Spatial-Temporal World Understanding

Benchmarks and evaluation for precise world understanding in multimodal embodied settings.

This project direction focuses on evaluating whether modern multimodal models can reason about the world with enough spatial and temporal precision for embodied tasks.

Current public representative paper:

(Li et al., 2025)

The broader goal is to expose the gap between generic multimodal understanding and the kind of grounded reasoning required by agents acting in real environments.

References

2025

Sti-bench: Are mllms ready for precise spatial-temporal world understanding?

Y. Li, Y. Zhang, T Lin, and 4 more authors

ICCV, 2025

Bib Paper

@inproceedings{li2025stibench,
  title = {Sti-bench: Are mllms ready for precise spatial-temporal world understanding?},
  author = {Li, Y. and Zhang, Y. and Lin, Tao and Liu, X. R. and Cai, W. and Liu, Z. and Zhao, B.},
  booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision},
  year = {2025},
}