Imperfect and Cooperative Guandan Game System

Guandan is a type of imperfect information Card games with four players which are divided into two teams. In recent years, great progress has been made in the field of imperfect information games, AI Pluribus and AI Suphx defeat multiple expert players in Texas Hold'em or Mahjong successfully. However, in Guandan game, the cooperation of two players in the same team must be considered. The cooperation relationship in Texas Hold'em or Mahjong is not exist. Therefore, the successful methods mentioned above are not suitable for solving Guandan game. In this paper, we design and implement Guandan game system based on the framework of UCT (Upper Confidence Bound Apply to Tree) algorithm. Firstly, we sample the multiple possible game states through Monte Carlo Sampling algorithm. Secondly, UCT algorithm is used to searching game tree based on every sampled game state. Finally, the best action is selected based on searching results. According to the cooperation relationships in Guandan game, the payoffs of the two players in the same team is considered in the simulation phase of UCT algorithm. It can reduce the card ability loss in the team. The experimental results show that the improved UCT algorithm is better than random strategy in terms of intelligence, and it has some meaning for solving the game problems involving cooperation relationship.