zestcode5's picture
Add README with task list and source attribution
438e50d verified
|
Raw
History Blame Contribute Delete
1.81 kB
---
license: apache-2.0
task_categories:
- robotics
tags:
- LeRobot
- ur3
- manipulation
- imitation-learning
configs:
- config_name: default
data_files: data/*/*.parquet
---
# ur3-merged-5tasks-v2
A multi-task UR3 manipulation dataset in LeRobot v2.1 format, merged from
five source datasets covering pick-and-place, unstacking, multi-item
sorting, and pushing primitives.
## Tasks
| task_index | description |
|:----------:|:------------|
| 0 | pick up the pink cylinder and place it in the orange box |
| 1 | pick up the white glass and put on a brown coaster |
| 2 | Remove cup from nested cups |
| 3 | pick multiple items then put in the orange basket |
| 4 | Single-finger push to blue marker |
## Source datasets
This dataset was assembled by merging the following:
- [`zestcode5/ur3-pink-cylinder-6D`](https://huggingface.co/datasets/zestcode5/ur3-pink-cylinder-6D) - base; provides task 0
- [`zestcode5/ur3-multiple-task-trimmed-idle`](https://huggingface.co/datasets/zestcode5/ur3-multiple-task-trimmed-idle) - split into per-task subsets; provides tasks 1, 2, 4
- [`zestcode5/ur3-multiple-item-basket-trimmed`](https://huggingface.co/datasets/zestcode5/ur3-multiple-item-basket-trimmed) - provides task 3
The schema, fps, and camera configuration follow the base dataset.
## Format
- LeRobot v2.1
- Robot: UR3
- One parquet file per episode under `data/chunk-*/`
- One mp4 file per camera per episode under `videos/chunk-*/<camera>/`
- Episode metadata in `meta/episodes.jsonl`, task list in `meta/tasks.jsonl`,
per-episode stats in `meta/episodes_stats.jsonl`
## Loading
```python
from lerobot.common.datasets.lerobot_dataset import LeRobotDataset
ds = LeRobotDataset("zestcode5/ur3-merged-5tasks-v2")
print(ds.num_episodes, ds.num_frames, ds.fps)
sample = ds[0]
```