Monte Carlo Tree Search Modification for Computer Games
Danil A. Chentsov, Sergey A. Belyaev · 2020
One of the most important tasks of artificial intelligence is to create a universal intelligent agent, which can succeed in solving of many problems instead of only one. In the field of gaming, such intelligent agents train in General Game Playing (GGP). In GGP, one can define a game declaratively in terms of game logic. The AI player then has to decide how to play the game and how to win. One of the most efficient algorithms is the Monte Carlo tree search algorithm (MCTS), but it can work inefficiently if one uses it in general without taking into account the specifics of the task. The paper proposes MCTS, which allows you to change the probability of choosing a solution. The authors evaluated the effectiveness of the proposed solution, and also compared it with other algorithms and methods based on the competitive GVG-AI engine in three games, such as Aliens, Frogs, Zelda.