The effects of map features in RTS gameplay : calculating the statistically invisible

Abstract

Levels in Real-Time strategy games are often skilfully crafted by experienced level-designers, either by hand, procedurally generated with highly evaluated systems, or through computer-assisted designing tools. These maps are then used by players, who adopt various strategies and employ them through their methods following a player persona, expecting no discrete advantages hidden within the map’s structural features. Map imbalance is a common topic in games research, with many works tending to approach the problem through creating new, more balanced maps for a specific platform [6; 26]. These are made to achieve a target level of balance and desired features. What has been seldom considered however, is the measurement of map features that cause this imbalance, and the relationships they hold with the competing players. Through an evolutionary process, the link between a measurable map feature and a player’s strategy and characteristics is captured. By gradually evolving maps through change in level structure, in such a way that a particular strategy and personality is set to always win, these features become exposed. Using 4 different AI bots with unique strategies and game-play approaches, different sets of biased maps may be obtained and compared, identifying the differences in maps which gives each of these agents a more likely chance of winning. These various Map features will be quantitatively summarised considering the aforementioned AI Agents in a generic RTS platform, objectively identifying balance through wins and losses, and to what degree they do so.