5position / README.md
smilegeng's picture
Upload README.md with huggingface_hub
4150073 verified
|
Raw
History Blame Contribute Delete
2.91 kB
---
license: apache-2.0
task_categories:
- robotics
tags:
- LeRobot
- so100
- tutorial
- time-series
- tsfile
- robot
- imitation-learning
- timeseries
configs:
- config_name: default
data_files:
- split: train
path: data/5position.tsfile
---
# 5position
This dataset is a TsFile conversion of
[`DAbraham/5Position`](https://huggingface.co/datasets/DAbraham/5Position), a
LeRobot SO100 robot dataset. The original dataset was created using
[LeRobot](https://github.com/huggingface/lerobot).
## Source Dataset
- Source dataset: [`DAbraham/5Position`](https://huggingface.co/datasets/DAbraham/5Position)
- License: apache-2.0
- Robot type: SO100
- Episodes: 50
- Frames: 43,346
- Tasks: 1
- Sampling rate: 30 fps
- Original data path: `data/chunk-{episode_chunk:03d}/episode_{episode_index:06d}.parquet`
- Original video path: `videos/chunk-{episode_chunk:03d}/{video_key}/episode_{episode_index:06d}.mp4`
## TsFile Conversion
The numeric frame data was converted to one TsFile:
```text
data/5position.tsfile
```
All episodes are stored in the same TsFile table, with the original source
columns `episode_index` and `task_index` declared as TAG columns. To query one
episode, filter by `episode_index`.
`Time` is generated as `round(timestamp * 1000)` in milliseconds and restarts per
episode. The source `timestamp` column is dropped because it is represented by
`Time / 1000` seconds. The source `frame_index` column is kept.
The source `index` column is renamed to `sample_index`.
Vector columns are flattened while preserving the source column name:
- `action` -> `action_0` ... `action_5`
- `observation.state` -> `observation_state_0` ... `observation_state_5`
Flattened vector values are stored as single-precision FLOAT fields.
## Schema
Roles in the converted TsFile:
- TIME: `Time`
- TAG: `episode_index`, `task_index`
- FIELD: `frame_index`, `sample_index`, `action_0` ... `action_5`,
`observation_state_0` ... `observation_state_5`
The converted dataset contains 43,346 rows and 17 columns.
## Video Policy
The original dataset contains three video streams:
- `observation.images.context`
- `observation.images.arm`
- `observation.images.base`
Videos are not included in this converted repository. They remain available in
the original Hugging Face dataset:
https://huggingface.co/datasets/DAbraham/5Position/tree/main/videos
## Metadata
The original `meta/` files are mirrored in this repository. `meta/info.json` is
updated so `data_path` points to `data/5position.tsfile`, and it includes a
`tsfile_conversion` object documenting the source dataset, converted data path,
time mapping, TAG columns, flattened fields, dropped fields, omitted video
features, and video alignment policy.
## Citation
The original dataset card does not provide a BibTeX citation.