一句话：提出非侵入式机器人测试系统 RoScript，用视觉测试脚本驱动物理机器人测试触摸屏应用，并从人操作视频自动录制脚本。
现有面向触摸屏应用的侵入式测试自动化技术（如 Appium、Sikuli）难以适用于 GoPro 等封闭或小众系统，而非侵入式方法能显著扩大测试自动化的适用范围。本文提出 RoScript——一个真正非侵入式的、由测试脚本驱动的机器人测试系统，用于触摸屏应用的测试自动化。RoScript 采用视觉测试脚本描述触摸屏应用上的 GUI 动作，并借助物理机器人驱动自动化测试执行；为降低测试脚本编写成本，系统还引入基于计算机视觉的非侵入式技术，从被测设备上人类操作视频中自动将屏幕点击动作录制为测试代码。由于不依赖底层操作系统或 GUI 框架，RoScript 几乎适用于任意平台上的触摸屏应用。作者在 6 种不同设备的 21 个触摸屏应用上开展实验，成功自动化了包含 650 多个不同 GUI 动作的 104 个测试场景；RoScript 在超过 90% 的测试脚本执行中准确完成 GUI 动作，并将约 85% 的人类屏幕点击动作准确录制为测试代码。
Existing intrusive test automation techniques for touch screen applications (e.g., Appium and Sikuli) are difficult to work on many closed or uncommon systems, such as a GoPro. Being non-intrusive can largely extend the application scope of the test automation techniques. To this end, this paper presents RoScript, a truly non-intrusive test-script-driven robotic testing system for test automation of touch screen applications. RoScript leverages visual test scripts to express GUI actions on a touch screen application and uses a physical robot to drive automated test execution. To reduce the test script creation cost, a non-intrusive computer vision based technique is also introduced in RoScript to automatically record touch screen actions into test scripts from videos of human actions on the device under test. RoScript is applicable to touch screen applications running on almost arbitrary platforms, whatever the underlying operating systems or GUI frameworks are. We conducted experiments applying it to automate the testing of 21 touch screen applications on 6 different devices. The results show that RoScript is highly usable. In the experiments, it successfully automated 104 test scenarios containing over 650 different GUI actions on the subject applications. RoScript accurately performed GUI actions on over 90% of the test script executions and accurately recorded about 85% of human screen click actions into test code.
DOI 原文 ·
@inproceedings{paperbot5613,
title = {RoScript},
author = {Ju Qian and Zhengyu Shang and Shuoyan Yan and Yan Wang and Lin Chen},
booktitle = {ICSE},
year = {2020},
doi = {10.1145/3377811.3380431}
}