Automation of Balance Adjustment Testing by Human-like Agent that Explores Map for First Person Shooter Game

Abstract

AI agent is applied to balance adjustment testing for First Person Shooter games (BAT4FPS). Current research can generate human-like AI agents for BAT4FPS, though they are limited to explore and play them just around a local position. This study proposes a BAT4FPS approach by a human-like AI agent exploring both locally and globally. We merge different AIs, Rule based AI for global exploration and Deep reinforcement learning based AI for local exploration, switching each other automatically. In this presentation, we show higher coverage of local and global positions than existing research.