Evaluating Procedurally Generated Levels : Developing metrics for 3D platforming challenges

Abstract

Evaluating the quality of procedurally generated outputs is still one the important open problems in the field of Procedural Content Generation (PCG). This paper presents an evaluative model for assessing 3D platformer levels, that leverage on graph theory and its proposed graph, vertex, and edge properties. The objective is to build upon the existing knowledge on PCG metrics, and try to develop new metrics and vocabulary towards a model that can automatically assess 3D PCG. The study offers a full look into how these metrics were procured, the design of the test, and how the test performed. The empirical study of the model shows some notable positive correlations between these proposed metrics and user rating, and also deems some of the metrics not so useful. Besides the empirical data, it also provides discussion on the pitfalls of trying generate or test such model - that on its own could provide value when working towards a more formalized evaluative model.