MobileDreamer: Generative Sketch World Model for GUI Agent

arXiv 26.1 2026 59.4 method, application

TLDR

MobileDreamer proposes a generative sketch world model for mobile GUI agents to forecast post-action states and improve long-horizon task performance.

Reasoning

Strengths include a novel textual sketch world model with order-invariant learning for spatial preservation and a rollout imagination strategy for action selection. Weaknesses are the domain specificity to GUI agents and lack of detailed efficiency analysis.

Read-first score

Read-first score 59.4, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 48.

Recency 8%

100

Uses a gentle age decay so recent papers surface without erasing older foundations. 2026

Topical relevance 42%

68.6

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

Methodology quality 25%

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

Reproducibility 25%

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=none

Field roles

Frontier

Rank sensitivity

Stability: volatile; rank range: 410.

Keyword Scores

world model

generative world model

world dynamics prediction

world simulator

model-based reinforcement learning world model

interactive world model

video world model

Deep Analysis

Innovations

Textual sketch world model that transforms digital images into key task-related sketches
Order-invariant learning strategy to preserve spatial information of GUI elements
Rollout imagination strategy for GUI agent to optimize action selection using world model predictions

Methodology

MobileDreamer proposes a world-model-based lookahead framework consisting of a textual sketch world model and a rollout imagination strategy. The world model learns to convert digital images into task-related sketches while using an order-invariant learning approach to maintain spatial awareness. The rollout imagination leverages the world model's predictions to guide the agent's action selection process.

Key Results

On Android World, MobileDreamer achieves state-of-the-art performance with a 5.25% improvement in task success rate. World model evaluations confirm that the textual sketch modeling accurately forecasts key GUI elements.

Resources

Paper