Everloom/AAWR-DROID

---
license: mit
task_categories:
- robotics
language:
- en
tags:
- active-perception
- robotics
- RL
size_categories:
- 10B<n<100B
---

# Active Perception AAWR Dataset in GRASP Lab Mock Kitchen

[Paper](https://huggingface.co/papers/2512.01188) | [Project page](https://penn-pal-lab.github.io/aawr/) | [Code](https://github.com/penn-pal-lab/aawr)

Here is our full dataset (13.5GB) in DROID format:
https://huggingface.co/datasets/Everloom/AAWR-DROID/

It includes 4 scene as below, each are cleaned with a white list (filter the idle frames)


## Sample Usage

To get started with AAWR, follow these steps for a toy simulated task, as described in the [code repository](https://github.com/penn-pal-lab/aawr):

### Setup

1. **Install Python dependencies**: You can refer to the `environment.yaml` file, or just manually install the dependencies. There's not too many, mainly `torch`, `gymnasium` and their related packages.

2. **Install the simulated xarm environment code, and generate demos**:
```bash
[e-mail]:edwhu/gym-xarm.git
pip install -e .
python generate_demos.py # generate demo data.

mkdir gymnasium_xarm_lift_bc
mv buffer.pkl gymnasium_xarm_lift_bc
```

### Train AAWR

Train AAWR on a simple simulated lifting task:

```bash
python src/train_il.py agent=asym_awr modality=all task=gymnasium_xarm_lift dataset_dir=data/gymnasium_xarm_lift_bc seed=41 use_wandb=false exp_name=aawr_xarm_lift horizon=1 offline_steps=20000 train_steps=40000 lr=1e-4 grad_clip_norm=10 awr_filter=indicator expectile=0.9 A_scaling=3 save_video=true switch_awr_filter=true
```

## Citation

If you find our project helpful, please cite us:

```bibtex
@inproceedings{
hu2025realworld,
title={Real-World Reinforcement Learning of Active Perception Behaviors},
author={Edward S. Hu and Jie Wang and Xingfang Yuan and Fiona Luo and Muyao Li and Gaspard Lambrechts and Oleh Rybkin and Dinesh Jayaraman},
booktitle={The Thirty-ninth Annual Conference on Neural Information Processing Systems},
year={2025},
url={https://openreview.net/forum?id=RkdTtznSAL}
}
```

## Additional Resources

Here is a website on searching behavior under 3rd camera view:
https://sites.google.com/view/rwrl-ap/home