Learning Strategies of Intelligent Virtual Agents in Video Games: Exploring Types of Decision-Making Using DQN

The chapter delves into the learning strategies of Intelligent Virtual Agents (IVAs) in video games, focusing on how the DQN algorithm influences their decision-making process, especially in videogames. An introduction to experimentation with IVAs in the context of video games is provided, highlighting their relevance in training agents for strategic and tactical decisions. The experimental design is detailed, including the algorithm parameters and study factors, to provide a comprehensive understanding of the experiment conditions. Three variants of the IVA are described, each with a different focus on decision-making strategy, to explore various tactical and strategic approaches. The chapter examines how the DQN algorithm in the video game can be considered a cognitive model, relating its functioning to fundamental human cognitive processes. The results of the experiments show how the discount factor (gamma) affects the performance of the DQN agent in different decision-making contexts, highlighting the effectiveness of a discount factor of 0.9 in maintaining high and consistent performance. In conclusion, it is emphasized that the choice of discount factor has a significant impact on the performance of the DQN agent, with a discount factor of 0.9 emerging as the most effective in maintaining optimal performance in the video game.