bridgedatav2_train

Converted from the Hugging Face dataset ZibinDong/bridgedatav2_train to Apache TsFile format.

Modalities: Time-series.

Source Dataset

The source dataset is BridgeDataV2 in LeRobot Dataset V3.0 format. The original dataset page links to BridgeDataV2 at https://rail.eecs.berkeley.edu/datasets/bridge_release/data/tfds/bridge_dataset/.

• License: apache-2.0
• Source scale: 61,529 episodes, 2,228,465 frames, 11,950 tasks, 5 fps
• Source robot type: WidowX250
• Source video stream: image

Converted Files

• data/bridgedatav2_train_1.tsfile
• data/bridgedatav2_train_2.tsfile
• data/bridgedatav2_train_3.tsfile
• meta/info.json: source metadata rewritten to describe the TsFile artifact
• meta/stats.json, meta/tasks.parquet, meta/episodes/chunk-000/file-000.parquet: source metadata mirrored for reference

Videos are not included in this repository. They remain in the original Hugging Face dataset under https://huggingface.co/datasets/ZibinDong/bridgedatav2_train/tree/main/videos.

TsFile Schema

• Time: millisecond timestamp, synthesized as round(timestamp * 1000).
• TAG columns: episode_index, task_index.
• FIELD columns: frame_index, sample_index, traj_idx, state_0..state_6, and action_0..action_6.

Vector columns are flattened by appending the element index. state[7] becomes state_0..state_6; action[7] becomes action_0..action_6. Vector values are stored as single-precision FLOAT.

The source timestamp column is dropped because it is redundant with Time. The source index column is renamed to sample_index. Source video feature column image is omitted from the TsFile table and remains available in the original dataset.

Read Example

Use the Apache TsFile Java or Python SDK to read the files under:

data/bridgedatav2_train_*.tsfile

The table name is bridgedatav2_train. Query a single episode by filtering the TAG:

WHERE episode_index = 0