Potential-Field-Based Unit Behavior Optimization for Balancing in StarCraft II

This article presents an evolutionary algorithm for optimizing the offensive behavior of opposing units in the real-time strategy game StarCraft II. Encounters between different unit groups are examined and described. The goal for each group is to deal maximal damage to the opposing group while receiving a minimal amount of damage at the same time. The actions each unit performs are determined by accumulating a number of predefined potential fields. Dependent on the statistics of the involved units, the parameters of these fields then fully describe the behavior of each individual unit. Since this includes a huge number of possibilities, the set of optimal parameter values for both groups in an encounter is obtained by applying an evolutionary algorithm.