Automatic AI design by the use of MCTS for the game Dead-End

Abstract

The goal for Artificial Intelligent (AI) in modern video game field is about creating AI that is challengeable and interesting. This paper aims at implementing a method in which AI is controlled by Monte-Carlo Tree Search (MCTS) instead of Finite State Machine (FSM). Regarding as an automatic AI design, NPC controlled by MCTS outperforms FSM-controlled-NPC in mainly three aspects. We predict that, in order to produce challengeable and interesting opponents by MCTS, the resulting performance of opponent is determined by the length of simulation time of the MCTS method. Thus, we can adjust the opponents' intelligence by changing the length of simulation time. This research is based on Dead-End.