STORM: Search-Guided Generative World Models for Robotic Manipulation
Wenjun Lin, Jensen Zhang, Kaitong Cai, Keze Wang · 2025
Read STORM: Search-Guided Generative World Models for Robotic Manipulation by Wenjun Lin, Jensen Zhang, Kaitong Cai, Keze Wang (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2512-18477.txt.
Description
Wenjun Lin and three co-authors present STORM, a framework that helps robots plan manipulations by imagining what their actions will look like. Earlier vision-language-action models reason in hidden, abstract codes. STORM instead grounds planning in explicit video rollouts, and the work was posted to arXiv in December 2025.
Three parts work together. A diffusion-based policy proposes varied candidate actions. A generative video model predicts each action's visual outcome and reward. Monte Carlo tree search then refines the plan by looking ahead. On the SimplerEnv benchmark STORM reaches a 51.0 percent average success rate, a new best, beating systems such as CogACT. Its reward-aware video prediction cuts the Frechet Video Distance by over 75 percent, and the system replans well after failures.
Catalog record
| CREATOR | Wenjun Lin, Jensen Zhang, Kaitong Cai, Keze Wang |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 31 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2512-18477