This paper was written as part of high distinction work for HIT3046 Artificial Intelligence for Games at Swinburne University of Technology.
Genetic programming has been successfully applied in numerous game-based contexts. Though a popular construct in game development, industry tends to favour manually coded rather than generated behaviour trees despite their demonstrated performance.
This work applies genetic programming to determine a behaviour tree capable of moving an agent through an environment to a goal location whilst avoiding detection.
Tests using five alternate game maps show the potential of the technique to solve this problem.