EpiDroid: Dependency-Guided Recomposition for Deep State Discovery in Mobile GUI Testing

A research team from Zhejiang University and Alibaba Group has introduced EpiDroid, a novel framework that significantly improves automated testing for complex mobile applications. The system addresses a critical limitation in current GUI testing tools: their inability to understand semantic dependencies between different application states. EpiDroid works as a black-box, pluggable layer that can augment existing testing explorers like those used in industrial settings.

EpiDroid's core innovation is its Dependency-Guided Recomposition approach. First, it distills raw testing traces into stable fragments to extract underlying state relationships. Then, using LLMs for impact reasoning, it performs deterministic replay focused on high-value mutable state elements. Through iterative feedback, EpiDroid refines a state-dependency graph to systematically reach deep application states that forward exploration alone cannot access.

The framework was evaluated on 20 real-world Android applications and integrated with both industrial tools and state-of-the-art research explorers. Results demonstrated consistent performance improvements across all baselines, with average code coverage increases of 10-28%. Most impressively, EpiDroid delivered 3-4 times more coverage gain compared to simply continuing the baseline tools from the same starting point, proving that dependency awareness unlocks deep states regardless of additional testing budget.