Modelling and Generating Strategy Games Mechanics

Strategy games are a popular genre of games with a long history, originating from games like Chess or Go. The word “strategy” is derived from the Greek word strathgic (strategos), meaning (the) general, resp. strathgeÐa, meaning “art of commanding”. Both terms directly hint at the subject of strategy games: warfare. Players must utilise armies and resources to their advantage on a battlefield to win a “virtual war”. The first strategy games were published as “Kriegspiele” (engl. wargames) in the late 18th century, intended for the education of young cadets. Since then strategy games were refined and transformed over two centuries into a medium of entertainment. Today’s computer strategy games have their roots in the boardand roleplaying games of the 20th century and enjoy great popularity. In this thesis, we use strategy games as an application for the procedural generation of game content. Procedural game content generation has regained some interest in recent years, both among the academic community and game developers alike. When the first commercial computer games were published in the early 1980s, technical limitations prevented game developers from shipping their titles with a large amount of pre-designed game content. Instead game content such as levels, worlds, stories, weapons, or enemies needed to be generated at program runtime to save storage space and memory resources. It can be argued that the generation of game content “on the fly” has gained popularity again for two reasons: first, game production budgets have risen rapidly in the past ten years due to the necessary labour to create the amount of game content players expect in games today. Secondly: the potential audience for games grew in the past years, with a diversification of player types at the same time. Thus game developers look for a way to tailor their games to individual players’ preferences by creating game content adaptively to how the player plays (and likes) a game. In this thesis we extend the notion of “procedural game content generation” by “game mechanics”. Game mechanics herein refer to the way that objects in a game may interact, what the goal of the game is, how players may manipulate the game world, etc. We present the Strategy Games Description Language (SGDL) and its adjacent framework, a tree-based approach to model the game mechanics of strategy games. The SGDL framework allows game designers to rapid prototype their game ideas with the help of our customisable game engine. We present several example games to demonstrate the capabilities of the language and how to model common strategy game elements. Furthermore, we present methods to procedurally generate and evaluate game mechanics modelled in SGDL in terms of enjoyability. We argue that an evolutionary process can be used to evolve the mechanics of strategy games using techniques from the field of machine learning. Our results show that automated gameplay combined with expert knowledge can be used to determine the quality of gameplay emerging from game mechanics modelled in SGDL, and that algorithms can augment the creativity of human game designers.