DQT

Overview

DQT (Deep Q-network Testing) is an automatic black-box Android GUI testing approach based on deep reinforcement learning and graph embedding. It leverages advanced graph embedding techniques for states and action encoding, employs a specially designed DQN for state-action evaluation, and conducts curiosity-driven exploration guided by a fine-grained dynamic reward function.

The artifact of DQT can be downloaded from GoogleDriveShare.

Publication

More details about DQT can be achieved in the ICSE 2024 paper "Deeply Reinforcing Android GUI Testing with Deep Reinforcement Learning".

@inproceedings{DBLP:conf/icse/LanLLP00L24,
  author       = {Yuanhong Lan and
                  Yifei Lu and
                  Zhong Li and
                  Minxue Pan and
                  Wenhua Yang and
                  Tian Zhang and
                  Xuandong Li},
  title        = {Deeply Reinforcing Android {GUI} Testing with Deep Reinforcement Learning},
  booktitle    = {Proceedings of the 46th {IEEE/ACM} International Conference on Software
                  Engineering, {ICSE} 2024, Lisbon, Portugal, April 14-20, 2024},
  pages        = {71:1--71:13},
  publisher    = {{ACM}},
  year         = {2024},
  url          = {https://doi.org/10.1145/3597503.3623344},
  doi          = {10.1145/3597503.3623344},
  timestamp    = {Mon, 24 Jun 2024 15:20:25 +0200},
  biburl       = {https://dblp.org/rec/conf/icse/LanLLP00L24.bib},
  bibsource    = {dblp computer science bibliography, https://dblp.org}
}

File Structure

DQT
├── app  The directory for your APKs, an example MyExpenses-r554-debug.apk is provided.
├── main  The directory for the executable DQT.
├── result  The directory for storing runtime testing results.
└── venv  The Python virtual environment.

For directories app and result, specifying other paths for them in the config file is also feasible.
For directory venv, you need to build your own Python virtual environment.

Experimental Environment

Here is the experimental environment we have tested.

Operating System

• Ubuntu 18.04
• Also tested on Ubuntu 22.04

Python

• The project is based on Python 3.7.
  • except for module system_event_trigger which only supports Python 2.7
• Make sure that Python 3.7 and Python 2.7 are all available in your Ubuntu
  • It is recommended to install Python 3.7 and Python 2.7 via apt command (e.g., use ppa:deadsnakes/ppa source), thus /usr/bin/python3.7 and /usr/bin/python2.7 will be directly available.
• Build your own Python3.7 virtual environment for the project.
  • It is recommended to build it under DQT/venv.
  • You may use command python3.7 -m venv venv.
• You may use the provided Python2.7 for module system_event_trigger directly.
  • While testing the tool on Ubuntu 22.04, this provided virtual environment works well.
  • You may build your own Python2.7 virtual environment to replace the provided one.
  • Anyway, make sure DQT/main/system_event_trigger/venv/bin/python2.7 performs normally. You can check this by directly running it to enter the Python2.7 interactive mode.
• For these two venvs, in general terms, there is no need to install any extra packages, as the executable DQT already includes necessary python packages.

Deep Learning

• GPU Driver: 470.161.03 (to support CUDA)
• CUDA: 11.4 / 11.8
  • Remember to add it to the system PATH
  • Check by command: nvcc -V
• cuDNN: 8.2.4 (compatible with CUDA)
• Torch: 1.10.0+cu113 (has already been provided with the executable DQT)

Note: Pay attention to the version compatibility between GPU Driver, CUDA, and cuDNN.

Android Environment

• Command Line Tool Used: ADB, AAPT, AAPT2
  • Make sure these tools are available on system variables to support command-line calls
  • It is recommended to add Android Supports such as tools, platform-tools, and build-tools to the system path.
• You may check the availability of these tools by these commands: adb version, aapt version, aapt2 version

Note: One convenient way to download these Android tools is via AndroidStudio.

Android Emulator

• Device: Google Pixel 2
• Resolution: 1080*1920
• Android Version: Android 9.0 (API Level 28)
• RAM: 4GB
• VM Heap: 2GB
• Internal Storage: 8GB
• SD Card: 1GB

Testing Config (config.yaml in DQT/main/)

• ALLOW_CLEAN: Allow the tool to clean app data during testing
  • Except for some special cases, this config is usually set to true.
  • Special cases need this config to be false, possible scenarios include:
    • Logged-in apps (apps that have been logged in and you don't want to lose this status while testing)
    • Database conflicts (avoid duplicate database builds)
    • Plug-in missing (apps that have been equipped with plug-ins and you don't want to lose them while testing)
  • Among our 30 experimental apps, these 5 apps are recommended to set false: Signal (login), Tachiyomi (prevent plug-in loss), K9Mail (login), Conversations (login), MoneyManagerEx (database problem)
  • Note that even if this is set to false, DQT will still clean app data at the very final phase of the test.
  • It is recommended to take a snapshot of the emulator to save the initial status before testing and start with the snapshot for every test
• APK_LOCATION: The absolute dir path of the APK.
• APK_NAME: The file name of the App Under Test (AUT). We have provided the app MyExpenses as an example.
• DEVICE_ID: The device ID for testing (e.g., emulator-5554).
• RESULT_LOCATION: The absolute dir path for saving runtime testing results.
• TIME_LIMIT: The testing time in seconds.

Run DQT

1. Unzip DQT.zip and enter the directory DQT
2. Build the two Python virtual environments
3. Prepare an Android Device (e.g., an Android Emulator), use the command adb devices to check
4. Enter the Python3 virtual environment (e.g., source ./venv/bin/activate)
5. Edit the config file DQT/main/config.yaml
6. Run DQT via a command like ./main/main

Reference Links

1. Python
   1. Python venv: https://docs.python.org/3.7/tutorial/venv.html
2. Deep Learning
   1. GPU Driver and CUDA: https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html
   2. CUDA and cuDNN: https://developer.nvidia.com/rdp/cudnn-archive
3. Android
   1. Logcat: https://developer.android.com/studio/command-line/logcat
   2. Emulator: https://developer.android.com/studio/run/emulator
   3. Emulator(cmd): https://developer.android.com/studio/run/emulator-commandline
   4. ADB: https://developer.android.com/studio/command-line/adb
   5. AAPT2: https://developer.android.com/studio/command-line/aapt2