---
task_categories:
- video-text-to-text
license: mit
dataset_info:
  features:
  - name: question
    dtype: string
  - name: index2ans
    struct:
    - name: A
      dtype: string
    - name: B
      dtype: string
    - name: C
      dtype: string
    - name: D
      dtype: string
  - name: answer
    dtype: string
  - name: video_path
    dtype: string
  - name: category
    dtype: string
  - name: subcategory
    dtype: string
  splits:
  - name: test
    num_bytes: 400189
    num_examples: 1214
  download_size: 105256
  dataset_size: 400189
configs:
- config_name: default
  data_files:
  - split: test
    path: data/test-*
---

# Physical AI Bench - Understanding

PAI-Bench (Physical AI Bench) is a comprehensive benchmark designed to evaluate physical AI generation and understanding capabilities across various real-world scenarios. This particular dataset, **PAI-Bench-U**, focuses specifically on **Video Understanding** tasks, comprising 2,808 real-world cases with task-aligned metrics.

- **Paper:** [PAI-Bench: A Comprehensive Benchmark For Physical AI](https://huggingface.co/papers/2512.01989)
- **Code:** [GitHub Repository](https://github.com/SHI-Labs/physical-ai-bench)

## Citation

If you use Physical AI Bench in your research, please cite:

```bibtex
@misc{zhou2025paibenchcomprehensivebenchmarkphysical,
      title={PAI-Bench: A Comprehensive Benchmark For Physical AI}, 
      author={Fengzhe Zhou and Jiannan Huang and Jialuo Li and Deva Ramanan and Humphrey Shi},\
      year={2025},\
      eprint={2512.01989},\
      archivePrefix={arXiv},\
      primaryClass={cs.CV},\
      url={https://arxiv.org/abs/2512.01989}, \
}
```