Combining Case-Based Memory with Genetic Algorithm Search for Competent Game AI

We use case-injected genetic algorithms for learning how to competently play computer strategy games. Case-injected genetic algorithms combine genetic algorithm search with a case-based memory of past problem solving attempts to improve performance on subsequent similar problems. The case-injected genetic algorithm improves performance on later problems in the sequence by learning from cases recorded earlier in the sequence. Since game-play in strategy games usually boils down to optimally allocating resources to achieve in-game mission objectives, we describe how a case-injected genetic algorithm player can play our game by solving the sequence of resource allocation problems generated by opponent moves during game-play. When retrieving and using cases recorded from human game-play, results show that case injection effectively biases the genetic algorithm toward producing plans that contain appropriate elements of plans produced by human players.