| # π§© VRQA-Bench |
|
|
| [](https://arxiv.org/abs/2606.03603) |
| [](https://github.com/yczhou001/PF-OPSD) |
|
|
| > A multiple-choice visual reasoning benchmark built on π **maze** navigation and π¦ **Sokoban** puzzles. Each sample provides an initial-state image, an optimal-solution video, and a question about the planned trajectory. |
|
|
| ## π·οΈ Categories |
|
|
| | Category | Task | Description | |
| |----------|------|-------------| |
| | `C1_turn_count` | π maze | Total number of direction changes along the path | |
| | `C2_turn_direction` | π maze | Number of LEFT / RIGHT turns | |
| | `C4_sokoban_push` | π¦ sokoban | Direction of a specific push | |
| | `C5_direction_count` | π maze | Number of steps in a given direction | |
| | `C6_push_dir_count` | π¦ sokoban | Number of pushes in a given direction | |
|
|
| ## π Annotation Fields |
|
|
| `question`, `options` (AβD), `answer` (correct key), `correct_value`, `task_type`, `input_image` / `video_path` (relative to each split dir), `_solution` (ground-truth solution trace). |
|
|
| ## π Citation |
|
|
| ```bibtex |
| @misc{zhou2026worldmodelsmeetlanguage, |
| title={World Models Meet Language Models: On the Complementarity of Concrete and Abstract Reasoning}, |
| author={Yucheng Zhou and Wei Tao and Yiwen Guo and Jianbing Shen}, |
| year={2026}, |
| eprint={2606.03603}, |
| archivePrefix={arXiv}, |
| primaryClass={cs.CV}, |
| url={https://arxiv.org/abs/2606.03603}, |
| } |
| ``` |
|
|