An Approach to First-Person Shooter Games Using Deep Reinforcement Learning

Saadati Razian, Jafar; Asgharzadeh-Bonab, Akbar; Mohamadi, Alireza; Jabarzadeh, Saber

doi:10.22034/ijwg.2024.453676.1083

An Approach to First-Person Shooter Games Using Deep Reinforcement Learning

Document Type : Original Article

Authors

Jafar Saadati Razian ¹

Akbar Asgharzadeh-Bonab ²

alireza mohamadi ³

Saber jabarzadeh ⁴

¹ Lecturer, Imam Ali Officer University, Tehran, Iran

² Department of science and technology studies. AJA Command and Staff University, Tehran, Iran.

³ Department of Science and Technology Studies, AJA Command and Staff University,Tehran, Iran

⁴ Department of science and technology, AJA Command and Staff University

10.22034/ijwg.2024.453676.1083

Abstract

Objective: The main objective of this research is to investigate and improve the performance of Double Deep Q-Learning Network models in first-person shooter games with a focus on intelligent competition.

Methodology: In this research, Deep Q-Network (DQN) and Deep Double Deep Q-Network (DDQN) models are used for the game War. First, the DQN and DDQN models are evaluated, and then their performance is improved using the Prioritized Experience Replay (PER) method. Three experimental game environments are used to evaluate and assess the models.

Findings: The findings of this research show that the proposed Double Deep Q-Network architecture with the Prioritized Experience Replay method has better performance than other proposed algorithms in this field.

Conclusion: The use of the Prioritized Experience Replay method in reinforcement learning has significant advantages that lead to improved performance of the artificial intelligence agent. This method, by utilizing high-quality data and experiences, focuses specifically on more informative experiences, thus significantly increasing sampling efficiency

Keywords

Deep learning

AI War Game

Deep reinforcement learning

Subjects