---
license: apache-2.0
task_categories:
- video-text-to-text
- robotics
language:
- en
tags:
- agent
size_categories:
- 1K<n<10K
---

# AerialVLN

This repository provides **a mirrored copy** of the AerialVLN dataset for research convenience and reproducibility.

The original repository can be accessed from [the official Kaggle repository](https://www.kaggle.com/datasets/shuboliu/aerialvln).

The AerialVLN dataset are associated with the paper [AerialVLN: Vision-and-Language Navigation for UAVs](https://openaccess.thecvf.com/content/ICCV2023/html/Liu_AerialVLN_Vision-and-Language_Navigation_for_UAVs_ICCV_2023_paper.html) (ICCV 2023).

All credits for the original dataset belong to the authors of AerialVLN. This repository is intended to facilitate easier access and usage in related research.

## Verified AerialVLN Datasets

We further provide [AerialVLN-Verified](https://huggingface.co/datasets/LeonMiao/AerialVLN-Verified) and [AerialVLN-S-Verified](https://huggingface.co/datasets/LeonMiao/AerialVLN-S-Verified), two verified versions of [AerialVLN](https://huggingface.co/datasets/LeonMiao/AerialVLN) and [AerialVLN-S](https://huggingface.co/datasets/LeonMiao/AerialVLN-S). These datasets were curated with the assistance of [ViRC-7B](https://huggingface.co/LeonMiao/ViRC-7B) and approximately 150 hours of human verification.

The verification process corrects noisy instruction texts and abnormal action sequences, including corrupted characters, collision actions, semantically inconsistent actions, and trajectories affected by takeoff failures. Collision-related actions were identified using Unreal Engine's Point-in-Frustum Test with Occlusion Query, while candidate corrected action sequences were sampled by [ViRC-7B](https://huggingface.co/LeonMiao/ViRC-7B) and manually verified. To preserve compatibility with the original navigation setting, only intermediate action subsequences were replaced, while the starting and ending actions and UAV poses were kept consistent with the original trajectories.

The verified datasets provide cleaner instruction-action pairs for aerial vision-and-language navigation research, supporting studies on robust instruction following, action-sequence learning, and trajectory-level evaluation.

If you find the verified datasets or [ViRC-7B](https://huggingface.co/LeonMiao/ViRC-7B) useful for your research, please consider citing:

```bibtex
@article{wang2025virc,
  title={ViRC: Enhancing Visual Interleaved Mathematical CoT with Reason Chunking}, 
  author={Lihong, Wang and Liangqi, Li and Weiwei, Feng and Jiamin, Wu and Changtao, Miao and Tieru, Wu and Rui, Ma and Bo, Zhang and Zhe, Li},
  journal={arXiv preprint arXiv:2512.14654},
  year={2025}
}
@inproceedings{liu2023aerialvln,
  title={AerialVLN: Vision-and-language Navigation for UAVs},
  author={Shubo Liu and Hongsheng Zhang and Yuankai Qi and Peng Wang and Yanning Zhang and Qi Wu},
  booktitle={International Conference on Computer Vision (ICCV)},
  year={2023}
}
```