WorldBagel: Uncovering the Power of Unified Multimodal Models for Vision-Language-Action-World Modeling

2026-07-03

AI Summary

This work addresses the limitations of existing world models in cross-modal perception, reasoning, and action coordination. It proposes WorldBagel, a framework built upon the unified multimodal architecture BAGEL, which for the first time systematically investigates the impact of architectural unification on vision–language–action–world (VLAW) modeling. By jointly modeling visual, linguistic, and action sequences in an end-to-end manner, WorldBagel is trained and evaluated on robotic platforms including LIBERO, Language Table, and Franka, yielding action representations with clearer structure and stronger semantic alignment. Experimental results demonstrate that WorldBagel consistently outperforms task-specific models across multiple benchmarks, confirming that a unified architecture not only enhances modeling performance but also exhibits superior generalization capabilities.

Abstract

World models aim to capture environment dynamics in ways that support perception, reasoning, and action, and have recently become a central direction in Vision-Language-Action-World (VLAW) modeling. Meanwhile, unified vision-language models have demonstrated strong multimodal generation capabilities, yet their potential as world models remains underexplored. In this work, we introduce \texttt{WorldBagel}, a unified VLAW framework built on BAGEL, a modern multimodal unified model, and use it to systematically investigate the role of unification in world modeling. Across multi-task robotic manipulation and cross-domain experiments, \texttt{WorldBagel} consistently outperforms task-specific alternatives and learns action representations that are more structured and semantically aligned with visual and linguistic context. Experiments on LIBERO, Language Table, and Franka show that unification is not only an architectural convenience, but also a key factor in learning effective VLAW models, leading to consistent empirical gains and deeper insights into multimodal world modeling. Code and model checkpoints will be released upon acceptance.

Problem

• world models
• unified multimodal models
• vision-language-action-world modeling
• VLAW
• multimodal generation

Innovation

• unified multimodal models
• world modeling
• vision-language-action
• action representation
• robotic manipulation