RPG Gym: RPG Game Simulator for Deep Learning

Abstract

Recently, artificial intelligence research using deep learning has been actively conducted. In particular, research related to artificial intelligence is being verified for performance in game environments, and being applied to improve the quality of content in games in the game industry. For example, game balance is adjusted using various data analysis and artificial intelligence methods. In addition, imitation learning or reinforcement learning is used to diversify game characters' behavior control and improve performance. However, there is no suitable simulator for conducting these studies and few simulators are similar to commercial games. OpenAI Gym or Mujoco that are mainly used to learn character behavior control, but these are very different from commercial games. Therefore, this paper proposes a new game simulator with an ARPG game structure that can be used in game balance adjusting and deep learning. The proposed simulator was implemented using Unreal Engine 5, a commercial game engine, and API for applying a deep learning framework is designed and provided. In addition, we conduct game balance experiments and character behavior control experiments using deep learning in the proposed simulator and measure performance.