Zusammenfassung: |
While several researchers have applied case-based reasoning techniques to games, none have addressed the problem of learning to win real-time strategy (RTS) games. Using a different approach, Ponsen and Spronck [10] report good results for learning to win a RTS game (Wargus), assuming a static opponent. We introduce a plan retrieval algorithm that removes this assumption by using three key sources of domain knowledge. We report that its implementation in the Case-based Tactician (CAT) significantly outperforms the best among a set of genetically evolved plans when tested against random WARGUS opponents. CAT communicates with WARGUS through TIELT, a testbed for integrating and evaluating decision systems with simulators. This is the first application of TIELT. We describe this application, our lessons learned, and our motivations for future work. (This article is a reduced version of the 2005 International Conference on Case-Based Reasoning paper of the same title.) |
