Performance Study of Minimax and Reinforcement Learning Agents Playing the Turn-based Game Iwoki

Abstract

Iwoki math is an abstract board game that consists on placing tiles and that combines the calculation of simple mathematical operations with the spatial perception of two-dimensional objects. Due to its inherent features, it is also a very challenging environment to test different artificial intelligence technologies and methods. In this paper, a series of intelligent agents with different reasoning and decision capacities have been developed based on different artificial intelligence techniques applied to game theory, such as Minimax or Reinforcement Learning. Their capabilities have been tested by playing games with each other, but also against human players, obtaining remarkable results. The experimental results ratify conclusions already known at a theoretical level but also provide a new contribution that could be the basis for future research. © 2021 Taylor & Francis.