---
license: other
pretty_name: Robotics PushT LeRobot
tags:
- lerobot
- robotics
- imitation-learning
- multimodal-behavior-cloning
---

# PushT LeRobot Dataset

LeRobot-formatted PushT robot pushing dataset used for multimodal behavior cloning experiments.

## Dataset Details

- Format: LeRobot v3.0
- Robot type: `pusht_sim`
- Episodes: 206
- Frames: 25650
- FPS: 10
- Tasks: 1
- Action shape: `[2]`
- State shape: `[24]`
- Videos: yes

The repository follows the standard LeRobot layout with `meta/`, `data/`, and, when available, `videos/` at the repository root.

## Loading

```python
from lerobot.common.datasets.lerobot_dataset import LeRobotDataset

dataset = LeRobotDataset("nct-tso/robotics_pusht_lerobot")
```

## Provenance and License

This release is a LeRobot-formatted adaptation of an upstream robotics benchmark dataset.

Upstream benchmark/source: PushT benchmark from the Diffusion Policy robotics benchmark lineage.

The Hub metadata uses `license: other` because the release combines converted LeRobot metadata/data files with upstream benchmark assets. The upstream benchmark assets retain their original licenses and attribution requirements. See `LICENSE` for the repository-level notice.

## Citation

If you use this dataset release, please cite both:

1. The LeRobot adaptation / dataset release:

   **VersatIL: A Unified, Composable Imitation Learning Library for Reproducible Autonomous Robot Policies**  
   Lorenzo Mazza, Ariel Rodriguez, Stefanie Speidel.  
   Citation details coming soon.

2. The relevant upstream benchmark papers or repositories for the original environment, task definition, and data source.