一句话：提出基于图像理解与深度Q网络的平台无关GUI自动化测试框架。
现有自动化GUI测试工具大多依赖特定平台框架，开发跨平台工具前需充分理解平台特性。UniRLTest从测试者视觉出发，利用不同平台GUI在控件图像和布局设计上的共性，通过计算机视觉技术捕捉截图中的控件并为每个页面构建控件树，进而生成可执行动作集。该方法采用深度Q网络，将Android GUI测试形式化为马尔可夫决策过程进行探索。作者在来自不同平台的25个应用上开展评估，结果表明UniRLTest在效率和有效性上均优于基线。
GUI testing has been prevailing in software testing. However, existing automated GUI testing tools mostly rely on frameworks of a specific platform. Testers have to fully understand platform features before developing platform-dependent GUI testing tools. Starting from the perspective of tester’s vision, we observe that GUIs on different platforms share commonalities of widget images and layout designs, which can be leveraged to achieve platform-independent testing. We propose UniRLTest, an automated software testing framework, to achieve platform independence testing. UniRLTest utilizes computer vision techniques to capture all the widgets in the screenshot and constructs a widget tree for each page. A set of all the executable actions in each tree will be generated accordingly. UniRLTest adopts a Deep Q-Network, a reinforcement learning (RL) method, to the exploration process and formalize the Android GUI testing problem to a Marcov Decision Process (MDP), where RL could work. We have conducted evaluation experiments on 25 applications from different platforms. The result shows that UniRLTest outperforms baselines in terms of efficiency and effectiveness.
DOI 原文 ·
@inproceedings{paperbot7394,
title = {UniRLTest: universal platform-independent testing with reinforcement learning via image understanding},
author = {Ziqian Zhang and Yulei Liu and Shengcheng Yu and Xin Li and Yexiao Yun and Chunrong Fang and Zhenyu Chen},
booktitle = {ISSTA},
year = {2022},
doi = {10.1145/3533767.3543292}
}