Enemy evaluation AI for 2D action-platform game

Abstract

Enemy plays an important role in providing challenge in action-platform games. Our intention is to generate enemies automatically. To ensure that our enemies can be used in actual games, we need to evaluate their qualities. Using human players for the evaluation is not feasible if we generate a lot of enemies or if we are going to use the algorithm in actual games. Our approach is to create a player AI that can play action-platform games with the same performance as human players. That AI's battle performance will then be used to evaluate the generated enemies. This paper presents the definition of battle performance and the AI implementation, combining FSM, search and rule-based heuristics. The AI is shown to perform similar to human players.