Procedural content generation of Angry Birds levels using Monte Carlo Tree Search

Monte Carlo Tree Search is a method for searching a decision-making process, usually employed in domains such as general game playing, where an artificial intelligence agent must decide the next move to make in a game simulation. There have been other domains that have been explored for MCTS, one of them being procedural level generation, which involves the automatic generation of game levels which are interesting to play, of an acceptable difficulty level, and not discernible from levels created by humans. In this report, I present a method for using MCTS to procedurally generate new Angry Birds levels, trained on a set of Angry Birds levels from existing games. This approach will scale for a requested level of difficulty by using multiple heuristics. I will examine the viability of the approach using playouts of AI agents on the generated levels, each with their own approach to winning the levels, used to simulate the experience level of human players (from naïve to advanced).