Optimizing Visual Properties of Game Content Through Neuroevolution

This paper presents a search-based approach to generating game content that satisfies both gameplay requirements and user-expressed aesthetic criteria. Using evolutionary constraint satisfaction, we search for spaceships (for a space combat game) represented as compositional patternproducing networks. While the gameplay requirements are satisfied by ad-hoc defined constraints, the aesthetic evaluation function can also be informed by human aesthetic judgement. This is achieved using indirect interactive evolution, where an evaluation function re-weights an array of aesthetic criteria based on the choices of a human player. Early results show that we can create aesthetically diverse and interesting spaceships while retaining in-game functionality. The games industry has often used procedurally generated content in order to increase replayability and cut down on development costs. With the gaming population increasing in size and diversity over recent years (Entertainment Software Association 2011), the need for original content suited to a wider assortment of players of different age, culture and taste has also increased. Experience-Driven Procedural Content Generation (Yannakakis and Togelius 2011) introduces a framework of methods for creating and evaluating experience-centric game content which offers personalized gaming experience within a wide variety of game genres. In this paper we introduce a number of aesthetic filters for evaluating generated content from the perspective of a player’s visual taste and preferences. Using these quantifiable visual properties as the fitness function of a genetic algorithm, the content generator can optimize game elements with the desired visual patterns as dictated either by the player (online) or by a designer (offline). Our approach is unique as it combines neuroevolution with constraint satisfaction (introduced in (Liapis, Yannakakis, and Togelius 2011)) in order to create content which fulfills some minimum requirements while posessing some visual properties deemed important for human perception by studies in cognitive psychology and neuroscience. Most importantly, the paper proposes a novel framework for adapting a user preference model of visual taste by adjusting the importance of each visual property in the content’s evaluation based on the choices of one or more players. This allows for an Copyright c © 2011, Association for the Advancement of Artificial Intelligence (www.aaai.org). All rights reserved. indirect form of interactive evolution, where the player’s choices affect the fitness function determining content quality, allowing for a more holistic aesthetic model. The presented framework is inspired by the Galactic Arms Race game (Hastings, Guha, and Stanley 2009), yet it is distinct in that it evolves the spaceships themselves rather than their weapons, controls the generative process through constraints and proposes an indirect form of preference modelling. This paper builds and extends upon the study presented in (Liapis, Yannakakis, and Togelius 2011) which focuses on the optimization of performance of generated spaceships in a space combat game. The current paper focuses on the optimization of visual properties of the spaceships’ form and on the personalisation of a hand-crafted aesthetic model to individual players. However, the methods suggested in this paper are quite generic and not explicitly designed for creating spaceships; therefore results are often abstract shapes.