Genetic Algorithms as a Basis for GOAP AI Tuning

    Gamers are expecting a more immersive gameplay experience with each new generation of hardware. One of the challenges game designers face in creating a more immersive experience is creating believable artificial intelligence (AI). Players are demanding a higher degree of believability out of their opponents. To meet this demand designers have to develop complex AI systems. As the complexity of the AI system increases it becomes more specialized towards the gameplay. This can be an issue when elements of gameplay change as it causes the developers to rebalance or redesign the AI.

    This thesis examines the viability of using a genetic algorithm to balance the priorities of a Goal-Oriented Action Planning (GOAP) AI system. This reduces the challenge of adapting AI as gameplay changes while still maintaining the ability for developers to specify how the AI plays the game. It could also potentially serve as a predictive development tool. While human developers and testers will still be an integral part of the development lifecycle, this thesis predicts that using an adaptive system to tune AI agents is viable for the purposes of gameplay balancing and could potentially provide insight into the gameplay itself. To test this thesis the project examines the implementation of such a system on bots in Unreal Tournament 3.