A Deep Q-Network Eith Experience Optimization (DQN-EO) for Atari's Space Invaders and Its Performance Evaluation

During recent years, the deep Q-Learning is used to solve different complex problems in different fields. However, Deep Q-Learning does not have a unified method for solving certain problems because different problems require specific settings and parameters. This paper proposes a Deep Q-Network with Experience Optimization for Atari’s “Space Invaders” environment called DQN-EO. Training and testing results are presented. The performance evaluation results show that while using the proposed algorithm the agent is better at avoiding enemy bullets by 37.7% (longer lifetime) and destroying enemy ships by 14.5% (higher score).