Learning How the World Evolves: Extrapolative Video World Models via Latent Dynamics Reasoning

Please visit our Github Repo for how to use these models.

Structure

LDR/
├── 128x128/
│   ├── single_task/{uniform,parabola,collision,looming,bouncing}.pt
│   └── joint_task/joint_5task.pt
└── 256x256/
    ├── single_task/{uniform,parabola,collision,looming,bouncing}.pt
    └── joint_task/joint_5task.pt

BibTeX

@article{li2026learning,
  title={Learning How the World Evolves: Extrapolative Video World Models via Latent Dynamics Reasoning},
  author={Li, Haodong and Liu, Shaoteng and Wang, Tianyu and Ge, Chongjian and Ji, Sihui and Zhang, Jiahan and Lin, Xin and Lu, Haolin and Lin, Zhe and Chandraker, Manmohan},
  journal={arXiv preprint arXiv:2608.09926},
  year={2026}
}