SyncWorld / README.md
yyuncong's picture
Link every author to their homepage, matching the project page
8566e96 verified
|
Raw
History Blame Contribute Delete
2.38 kB
metadata
license: other
license_name: openmdw-1.1
license_link: https://github.com/NVIDIA/Cosmos/blob/main/LICENSE
pipeline_tag: image-to-video
tags:
  - robotics
  - world-model
  - forward-dynamics
  - image-to-video

SyncWorld: Visual Calibration Enables World Models as Zero-Shot Simulators

Yuncong Yang, Zhengtao Han, Furkan Özyurt, Zeyuan Yang, Han Yang, Junyi Cao, Haoyu Zhen, Yilun Du, Chuang Gan

Paper PDF Project Page Code

SyncWorld is an action-conditioned world model that serves as a zero-shot simulator across unseen environments without any additional training. It leverages a visual calibration episode — paired frames and actions that showcase all the controllable degrees of freedom — to specify the setup-specific Action–Visual Mapping in context, and learns to interpret actions through visual evidence and to fall back on interaction history when explicit calibration is unavailable.

hf download yyuncong/SyncWorld --local-dir ./SyncWorld-ckpt

See the SyncWorld repository for installation and how to run evaluation, and yyuncong/SyncWorld-Evaluation for the evaluation sets.