
• Episodes: 165
• Frames: 70085

Sources

• Previous blue-cube dataset: 110 trials
• blue_green_no_obs_20260709_141129: 25 trials
• blue_green_obs_20260709_145500: 12 trials
• blue_green_obs_20260709_150614: 10 trials
• blue_green_obs_20260709_153500: 4 trials
• blue_green_obs_20260709_154051: 4 trials

Robot rows are aligned with tail alignment. In tail mode, the last video frame maps to the last robot state and earlier frames are sampled backward using 1000 Hz robot state frequency and 30 FPS video frequency.

Gripper Postprocessing

The July 9 recordings were postprocessed with the local gripper interpolation pipeline. Close and terminal release phases are shaped over 0.5 seconds so the gripper width has clear, smooth grasp and release actions.