We use state of the art deep reinforcement learning agents to test games. Its not a simple automated or scripted testing framework, but our AI agent plays games like a human.
For example, performance statistics like CPU usage, memory usage, network usage and frame rendering are shown below for Nintendo's SuperMario on a Huawei phone.
Live stream will appear when training is ongoing.
The deep reinforcement learning framework for games allows us to design an AI engine which is game agnostic. It doesn't have to understand game screen or inputs to learn how to play the game. The article covers the first important part of such a system - recognizing game score.
It’s well known that mobile games are resource hungry apps. That’s why they become the best candidate for testing underlying system performance. We test out two games - Flappy Bird and SuperMario on two Google Pixels running Android 7 and 8 each.