Optimization and simplification of dynamic scripting with evolution strategy and fuzzy control in a fighting game AI

We develop a combination of evolution strategy (ES) and fuzzy control (FC) to optimize and simplify an existing dynamic scripting (DS) AI called CodeMonkey for fighting game FightingICE, recently used as a platform in international game AI competitions. One of the major issues in DS is that the user has to decide a lot of parameters whose process is not simple. In addition, a complex user-defined-action-rule set for DS is required. The purpose of our work is to automate this troublesome process and improve the performance of the DS technique. We apply (1+1)-ES with the one-fifth rule for tuning parameters and FC for easing definition of additional action rules. We verify our AI's performance in FightingICE. The test results show that although the proposed AI does not require complex parameter and rule setting by the user, it defeats the original CodeMonkey.