An Autonomous Player Agent for Game Balance Insight on an Educational Video Game

This paper explores the use of Reinforcement Learning (RL) to address the challenge of game balance in Educational Video Games (EVG). We adapt Eco City, a city management simulator EVG where the player enacts city policies to maintain the city for as long as possible, as the environment for study. Our RL agent, based on a Deep Q-Network (DQN), is trained to maximize its duration of survival. We analyze the behavior patterns of the agent during both training and evaluation to gain insights into the game dynamics and player strategies. The results show that the agent outperforms random play and exhibits a dominant strategy that effectively ignores half of the possible actions in the game. The analysis of the agent's behavior provides valuable information for game balancing efforts, such as adjusting policy effects and introducing new policies. We demonstrate the potential and the implications of our RL approach to augment the game balance process for developers, particularly in the educational gaming domain.