EVO-WAM

EVO-WAM

Evolving World Action Models
through Video-Action Verification

Paper GitHub Watch demo
Explore
Shiyang Zhou1,2,‡Xionghao Wu3,‡Wenbo Li4,†Shenghe Zheng5Jiyao Zhang6Songsong Yu7Yijun Yang5,8Jianhui Liu9Haoze Sun10Senqiao Yang11Li Jiang12,2Jingyong Su1Haoyang Huang4Zhuotao Tian1,2,*
1HITSZ2SLAI3THU4JD5HKUST6PKU7SJTU8HKUSTGZ9HKU10UBC11CUHK12CUHKSZ

shiyangzhou@stu.hit.edu.cnfenglinglwb@gmail.comtianzhuotao@hit.edu.cn

‡ Equal contribution† Project lead* Corresponding author

01 / RESULTS

Imagination
improves generalization.

Mean task success (%) · R0–R4

SIMULATION Cosmos3

26.9→68.0%

+41.1 percentage points

Cosmos3 simulation success rates: R0 26.9%, R1 58.3%, R2 66.6%, R3 63.6%, R4 68.0%. A gain of 41.1 percentage points from R0 to R4.

7 unseen RoboTwin 2.0 tasks

SIMULATION DreamZero

28.5→46.4%

+17.9 percentage points

DreamZero simulation success rates: R0 28.5%, R1 36.5%, R2 42.3%, R3 45.1%, R4 46.4%. A gain of 17.9 percentage points from R0 to R4.

7 unseen RoboTwin 2.0 tasks

REAL WORLD Cosmos3

20.0→76.7%

+56.7 percentage points

Cosmos3 real-robot success rates: R0 20.0%, R1 60.0%, R2 76.7%, R3 73.3%, R4 76.7%. A gain of 56.7 percentage points from R0 to R4.

3 unseen long-horizon composite tasks

Results & evaluation ↗

02 / THE STORY

Self-evolution in action.

Media credits

Imagine the task.
Verify the experience.
Improve the policy.

EVO-WAM learns from its own generated video–action trajectories, retaining prefixes that pass task-completion and action-consistency checks.

Self-training uses no additional expert demonstrations or external execution of candidate actions.

Method

03 / IN ACTION

From imagined experience
to better execution.

Place Ducks

BeforeR1
Task incompleteDownload
AfterR2
Task completed Download

Separate trials, similar layouts. Inference pauses removed.

04 / THE METHOD

Generate. Verify.
Improve.

Duck placement

Candidate 1
Candidate 2
Candidate 3
Candidate 4
Full framework EVO-WAM framework: autoregressive video-action-state generation, visual goal evaluation and inverse-dynamics consistency verification, followed by supervised fine-tuning with accepted prefixes and original data.

Paper, Figure 2.

05 / RESOURCES

Explore EVO-WAM.

PaperarXiv2609.38057RepositoryGitHub
BibTeX
@misc{zhou2026evowamevolvingworldaction,
      title={EVO-WAM: Evolving World Action Models through Video-Action Verification},
      author={Shiyang Zhou and Xionghao Wu and Wenbo Li and Shenghe Zheng and Jiyao Zhang and Songsong Yu and Yijun Yang and Jianhui Liu and Haoze Sun and Senqiao Yang and Li Jiang and Jingyong Su and Haoyang Huang and Zhuotao Tian},
      year={2026},
      eprint={2609.38057},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2609.38057},
}
Download BibTeX