World4Omni: A Zero-Shot Framework from Image Generation World Model to Robotic Manipulation

arXiv 25.6 2025 40.5 method, application

TLDR

A zero-shot framework using image-generative VLMs as world models to generate goal states for generalizable robotic manipulation.

Reasoning

The paper presents a novel approach leveraging image-generative VLMs as world models for zero-shot robotic manipulation, with real-world experiments demonstrating strong performance. However, the abstract lacks details on limitations and the scope is narrow, focusing only on goal state generation.

Read-first score

Read-first score 40.5, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 19.

Recency 8%

86.7

Uses a gentle age decay so recent papers surface without erasing older foundations. 2025

Methodology quality 25%

Screens visible abstract and analysis fields for experiment, dataset, baseline, metric, and limitation evidence. markers=experiment

Reproducibility 25%

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=github

Topical relevance 42%

27.1

Uses existing LLM keyword relevance scores normalized to 0-100. world model,world simulator,generative world model,interactive world model,video world model,world dynamics prediction,model-based reinforcement learning world model

Field roles

Frontier

Rank sensitivity

Stability: volatile; rank range: 280.

Keyword Scores

generative world model

world model

world simulator

world dynamics prediction

interactive world model

video world model

model-based reinforcement learning world model

Deep Analysis

Innovations

Leveraging Image-Generative VLMs as world models to generate desired goal states for robotic manipulation
Using object state representation as a golden interface to separate high-level and low-level policies, enabling training-free low-level control
Introducing a Reflection-through-Synthesis process that iteratively validates and refines the generated goal image before execution

Methodology

Goal-VLA is a zero-shot framework that uses Image-Generative VLMs as world models to generate desired goal states. From these generated images, the target object pose is derived, which serves as spatial cues for training-free low-level control. The system is evaluated in both simulated and real-world environments.

Key Results

The framework achieves strong performance and inspiring generalizability in manipulation tasks across simulated and real-world experiments.

Resources

Paper