Datasets:
license: odc-by
dataset_info:
config_name: ObjectNavType
features:
- name: path
dtype: string
- name: shard_id
dtype: int64
- name: offset
dtype: int64
- name: size
dtype: int64
- name: inflated_size
dtype: int64
splits:
- name: pkgs
num_bytes: 4065356
num_examples: 72620
download_size: 2028907
dataset_size: 4065356
configs:
- config_name: ObjectNavType
data_files:
- split: pkgs
path: ObjectNavType/pkgs-*
tags:
- object navigation
- indoor
- vla
- cross-embodiment
RING-data
🌐 Project page 💻 GitHub 📄 Paper
This dataset contains training episode data (actions, visual inputs, and other sensor data) with randomized embodiments for ObjectNavType tasks.
Please note that every package indexed by the parquet file can contain several instances of episode data.
Data access
We recommend to use bulk_download.py. For example,
python bulk_download.py --max_shards 1 --all /path/to/ring_data/
will download and extract one shard from the single available task configuration (ObjectNavType) under /path/to/ring_data.
To launch it you might need to install some dependencies, e.g., by:
pip install zstandard datasets huggingface_hub tqdm
License
This dataset is licensed under ODC-BY 1.0. It is intended for research and educational use in accordance with Ai2's Responsible Use Guidelines. Some scenes in the episode data were generated using a subset of Ai2's Objaverse, consisting of CC-licensed 3D objects from third party sources. The original license and creator attribution for the Objaverse objects used in this dataset are included in the scene metadata. Please refer to the instructions below on how to download this information. The scripts are licensed under Apache 2.0.
Scene license info
Assuming part or all of the data has been downloaded with the bulk_download.py script, e.g. by using the command shown above, we can use license_info.py to access license info for each scene. For example,
python license_info.py --split val --house-index 3 /path/to/ring_data/
prints the available information for scene with index 3 from the val split including licensing and annotation sources (where appropriate) for each included object.
Note that each of the directories under the downloaded train and val subdirectories are named as the formatted index for the corresponding scene.
BibTeX
@article{eftekhar2024onering_arxiv,
title={The One RING: a Robotic Indoor Navigation Generalist},
author={Eftekhar, Ainaz and Weihs, Luca and Hendrix, Rose and Caglar, Ege and Salvador, Jordi and Herrasti, Alvaro and Han, Winson and VanderBilt, Eli and Kembhavi, Aniruddha and Farhadi, Ali and Krishna, Ranjay and Ehsani, Kiana and Zeng, Kuo-Hao},
year={2024},
eprint={2412.14401},
archivePrefix={arXiv},
primaryClass={cs.RO},
url={https://arxiv.org/abs/2412.14401},
}