3D Ludii Games : the case of the Shibumi set

Details

Supervisors

Piette, Eric

Faculty

Ecole polytechnique de Louvain

Degree label

Master [60] en sciences informatiques

Abstract

Given the General Game Playing (GGP) community’s recent interest in modeling, analyzing, and studying 3D board games within the Ludii GGP system, this thesis focuses on two main phases. The first phase investigates the integration of 3D games into Ludii. The second phase explores the experimentation and development of RL techniques and heuristics to enhance AI agents’ performance in these games. Initially, this thesis will focus on Shibumi games, which are generic enough to expand to any 3D game. These games offer strategies that are easy to learn but take a lifetime to master. As for the current literature, Ludii stands as the first General Game Playing system capable of modeling, visualising, and playing any finite extensive-form game through ludemic modeling and a class grammar approach. However, Ludii was developed as part of the Digital Ludeme Project (DLP), which primarily focuses on traditional board games. As a result, the language, state representation, interface, and AI methods used in Ludii’s 1300+ games are not fully optimised for modeling 3D games invented in the last 50-70 years, which fall outside the scope of the DLP. This paper concludes by highlighting certain limitations as well as several promising re- sults of the carried out implementations. In addition, a number of suggestions and ideas for improvements are put forward in order to improve the obtained results and go further. Done implementations are publicly available at: https://github.com/cedantoine/Ludii/tree/dev.