Nurisk: VQA for Risk Assessment in Autonomous Driving

Nurisk is a visual question answering dataset focusing on risk assessment for autonomous driving. Each row contains:

• image: a BEV image
• question: a driving-related question
• answer: the ground truth answer

Paper

NuRisk: A Visual Question Answering Dataset for Agent-Level Risk Assessment in Autonomous Driving — see the paper on arXiv:2509.25944 .

Framework

Dataset Structure

• Splits: train, validation
• Columns: image, question, answer

Usage

from datasets import load_dataset

ds = load_dataset("Yuan-avs/Nurisk")
sample = ds["train"][0]
print(sample["question"])  # text
print(sample["answer"])    # text
img = sample["image"]       # PIL.Image.Image
img.show()

Notes

• Images may be referenced multiple times across different questions.
• The dataset viewer only exposes image, question, and answer.

License

Please specify the license applicable to the images and annotations.

Citation

If you use this dataset, please cite the authors accordingly.

@article{gao2025nurisk, title={NuRisk: A Visual Question Answering Dataset for Agent-Level Risk Assessment in Autonomous Driving}, author={Gao, Yuan and Piccinini, Mattia and Brusnicki, Roberto and Zhang, Yuchen and Betz, Johannes}, journal={arXiv preprint arXiv:2509.25944} , year={2025} }

If you are interested in foundation model-based scenario generation and scenario analysis, you may also refer to our comprehensive survey, which provides the first overview of this emerging research area.

Publication details: 📅 June 2025 – Released on arXiv. 📊 The accompanying repository categorizes 342 papers, including: 93 on scenario generation 54 on scenario analysis 55 on datasets 21 on simulators 25 on benchmark challenges 94 on other related topics

📄 Paper: https://arxiv.org/abs/2506.11526

💻 GitHub: https://github.com/TUM-AVS/FM-AD-Survey