Coevolution in Hierarchical AI for Strategy Games

Real-Time Strategy games present an interesting problem domain for Artificial Intelligence research. We review current approaches to developing AI systems for such games, noting the frequent decomposition into hierarchies similar to those found in real-world armies. We also note the rarity of any form of learning in this domain – and find limitations in the work that does use learning. Such work tends to enable learning at only one level of the AI hierarchy. We argue, using examples from real-world wars and from research on coevolution in evolutionary computation, that learning in AI hierarchies should occur concurrently at the different strategic and tactical levels present. We then present a framework for conducting research on coevolving the AI

• Type: Conference
• Year: 2005
• Where: CIG
• Authors: Daniel Livingstone
• Keywords: AI Systems, CIG 2005, Interesting Problem Domain, Real-Time Strategy Games