Masters by Research | Theorised Best Parent Optimisation

Abstract

Discovering and delivering what players want in games is a constant struggle that begins during the development process and extends into post-production as developers try to improve their game. In item-driven games, player retention and engagement is traditionally founded upon the quality and variety of the items available. This is particularly apparent in First Person Shooter (FPS) games where items tend to entirely dictate the actions available to the player. To this end, developers have investigated the use of Procedural Content Generation (PCG) tools to quickly and efficiently create large amounts of content for their games. PCG has been traditionally controlled by generating outputs within very tight restrictions to ensure that all generated outputs function correctly and within the developer defined bounds. However, this directly contradicts the purpose of PCG item generation which is to create new and unique procedurally emergent items. To solve for this, new techniques have been developed leveraging the use of Genetic Algorithms to 'evolve' items that conform to design constraints but can remain novel and creative within those bounds.

Further, if these design constraints are directly reflective of the current player's interactions with the game, items can be generated that empirically suit the player's game playing style; this is one example of Experience-driven Procedural Content Generation. Genetic Algorithms rely on the discovery of the best population member to pass their genetic code onto the next generation; finding a perfect solution using this method requires large population numbers and many generations to evolve. This is important when adapting these types of algorithms into the context of an FPS. All inputted data comes from the player's interactions with each generated item, requiring the system to be able to function effectively without the player sorting through thousands of items. However, this could be remedied if each generation the theoretically ideal best parent was estimated and used in creating the new generation. This solution would reduce both the population size requirement as well as the amount of generations, whilst still providing results varied enough to promote playing style creativity and exploration.