Abstract
Physical world modeling requires predicting how interactions change a scene, not merely generating coherent motion. We propose STRIKE, a framework that separates visual state transition learning from dense video generation. We construct event-aligned supervision by extracting observed states from training videos and pairing them with transition descriptions and temporal offsets. An image-based transition model learns to predict the next scene configuration from the current image, a local transition specification, and elapsed time. At inference, a pretrained vision-language planner predicts time transition specifications, and recursive application of the learned transition model produces a sequence of future visual states. A separately trained dynamic model then generates the complete rollout conditioned on these states and their temporal locations. Experiments on Physics-IQ Verified, PhyGenBench, Pisa-Experiments, and RoboTwin2.0 show improvements of STRIKE over the corresponding video-backbone baselines in benchmark measures of physical consistency and manipulation-video fidelity. These results support learned visual state transitions as an effective intermediate representation for physical world modeling.
Keywords
Subject
Publication details
- Journal
- Not available
- Open access
- Green open access
Cite this article
APA 7
Teng, W., Xu, T., Meng, D., Li, Y., Herau, Q., Hu, Y., Zhao, Y., & Zhan, W. (2026). STRIKE: Learning Visual State Transitions for Physical World Modeling. https://omanscience.com/en/articles/strike-learning-visual-state-transitions-for-physical-world-modeling
MLA 9
Teng, Wenbin, et al. "STRIKE: Learning Visual State Transitions for Physical World Modeling." https://omanscience.com/en/articles/strike-learning-visual-state-transitions-for-physical-world-modeling.
Chicago (author–date)
Teng, Wenbin, Tianshuo Xu, Depu Meng, Yuelei Li, Quentin Herau, Yihan Hu, Yajie Zhao, and Wei Zhan. 2026. "STRIKE: Learning Visual State Transitions for Physical World Modeling." https://omanscience.com/en/articles/strike-learning-visual-state-transitions-for-physical-world-modeling.
Harvard
Teng, W., Xu, T., Meng, D., Li, Y., Herau, Q., Hu, Y., Zhao, Y. and Zhan, W. (2026) 'STRIKE: Learning Visual State Transitions for Physical World Modeling', Available at: https://omanscience.com/en/articles/strike-learning-visual-state-transitions-for-physical-world-modeling.
Vancouver
Teng W, Xu T, Meng D, Li Y, Herau Q, Hu Y, et al. STRIKE: Learning Visual State Transitions for Physical World Modeling. https://omanscience.com/en/articles/strike-learning-visual-state-transitions-for-physical-world-modeling
IEEE
W. Teng, T. Xu, D. Meng, Y. Li, Q. Herau, Y. Hu, Y. Zhao, and W. Zhan, "STRIKE: Learning Visual State Transitions for Physical World Modeling," https://omanscience.com/en/articles/strike-learning-visual-state-transitions-for-physical-world-modeling.