Guided, Stochastic Model-Based GUI Testing of Android Apps

Mobile apps have become ubiquitous and drastically increased in number over the recent years. However, it is challenging to guarantee their quality. First, they are event-centric programs with rich graphical user interfaces (GUIs), and interact with complex environments (e.g., users, devices, and other apps). Second, they are typically developed under the time-to-market pressure, thus may be inadequately tested before releases. When performing testing, developers tend to exercise those functionalities or usage scenarios that they believe to be important, but may miss bugs that their designed tests fail to expose.

Stoat's Work Flow

Stoat (STOchastic model App Tester) is a novel guided approach to perform stochastic model-based testing on Android Apps. The idea is to thoroughly test the functionalities of an app from its GUI model, and validate the app’s behavior by enforcing various user/system interactions.

Stoat operates in a unique two-phase process to test an app: (1) construct a stochastic app model (in the form of stochastic finite state machine); and (2) iteratively mutate/refine the stochastic model (by perturbing the probability values of the model transitions), and guide test generation toward achieving high code, model coverage as well as exhibiting diverse event sequences.


1. Model Construction: Stoat uses a dynamic analysis technique, enhanced by a weighted UI exploration strategy and static analysis, to effectively and efficiently construct app models.

2. Guided model-based testing for mutliple apps running on multiple devices at the same time.

**You can also view these demos on YouKu, model construction, and guided model-based testing. But we reconmmend viewing them on Youtube, which is more clear.

Implementation Details

Stoat's supported System level events.

Downloading Stoat

Download Stoat tool and user manual here.


Stoat has been compared with the state-of-the-art.

 Tool Name  Approach Description
 MobiGUITAR  model-based GUI exploration  MobiGUITAR is the extension of AndroidRipper, which implements a systematic exploration and a random exploration strategies when constructing models.
 PUMA  model-based GUI exploration  PUMA uses a generic UI automator to sequentially explore GUIs, and stops exploring when all app states have been visited.
 Tool Name  Approach   Description
 Monkey  random fuzzing Monkey emits a stream of random input events, including both UI and system events, to maximize code coverage.
 A3E  systematic UI exploration A3E systematically explores app pages and emits UI events by following a depth-first strategy.
 Sapienz  search-based testing Sapienz uses Monkey to generate the initial test population, and exploits genetic algorithms to optimize the tests to maximize code coverage while minimizing test lengths.

In Study 1 and 2, Stoat is evaluated on 93 open-source Android apps from F-droid (a popular open-source Android app repository): 68 benchmark apps (widely used in previous research work) and 25 randomly selected apps (to avoid potential evaluation bias).


1. Statistics of the models produced by MobiGUITAR (the systematic strategy --- "M-S", the random strategy --- "M-R" ), PUMA ("PU"), and Stoat ("St") on the 93 open source apps. Note each tool is allocated with one hour for each app and is run on one emulator.

* Line Coverage, #Model States, #Model Transitions.


2. Comparison with AMOLA
AMOLA is also a model-based testing tool for Android apps, which uses multi-level GUI comparison criteria. But the tool is not publicly available, we compare AMOLA and Stoat on the same benchmark apps that it used in its paper.

*Stoat use the similar criterion with AMOLA (C-Lv4)

 App Name     AMOLA
org.jtb.alogcat  15  247  10  54  56%  63%
com.example.anycut  8 33  7  40  55%  67%
com.evancharlton.mileage 69 532  30  262  33%  39%
cri.sanity  2  7  16  202  N/A  12%
org.jessies.dalvikexplorer  30  301  29  348  64%  74%
i4nc4mp.myLock  5 51 5  51  11%  28%
com.bwx.bequick  60 250 8  132  39% 40% 20 96 13 76  17%  21%
net.mandaria.tippytipper  13  102  17  117  61%  81%
de.freewarepoint.whohasmystuff  24  143 18 103  51%  82%

3. Example app models from Stoat

1. Results of the code coverage achieved by A3E (denoted by "A"), Monkey ("M"), Sapienz ("Sa"), and Stoat ("St") grouped by app size on the 93 open source app. Note each tool is allocated with three hours for each app and run on one emulator.



2. Testing statistics of A3E, Monkey, Sapienz, and Stoat in fault detection on the 93 open source app. The configuration is same as the above.

Tool  #Buggy Apps #Unique Crashes
A3E  8  8
Monkey  40  76
Sapienz  43  87
Stoat  68  249


Find more detailed information in our publication (* denotes corresponding author)

Lingling Fan, Ting Su*, Sen Chen, Guozhu Meng, Yang Liu, Lihua Xu, Geguang Pu

The 33rd IEEE/ACM International Conference on Automated Software Engineering (ASE'18), Montpellier, France.


Lingling Fan, Ting Su*, Sen Chen, Guozhu Meng, Yang Liu, Lihua Xu, Geguang Pu, Zhendong Su

The 40th International Conference on Software Engineering (ICSE'18), Gothenburg, Sweden.


Chunyang Chen, Ting Su*, Guozhu Meng, Zhenchang Xing, Yang Liu

The 40th International Conference on Software Engineering (ICSE'18), Gothenburg, Sweden.


Ting Su, Guozhu Meng, Yuting Chen, Ke Wu, Weiming Yang, Yao Yao, Geguang Pu, Yang Liu, Zhendong Su

European Software Engineering Conference and ACM SIGSOFT Symposium on the Foundations of Software Engineering (ESEC/FSE), September 2017, Paderborn, Germany


Ting Su

International Conference on Software Engineering Companion (ICSE '16), May 2016, Austin, Texas, USA (ACM Student Research Competition First Place)



Contact: Ting Su