Monster Carlo: An MCTS-based Framework for Machine Playtesting Unity Games
Oleksandra G. Keehl, Adam M. Smith · 2018
We describe a Monte Carlo Tree Search (MCTS) powered tool for assessing the impact of various design choices for in-development games built on the Unity platform. MCTS shows promise for playing many games, but the games must be engineered to offer a compatible interface. To circumvent this obstacle, we developed a support library for augmenting Unity games, and Python templates for running machine playtesting experiments. We also propose ways for designers to use this tool to ask and answer designs questions. To illustrate this, we integrated the library with It's Alive!, a game in development by the authors, and 2D Roguelike, an open source game from the Unity asset store. We demonstrate the tool's ability to answer both game design and player modeling questions; and provide the results of system validation experiments.