Survey of Opponent Modeling: from Game AI to Combat Deduction

Significant advances in game AI and military intelligent decision making have been achieved through the rapid development of intelligent games. However, the actual combat environment is highly complex, encompassing large state space and more unseen battlefield factors. As a result, applying game AI techniques to intelligent combat deduction still presents considerable challenges. Opponent modeling is one of effective approaches to addressing the complexity of game confrontation. Firstly, we focus on opponent modeling technology in game AI, analyzed the benefits and drawbacks of the relative opponent modeling technology used in game AI. And then we discuss the difficulties of transferring opponent modeling to combat deduction. Finally, we present the path for opponent modeling’s future development.

I. Introduction

As the Drosophila of artificial intelligence, computer games have long been as testing grounds for advances in artificial intelligence [1]. The competition between humans and computers serves as a significant indicator of computer development. In 2017, AlphaGo [2], developed by DeepMind, achieved a groundbreaking victory over the human Go champion Lee Sedol. These achievements demonstrate that human have solved the complete information game like chess and Go, prompting researchers to shift their attention to the realm of incomplete information games represented by Texas Hold’ em. In 2019, Pluribus [3], developed by Carnegie Mellon University in the United States, reached another milestone in the field of artificial intelligence by defeating professional human players in multiplayer no-limit Texas Hold’ em. Additionally, in 2019, AlphaStar [4] achieved the master level and outperformed 99.8% of human players. The above achievements indict series of breakthroughs in the domain of intelligent game rivalry, as depicted in Fig. 1.