New AI System Generates Mobile App Test Cases by Extracting Business Logic

Researchers have introduced LogiDroid, an automated approach to generating functional test cases for mobile applications by parsing business logic from requirements and adapting it to real-world app interfaces.

The Challenge#

Functional testing—verifying that an app's behavior matches what users expect—remains largely manual and labor-intensive. Two bottlenecks complicate automation: extracting usable business logic from unstructured requirements documents is difficult, and translating that logic into test cases for different mobile interfaces requires bridging a substantial semantic gap between the specification and the GUI environment.

How LogiDroid Works#

The system operates in two stages. Knowledge Retrieval and Fusion uses LLM-based agents to build a test dataset, locate relevant existing test cases, and convert unstructured requirements into structured business logic. Context-Aware Test Generation then deploys two additional agents that jointly examine the extracted logic and the live GUI to incrementally produce test cases tailored to the target app—including verification assertions that check whether functionality behaves correctly.

Experimental Results#

LogiDroid was tested on 28 real-world applications and 190 functional requirements across two benchmark datasets:

• On the FrUITeR dataset, LogiDroid successfully tested 40% of functional requirements—a gain of 25% over prior state-of-the-art methods
• On the Lin dataset, it achieved 65% success—an improvement of 55% relative to existing approaches

These gains suggest that extracting and contextualizing business logic substantially reduces the manual effort needed to validate mobile app behavior.

Publication#

The research was submitted to arXiv on 27 February 2026, with a revised version released on 14 July 2026.