Automated Testing in Virtual Worlds
Summary
In this study we train an agent in the video game Minecraft, to gather items in the world space and combine them to automatically test the functionality of the crafting system of the game. We are inspired by the ideas of reinforcement learning to teach the agent a certain behavior based solely on previous experience, while we test his ability to correctly predict the exact recipe for crafting specific items. We perform experiments with different test cases to explore the success of such process and the time needed to test the crafting feature. Keeping in mind that our main purpose is to explore functional testing in virtual worlds, we conclude that such an automation process can certainly ease the workload of testing.