Testing of games through software agents managed by artificial neural networks

Abstract

Abstract With the development of artificial neural networks, their use in different fields is increasing to optimize different activities. The availability of rich programming languages, various existing machine learning software libraries, along with the virtualization technologies has made intelligent technologies accessible not only to large corporations but also to individual programmers. Game development is a complex process involving many specialists, experts in various fields. Single-game testing is an iterative process that covers different game development opportunities, different player abilities, as well as player interaction when working in a team. The question is, is it possible to use intelligent software agents in the game testing process in order to optimize the human, time and financial resources needed? This article introduces our attempt to create an intelligent autonomous agent that plays the Brawl Stars game, managed by an artificial neural network. Initially, the agent is trained by collecting frames from a game that is being played by an experienced player. The frames are then processed and analyzed, and a neural network is trained, which directs the agent to conduct an autonomous game. The developed agent was used in two experiments that confirmed the great potential for the use of artificial neural networks in game testing.