THULab/putcab_ab_randomdelta_train50

PutCab-AB-RandomDelta-Train50 (TsFile)

Apache TsFile version of Shiki42/PutCab-AB-RandomDelta-Train50.

Overview

A timing-balanced 50-episode subset of RoboTwin PutCab-AB demonstrations, recorded on a simulated ALOHA bimanual robot. Selection is conditioned on the normalized reference offset u_ref = (delta + T_B_ref) / (T_A_ref + T_B_ref) (where delta is right/B start minus left/A start): ten intervals [0,.1) ... [.9,1) each contribute five trajectories, giving 25 negative and 25 positive deltas. Reused trajectories keep independent active-arm v3 timing; the new V2 trajectories add recorded feedback waits (safe-waypoint lift, drawer-readiness re-check). observation.state holds 14 joint/gripper drive targets and action is the next observation target with the terminal target repeated. Physics runs at 250 Hz with observations every 15 ticks, i.e. 50/3 fps.

• Robot: ALOHA (simulated, bimanual)
• Episodes: 50 · Frames (rows): 11,614 · Tasks: 31
• Sampling rate: 16.666666666666668 fps
• Cameras: three 640x480 RGB cameras
• Converted TsFile: 11,614 rows in a single data/putcab_ab_randomdelta_train50.tsfile, one device per episode_index (WHERE episode_index=0 selects episode 0).

Schema (TsFile structure)

• Time (INT64, milliseconds) — round(timestamp * 1000); restarts at 0 for each episode.
• episode_index (TAG) — episode / device dimension.
• task_index (TAG) — task dimension (constant 0 here).
• frame_index (FIELD, INT64) — source frame counter; sample_index (FIELD, INT64) — source index column (renamed).
• observation.state_{0..13} (FIELD, FLOAT) — joint/gripper drive targets (7 per arm)
• action_{0..13} (FIELD, FLOAT) — next observation target (7 per arm)

Vector columns were flattened to scalar fields: a . in the source column name became _ and the element index is appended (e.g. observation.state → observation_state_0 ... observation_state_{n-1}). Values are single-precision FLOAT. The source timestamp column is dropped because it equals Time ÷ 1000 seconds. No other columns or rows were removed.

Usage

Install the Apache TsFile Python SDK (pip install tsfile) and read a converted file:

from pathlib import Path
from tsfile import TsFileReader

path = Path("data/putcab_ab_randomdelta_train50.tsfile")
with TsFileReader(str(path)) as reader:
    schemas = reader.get_all_table_schemas()
    print("tables:", list(schemas))
    table_name = next(iter(schemas))
    table = schemas[table_name]
    columns = [column.get_column_name() for column in table.get_columns()]
    print("columns:", columns)
    field_names = [
        column.get_column_name()
        for column in table.get_columns()
        if column.get_column_name() not in {"Time", "time"}
    ]
    if field_names:
        with reader.query_table(table_name, field_names[:3], batch_size=1024) as result:
            batch = result.read_arrow_batch()
            if batch is not None:
                print(batch.to_pandas().head())

Source & license

• Original dataset: https://huggingface.co/datasets/Shiki42/PutCab-AB-RandomDelta-Train50
• Author / publisher: Shiki42
• License: not declared by the original dataset; please defer to the original.
• Camera video streams are not included in this repository; they remain at the original dataset's videos/ directory.