MobileWorldBench: Towards Semantic World Modeling For Mobile Agents
This work addresses the problem of practical world modeling for mobile agents in GUI settings, offering a domain-specific solution that is incremental in its formulation.
The paper tackles the challenge of world modeling for mobile GUI agents by proposing a semantic approach using natural language state transitions instead of pixel-space predictions, resulting in improved task success rates through a novel framework and a large-scale dataset.
World models have shown great utility in improving the task performance of embodied agents. While prior work largely focuses on pixel-space world models, these approaches face practical limitations in GUI settings, where predicting complex visual elements in future states is often difficult. In this work, we explore an alternative formulation of world modeling for GUI agents, where state transitions are described in natural language rather than predicting raw pixels. First, we introduce MobileWorldBench, a benchmark that evaluates the ability of vision-language models (VLMs) to function as world models for mobile GUI agents. Second, we release MobileWorld, a large-scale dataset consisting of 1.4M samples, that significantly improves the world modeling capabilities of VLMs. Finally, we propose a novel framework that integrates VLM world models into the planning framework of mobile agents, demonstrating that semantic world models can directly benefit mobile agents by improving task success rates. The code and dataset is available at https://github.com/jacklishufan/MobileWorld