Adaptive Spatial Reasoning for Turn-based Strategy Games

The quality of AI opponents often leaves a lot to be desired, which poses many attractive challenges for AI researchers. In this respect, Turn-based Strategy (TBS) games are of particular interest. These games are focussed on high-level decision making, rather than low-level behavioural actions. For efficiently designing a TBS AI, in this paper we propose a game AI architecture named ADAPTA (Allocation and Decomposition Architecture for Performing Tactical AI). It is based on task decomposition using asset allocation, and promotes the use of machine learning techniques. In our research we concentrated on one of the subtasks for the ADAPTA architecture, namely the Extermination module, which is responsible for combat behaviour. Our experiments show that ADAPTA can successfully learn to outperform static opponents. It is also capable of generating AIs which defeat a variety of static tactics simultaneously.