OAT: An Optimized Android Testing Framework Based on Reinforcement Learning.

Automated testing of Android applications is always a challenging task. Deep reinforcement learning can continuously optimize the current exploration strategy through the interaction with the application under test and can explore application states that are difficult to reach in the testing process. However, existing state-of-the-art deep reinforcement learning techniques rely on coarse GUI state definitions, which make them hard to explore interesting application states even with the guidance of reward function. In this work, we propose OAT, an optimized automated testing tool for Android applications based on deep reinforcement learning. OAT is designed with a pair of fine-grained state representation and reward function to provide more effective reward incentives for reinforcement learning. OAT also adopts the Monte Carlo Tree Search (MCTS) strategy to more effectively explore promising GUI states. Our experimental evaluation shows that OAT is more effective than the state-of-the-art Android application testing techniques in terms of both code coverage and fault detection.