AUITestAgent is an automated GUI testing tool for mobile applications that translates natural language requirements into executable test cases. It is designed for researchers and developers seeking to automate GUI functional testing and bug detection in mobile apps, offering a novel approach to converting natural language specifications into interactive UI test sequences.

How It Works

AUITestAgent employs a decoupled architecture, separating GUI interaction generation from function verification. It uses dynamically organized agents to parse natural language requirements, extracting relevant GUI interactions. A multi-dimensional data extraction strategy then analyzes interaction traces to verify that the app's behavior aligns with the specified requirements, aiming for improved performance in domain-specific UI testing.

Quick Start & Requirements

• Installation and usage instructions are not explicitly detailed in the README.
• Requires mobile applications for testing and likely a Python environment.
• Links to demos are provided:
  • Meituan Task: https://github.com/user-attachments/assets/48341d06-bc05-4b71-accd-c8a1c7215834
  • Facebook Task: https://github.com/user-attachments/assets/8c0a33ab-11ab-4f95-b767-678472e8d902

Highlighted Details

• Achieves 100% task completion for easy interactions, 80% for moderate, and 50% for difficult tasks across 8 commercial apps.
• Demonstrates 94% alignment of generated interactions with ground truth manual interactions.
• Reports 90% recall for injected GUI functional bugs with a 4.5% false positive rate.
• Successfully detected unseen bugs in Meituan, indicating practical utility.

Maintenance & Community

• Developed by a joint team from Fudan University and Meituan.
• Related projects include vision-ui, Appaction, and AutoConsis.
• Citation details are provided for the paper "AUITestAgent: Automatic Requirements Oriented GUI Function Testing".

Licensing & Compatibility

• The README does not specify a license.

Limitations & Caveats

• Detailed installation and setup instructions are not provided, requiring further investigation.
• Performance metrics are based on specific benchmarks and may vary with different applications or task complexities.