Finding Game Levels with the Right Difficulty in a Few Trials through\n Intelligent Trial-and-Error

Abstract

Methods for dynamic difficulty adjustment allow games to be tailored to\nparticular players to maximize their engagement. However, current methods often\nonly modify a limited set of game features such as the difficulty of the\nopponents, or the availability of resources. Other approaches, such as\nexperience-driven Procedural Content Generation (PCG), can generate complete\nlevels with desired properties such as levels that are neither too hard nor too\neasy, but require many iterations. This paper presents a method that can\ngenerate and search for complete levels with a specific target difficulty in\nonly a few trials. This advance is enabled by through an Intelligent\nTrial-and-Error algorithm, originally developed to allow robots to adapt\nquickly. Our algorithm first creates a large variety of different levels that\nvary across predefined dimensions such as leniency or map coverage. The\nperformance of an AI playing agent on these maps gives a proxy for how\ndifficult the level would be for another AI agent (e.g. one that employs Monte\nCarlo Tree Search instead of Greedy Tree Search); using this information, a\nBayesian Optimization procedure is deployed, updating the difficulty of the\nprior map to reflect the ability of the agent. The approach can reliably find\nlevels with a specific target difficulty for a variety of planning agents in\nonly a few trials, while maintaining an understanding of their skill landscape.\n