chess_game_004_white (TsFile)

This dataset is an Apache TsFile conversion of Chojins/chess_game_004_white, a LeRobot robotics dataset for moving blue chess pieces according to highlighted squares.

Source Dataset

• Original dataset: Chojins/chess_game_004_white
• License: apache-2.0
• Source task category: robotics
• Source tags: LeRobot, chess, game
• LeRobot codebase version: v2.1 in downloaded meta/info.json
• Robot type: SO100
• Episodes: 50
• Frames: 23,035
• Tasks: 1
• Task: Move the blue chess pieces as indicated by the highlighted squares
• FPS: 30
• Source data path: data/chunk-{episode_chunk:03d}/episode_{episode_index:06d}.parquet
• Source video path: videos/chunk-{episode_chunk:03d}/{video_key}/episode_{episode_index:06d}.mp4
• Source video streams: observation.images.laptop and observation.images.phone, both 480x640 RGB AV1 videos at 30 fps.

Converted Data

• TsFile: data/chess_game_004_white.tsfile
• Table name: chess_game_004_white
• Rows converted: 23,035
• Episode count: 50
• Time precision: milliseconds
• Time mapping: Time = round(timestamp * 1000), restarting per episode
• TAG columns: episode_index, task_index
• Videos are not included in this repository. Use the original dataset's videos/ directory for visual streams.

Schema Mapping

Source column	Converted column(s)	Role	Notes
timestamp	Time	TIME	Converted from seconds to integer milliseconds and not retained as a duplicate field.
episode_index	episode_index	TAG	Source episode index.
task_index	task_index	TAG	Source task index.
frame_index	frame_index	FIELD	Source frame index.
index	sample_index	FIELD	Renamed to avoid generic index naming.
action	action_0 ... action_5	FIELD	Flattened float32 vector.
observation.state	observation_state_0 ... observation_state_5	FIELD	Flattened float32 vector.
observation.images.laptop	omitted	video	Not converted or uploaded; available in the source dataset.
observation.images.phone	omitted	video	Not converted or uploaded; available in the source dataset.

The six action/state dimensions follow the source feature names: main_shoulder_pan, main_shoulder_lift, main_elbow_flex, main_wrist_flex, main_wrist_roll, and main_gripper.

Conversion Notes

• The source LeRobot parquet episodes were merged into one TsFile table. Episodes remain queryable through the source TAG columns episode_index and task_index.
• Vector columns preserve the full source column name when flattened: . is replaced with _, and the element index is appended.
• The source timestamp column is dropped because it is redundant with Time / 1000 seconds.
• meta/ is mirrored from the source, with meta/info.json updated so data_path points to the converted TsFile and tsfile_conversion documents the mapping.
• No source numeric rows were dropped. Local validation confirmed 23,035 staged rows and 23,035 TsFile metadata rows.

Minimal Read Example

from tsfile import TsFileReader
from tsfile import tag_eq

path = "data/chess_game_004_white.tsfile"
reader = TsFileReader(path)

# Inspect table schemas.
print(reader.get_all_table_schemas().keys())

# Query numeric fields. To focus on one episode, pass a tag filter such as
# tag_filter=tag_eq("episode_index", "0") if supported by your SDK version.
with reader.query_table(
    "chess_game_004_white",
    ["frame_index", "sample_index", "action_0", "action_1", "observation_state_0"],
    batch_size=1024,
) as rs:
    batch = rs.read_arrow_batch()
    print(batch)

reader.close()