ad1t7a/UMI-Fanqi-Processed-Task-Dataset

---
task_categories:
- robotics
tags:
- code
size_categories:
- 100B<n<1T
---
# Robotic Manipulation Datasets for Four Tasks

[[Project Page]](https://data-scaling-laws.github.io/)
[[Paper]](https://huggingface.co/papers/2410.18647)
[[Code]](https://github.com/Fanqi-Lin/Data-Scaling-Laws)
[[Models]](https://huggingface.co/Fanqi-Lin/Task-Models/)
[[Raw GoPro Videos]](https://huggingface.co/datasets/Fanqi-Lin/GoPro-Raw-Videos)

This repository contains in-the-wild robotic manipulation datasets collected using [UMI](https://umi-gripper.github.io/), and processed through a SLAM pipeline, as described in the paper "Data Scaling Laws in Imitation Learning for Robotic Manipulation". The datasets cover four tasks:
+ Pour Water
+ Arrange Mouse
+ Fold Towel
+ Unplug Charger

## Dataset Folders:
**arrange_mouse** and **pour_water**: Each folder contains data from 32 unique environment-object pairs, with 120 demonstrations per pair.

**fold_towel** and **unplug_charger**: Each folder contains data from 32 unique environment-object pairs, with 60 demonstrations per pair.

**pour_water_16_env_4_object** and **arrange_mouse_16_env_4_object**: These folders contain data from 16 environments, with 4 different manipulation objects per environment, and 120 demonstrations per object.

Note that due to the size of the pour_water_16_env_4_object/dataset.zarr.zip file (over 50GB), it has been split into two parts. You can restore the full dataset using the following command:

```shell
cat pour_water_16_env_4_object/dataset_part_* > pour_water_16_env_4_object/dataset.zarr.zip
```

## Additional Information
+ Each dataset is a merge of smaller datasets (one per environment-object pair). Inside each folder, you will find a **count.txt** file that lists the number of demonstrations in each smaller dataset.
+ These datasets can be used to train policies that generalize effectively to novel environments and objects.
+ For more details on how to use our datasets, please refer to our [code](https://github.com/Fanqi-Lin/Data-Scaling-Laws).