eval_robustness_e9_3_full_fk_local_240k

eval_robustness_e9_3_full_fk_local_240k (TsFile)

Apache TsFile version of nduque/eval_robustness_e9_3_full_fk_local_240k.

Overview

This is an evaluation-robustness teleoperation dataset recorded on a Koch robot arm and packaged in the LeRobot v2.1 layout. Each row is one control step: a 6-dimensional joint/gripper state, the 6-dimensional commanded action, and a 4-dimensional local end-effector pose. The source dataset card does not name the manipulation task; the metadata records a single task (task_index = 0).

• Robot: Koch
• Episodes: 177 · Frames (rows): 74,419 · Tasks: 1
• Sampling rate: 30 fps
• Cameras: observation.images.front and observation.images.above (720x1280x3, AV1)
• Converted TsFile: 74,419 rows in a single data/eval_robustness_e9_3_full_fk_local_240k.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).
• action_{0..5} (FIELD, FLOAT) — commanded joint/gripper targets
• observation.state_{0..5} (FIELD, FLOAT) — measured joint/gripper state
• observation.ee_pos_{0..3} (FIELD, FLOAT) — local end-effector pose (x, y, z, yaw)

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/eval_robustness_e9_3_full_fk_local_240k.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/nduque/eval_robustness_e9_3_full_fk_local_240k
• Author / publisher: nduque
• License: apache-2.0
• Camera video streams are not included in this repository; they remain at the original dataset's videos/ directory.