Resumen

The developers focus on testing applications, which can be a time-consuming task. To address this issue, we developed AgentDroid, a tool that utilizes reinforcement learning techniques to automate test execution. So far, the results have been impressive, outperforming state-of-the-art RL-based automated testing tools for Android, such as ARES. In fact, AgentDroid achieved a 20% improvement in cumulative coverage compared to ARES. However, its effectiveness has only been evaluated on a single application, making it challenging to find compatible apps for testing. To address this, we tested 61 open-source apps and successfully executed 11 to verify that the tool's performance was consistent. During this experimentation, we also identified and corrected bugs in the tool, improved error detection, and generated code coverage reports at the package, class, and method levels.