STORM: Search-Guided Generative World Models for Robotic Manipulation

TL;DR

STORM introduces a unified framework combining diffusion-based action generation, visual world modeling, and search planning to improve robotic manipulation, achieving state-of-the-art success rates and robust re-planning capabilities.

Contribution

It is the first to integrate diffusion-based policies with explicit visual world models and search-based planning for robotic manipulation.

Findings

Achieves 51.0% success rate on SimplerEnv benchmark.

Reduces Frechet Video Distance by over 75%.

Demonstrates robust re-planning and failure recovery.

Abstract

We present STORM (Search-Guided Generative World Models), a novel framework for spatio-temporal reasoning in robotic manipulation that unifies diffusion-based action generation, conditional video prediction, and search-based planning. Unlike prior Vision-Language-Action (VLA) models that rely on abstract latent dynamics or delegate reasoning to language components, STORM grounds planning in explicit visual rollouts, enabling interpretable and foresight-driven decision-making. A diffusion-based VLA policy proposes diverse candidate actions, a generative video world model simulates their visual and reward outcomes, and Monte Carlo Tree Search (MCTS) selectively refines plans through lookahead evaluation. Experiments on the SimplerEnv manipulation benchmark demonstrate that STORM achieves a new state-of-the-art average success rate of 51.0 percent, outperforming strong baselines such as CogACT. Reward-augmented video prediction substantially improves spatio-temporal fidelity and task relevance, reducing Frechet Video Distance by over 75 percent. Moreover, STORM exhibits robust re-planning and failure recovery behavior, highlighting the advantages of search-guided generative world models for long-horizon robotic manipulation.