THULab/cucumber_subtask_grab_dagger_iter2_task_unified

Cucumber Subtask Grab DAgger Iteration 2 (TsFile)

Apache TsFile version of VibeCuisine/cucumber-subtask-grab-DAgger-iter2-task-unified. The source is a LeRobot v3.0 vibeboard_follower_tilt robot dataset.

Overview

The dataset contains one train split with robot trajectories from a cucumber subtask-grab DAgger collection. Source metadata reports 150 episodes, 13,484 frames, one task index, and a 20 FPS sampling rate. The materialized collection was built from the source components VibeCuisine/cucumber-subtask-grab-DAgger-iter1-v1-trim and VibeCuisine/jetson1-062526-grab-doris.

• Episodes: 150 (episode range 0:150)
• Rows: 13,484
• Sampling rate: 20 Hz
• Converted files: 1 TsFile

Schema (TsFile structure)

• Time (INT64, milliseconds) - round(timestamp * 1000), restarted for each episode.
• episode_index (TAG, INT64) - source episode identifier; query an episode with WHERE episode_index = 0.
• task_index (TAG, INT64) - source task identifier.
• frame_index, sample_index (FIELD, INT64) - frame position and source index (renamed to sample_index).
• action_0..action_6 (FIELD, FLOAT) - seven action values in the source order shoulder_pan.pos, shoulder_lift.pos, elbow_flex.pos, wrist_flex.pos, wrist_roll.pos, gripper.pos, tilt.pos.
• observation_state_0..observation_state_6 (FIELD, FLOAT) - seven robot state values in the same source order.

The source timestamp field is omitted because it is represented by Time. Camera features (observation.images.top, observation.images.wrist, and observation.images.base) are not converted; their videos remain in the original dataset's videos/ directory.

Usage

Read data/cucumber_subtask_grab_dagger_iter2_task_unified_train.tsfile with the Apache TsFile Java or Python SDK. The table name is cucumber_subtask_grab_dagger_iter2_task_unified_train.

Source & license

• Original dataset: VibeCuisine/cucumber-subtask-grab-DAgger-iter2-task-unified
• Author / publisher: VibeCuisine
• Framework: LeRobot
• License: Apache-2.0