Datasets:

jzpeng
/

windfromthenorth_scripted_atomic_step_pose_0.6

license: apache-2.0
authors:
  - Andy Tang
task_categories:
  - robotics
tags:
  - LeRobot
  - lego-atomic-step
  - scripted
  - simulation
  - ur5
  - v3
  - blind
  - tsfile
  - timeseries
  - tabular
  - format:tsfile
modality:
  - tabular
  - text
  - timeseries
pretty_name: Scripted Atomic Step Pose 0.6 TsFile
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/windfromthenorth_scripted_atomic_step_pose_0_6.tsfile
size_categories:
  - 100K<n<1M

Scripted Atomic Step Pose 0.6 TsFile

This Apache TsFile dataset is based on windfromthenorth/scripted_atomic_step_pose_0.6, a LeRobot v2.1 simulation dataset containing UR5/WSG50 trajectories for placing a LEGO brick on an assembly board.

Source and Attribution

Original dataset: windfromthenorth/scripted_atomic_step_pose_0.6
Author and repository publisher: Andy Tang (windfromthenorth)
License: Apache-2.0
Task: place the lego brick on the assembly board
Robot type: ur5_wsg50_lego_atomic_step
LeRobot version: v2.1
Split: train (0:955)
Sampling rate: 20 fps
Scale: 955 episodes, 159,935 frames, 1 task, 955 source Parquet shards
Source layout: data/chunk-{episode_chunk:03d}/episode_{episode_index:06d}.parquet
Paper, external homepage, and completed citation: not provided by the source card

The TsFile table is windfromthenorth_scripted_atomic_step_pose_0_6 and contains 159,935 rows in data/windfromthenorth_scripted_atomic_step_pose_0_6.tsfile. Source metadata is retained under meta/, with meta/info.json updated for the TsFile table.

Schema and Mapping

Time = round(timestamp * 1000) milliseconds. Time starts at zero and advances in 50 ms steps within every episode.

TsFile column	Role	Type	Source mapping
`Time`	TIME	TIMESTAMP	`round(timestamp * 1000)` ms
`episode_index`	TAG	STRING	Original INT64 episode index
`task_index`	TAG	STRING	Original INT64 task index
`frame_index`	FIELD	INT64	Preserved
`sample_index`	FIELD	INT64	Renamed from `index`
`action_0` ... `action_6`	FIELD	FLOAT	Flattened from `action[7]`
`observation_state_0` ... `observation_state_19`	FIELD	FLOAT	Flattened from `observation.state[20]`

Dots in source names are replaced by underscores. The source timestamp is not repeated because it is represented by Time / 1000 seconds. The constant prompt string is not stored as a measurement; its value remains in meta/tasks.jsonl. No numeric row, episode, task, state dimension, or action dimension is omitted.

Videos

The original repository has no video location: its meta/info.json declares total_videos: 0 and video_path: null, and its remote file tree has no videos/ directory. Therefore this TsFile dataset contains no videos. The numeric rows are indexed by episode_index and frame_index for any later external alignment.

Minimal Read Example

from tsfile import TsFileReader

reader = TsFileReader("data/windfromthenorth_scripted_atomic_step_pose_0_6.tsfile")
with reader.query_table(
    "windfromthenorth_scripted_atomic_step_pose_0_6",
    ["episode_index", "task_index", "frame_index", "sample_index",
     "action_0", "observation_state_0"],
    batch_size=65536,
) as result:
    batch = result.read_arrow_batch()
    print(batch.to_pandas().head())
reader.close()

Citation

The source card does not provide a paper or completed BibTeX entry. Cite the original Hugging Face dataset and Andy Tang (windfromthenorth) when using this dataset.