---
license: apache-2.0
task_categories:
- robotics
tags:
- vision-language-action
- vlm
- zero-shot
---

This repository contains the OTTER dataset, used in the paper [OTTER: A Vision-Language-Action Model with Text-Aware Visual Feature Extraction](https://huggingface.co/papers/2503.03734).

**Project Page:** [https://ottervla.github.io/](https://ottervla.github.io/)

**Code:** [https://github.com/Max-Fu/otter](https://github.com/Max-Fu/otter)


**Dataset Download:**

The OTTER dataset is hosted on Hugging Face and is in TFDS format.  It supports pre-training on Open X-Embodiment.  You can download the datasets using the Hugging Face CLI:

```bash
# first install huggingface-cli
pip install -U "huggingface_hub[cli]"
# download the datasets
mkdir -p dataset
pushd dataset
huggingface-cli download mlfu7/icrt_pour --repo-type dataset --local-dir .
huggingface-cli download mlfu7/icrt_drawer --repo-type dataset --local-dir .
huggingface-cli download mlfu7/icrt_poke --repo-type dataset --local-dir .
huggingface-cli download mlfu7/icrt_pickplace_1 --repo-type dataset --local-dir .
huggingface-cli download mlfu7/icrt_stack_mul_tfds --repo-type dataset --local-dir .
huggingface-cli download mlfu7/icrt_pickplace --repo-type dataset --local-dir .
popd
```

A converted LeRobot version of the dataset is available for fine-tuning Pi0 ([here](https://huggingface.co/datasets/mlfu7/pi0_conversion)).  Fine-tuning scripts are provided [here](https://github.com/Max-Fu/openpi).