PhD initial subject: "Procedural generation of military compound maximizing the perception of variability in the player", in collaboration with Ubisoft Bordeaux.
Subject eventually presented in the manuscript: "Procedural generation of game levels combining constructive approach and optimisation".
Links
Click here to see the paper published in IEEE Transactions on Games (English) Click here to access a public release of the paper (English)
The research work is positioned in the field of procedural content generation in video games. This study focuses more specifically on the questions related to the diversity and quality of the generated content. This subject raised our interest and led us to study numerous methods and algorithms, through a literature review of procedural content generation. We considered the following problematic: "How to achieve a generation method that offers a high degree of diversity of game experiences, while maintaining a certain structural quality in its results?" Our study is centred in particular on the level design and the placement of objects in game levels. We also targeted a generation of non-linear open 3D levels for a first-person shooter, with a settlement infrastructure whose contents are positioned on a 2D grid. This thesis introduces a new method, named Genetic-WFC, in order to answer our research question. It is a procedural generation pipeline that combines a genetic algorithm and a simulation-based evaluation with the Wave Function Collapse, a local adjacency constraint propagation algorithm, to generate levels targeting specific game experiences. Various experiments on our approach have been performed in order to establish, among other things, its performance against other similar methods and to explore the diversity of game experiences that our generation algorithm can provide. We conclude by mentioning several directions for improvement and continuing research that can be further explored.