|
--- |
|
license: mit |
|
task_categories: |
|
- robotics |
|
tags: |
|
- code |
|
size_categories: |
|
- 100B<n<1T |
|
--- |
|
# Raw GoPro Videos for Four Robotic Manipulation 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/) |
|
[[Processed Dataset]](https://huggingface.co/datasets/Fanqi-Lin/Processed-Task-Dataset) |
|
|
|
This repository contains raw GoPro videos of robotic manipulation tasks collected in-the-wild using [UMI](https://umi-gripper.github.io/), as described in the paper "Data Scaling Laws in Imitation Learning for Robotic Manipulation". The dataset covers four tasks: |
|
+ Pour Water |
|
+ Arrange Mouse |
|
+ Fold Towel |
|
+ Unplug Charger |
|
|
|
## Dataset Folders: |
|
**arrange_mouse** and **pour_water**: Each folder contains data collected from 32 environments. |
|
+ The first 16 environments have 4 different object folders per environment, each containing 120 GoPro videos. |
|
+ The remaining 16 environments have one object folder per environment, each containing 120 GoPro videos. |
|
|
|
**fold_towel** and **unplug_charger**: Each folder contains data from 32 unique environment-object pairs, with 60 GoPro videos per pair. |
|
|
|
## Usage |
|
The raw GoPro videos can be processed using the provided [code](https://github.com/Fanqi-Lin/Data-Scaling-Laws) to create the [processed dataset](https://huggingface.co/datasets/Fanqi-Lin/Processed-Task-Dataset) for direct use in policy learning. |
