SEMANTIC STRUCTURES FOR RTS ARMY PREDICTION

摘要

In this paper we describe a semantic structure based approach for AI for Real Time Strategy games. The semantic structures provide a shared knowledge base across the different components of an RTS AI. We outline how the a semantic structure works in conjunction with decision making algorithms such as utility systems. We also introduce a prototype implemented in StarCraft Ⅱ to illustrate how the semantic structures may be used in practice, and provide some detail on how this is implemented in the chosen game platform. Tests demonstrate that the AI is able to choose actions according to high level strategies adapting according to the current game state. Additionally the test conditions used additionally involve human-AI cooperation. Finally, we present an evaluation of the AI, making a detailed study of a typical instance of game play, exploring the AI actions and decision making process.