PROPOSING AN INTELLIGENT AGENT FOR THE FOUR-SIDED DOMINOES GAME USING THE EXPECTIMINIMAX ALGORITHM

摘要

This paper presents an intelligent agent to play the four-sided dominoes game, typically played in Amazonas state, Brazil. In this work, a version of the double-6 dominoes game is used, played by four players, forming two pairs. This game is of imperfect information and is non-deterministic, if we consider having pieces for any valid side as a chance event. The proposed agent is based on the expectiminimax technique. The methodology used for finding the best choice includes a depth-limited search with phase-related search, and a chance model based on the hidden pieces. We propose and test four strategies with different depths in the phases of the match, and tested each one against a pair playing with a basic strategy and a best strategy obtained by a genetic algorithm. In 10 simulations of 5,000 matches, the best strategy of this study obtained 72.04% of wins against a basic strategy and 58.34 % against the best strategy obtained by a genetic algorithm.