This thesis presents an approach to automatic video game level design consisting of a novel computational model of player enjoyment and a generative system based on evolutionary computing. The model is grounded in player experience research and game design theory and is used to estimate the entertainment value of game levels as a function of their constituent rhythm groups: alternating periods of high and low challenge. In comparison to existing, bottom-up techniques such as rule-based systems, the model affords a number of distinct advantages: it can be generalized to different types of games; it provides adjustable parameters representing semantically meaningful concepts such as difficulty and player skill; and it can facilitate mixed-initiative collaboration between the automated system and a human designer. The generative system represents a unique combination of genetic algorithms and constraint solving methods and leverages the model to create fun levels for two different games.
展开▼
