Design A General Game Playing Player

Abstract

The idea of GGP (General Game Playing) is to build a computer player, which can win as more games as possible. This player is different from some specialized game players, like Deep Blue that can only play Chess. It needs to learn how to play the game automatically after it gets the rule description of the game. The major part of creating a player is to design a thinking method or strategy to let it learn and play games well. There are two major approaches to design the GGP agents, one is the Heuristics approach and another one is the Simulation approach. In this project, a GGP player is designed based on the Simulation approach. More specifically, it is mainly implemented by MCTS (Monte Carlo Tree Search) and UCT (Upper Confidence bounds for trees). Moreover, some supportive algorithms are also implemented to improve the performance of the player.