Survival of the Fittest Bot: strategy selection using unit fitness in Starcraft

Abstract

In real time strategy games(RTSG) the decision what units to produce is a crucial part of the tactical game play. With this thesis we want to test a new way to make this decision based on the notion of unit fitness. Based on the observed enemy Strategy we give values to different properties of Unit types to be able to find the type most useful in the situation. To test this a approach we implemented a bot playing Starcraft and utilising this strategy to determine the units to produce. We then compared the results of matches against the build in Ai with the result the same bot achieves with random or fixed strategies. The results show that the approach performs significantly better.