Evolutionary optimization of enemy behaviors in RPGS
Authors: Samule Taripin and Jason Teo Tze Wi
Computer games have long been the test bed of choice for implementers of artificial intelligence (AI) techniques since such games exhibit myriad complex, dynamic and highly non-linear elements that require human-like intelligence in order to be successful in achieving the objectives of the game. One of the most popular genres of current computer games is role-playing games (RPGs). As such, numerous recent studies in game AI have attempted to implemented advanced AI techniques such as evolutionary algorithms towards solving different aspects of RPGs such as the main hero’s behaviours for quest completion, automated map and dungeon layout generation, and automated storyline generation among others. In RPGs, one of the most important and often en-countered aspects of practically all hero quests involves confrontations with game-generated enemies most commonly in the form of hostile monsters that attempt to defeat the hero through repeated waves of single or group attacks. However, this is one aspect of RPGs that has yet to be investigated for augmentation using AI techniques. Most enemies in RPGs are generated based on a fixed set of classes with behaviours that are easily predictable by experienced RPG players. This project uses evolutionary optimization algorithms in the form of a Genetic Algorithm (GA) to increase behaviour diversity for the same type of monster in the open source game platform called FLARE thereby giving them the ability to adapt against the player’s abilities. The main objective of the project is to design the GA that can evolve enemy behaviours inside FLARE that attempt to defeat the game player. Experiments are conducted to tune the GA while showing its ability to create evolving enemy in RPGs. Results show highly promising evolution of enemy behaviors in FLARE when tested against a hand-crafted AI hero as well as against 5 human players controlling the hero.