---
license: cc-by-4.0
task_categories:
- robotics
tags:
- lerobot
- imitation-learning
- robotics
- ur3e
- bimanual
pretty_name: "salt - WetLabRoboData imitation-learning dataset"
---

# salt

Teleoperation demonstrations for the **salt** task, part of the
[WetLabRoboData / ChemLab](https://huggingface.co/WetLabRoboData) imitation-learning
benchmark (UR3e bimanual arm, 3 cameras, GELLO teleop).

| Field                  | Value |
|------------------------|-------|
| Category               | short-horizon |
| Train episodes         | 65 |
| Mean episode length    | 56.9 s |
| Robot                  | UR3e bimanual |
| Cameras                | front, left_wrist_top, right_wrist_top |
| Teleop                 | GELLO |
| Format                 | LeRobot v3 |

## Related

- Scratch policy: [`WetLabRoboData/diffusion-salt-scratch`](https://huggingface.co/WetLabRoboData/diffusion-salt-scratch)

- Full task-collection landing page: [WetLabRoboData collections](https://huggingface.co/WetLabRoboData#collections)

## Load

```python
from lerobot.datasets import LeRobotDataset
ds = LeRobotDataset("WetLabRoboData/lerobot-data-salt")
```

## Provenance

Reorganized from `WetLabRoboData/lerobot-data-smrithi-salt_20260725` on 2026-10-04. The original
raw mirror (data + checkpoints + eval all in one repo) is preserved in the
`old/` subfolder of that source repo for traceability.