Implementing a Video Game Artificial Intelligence with Deep Reinforcement Learning

Matthew Cox

Read the ENTIRE PRESENTATION HERE

As the technology within the realm of Deep Learning progresses, it is necessary to understand all the most recent research and libraries in order to properly showcase the practical applications of Deep Learning methods with the latest technology. Of these Deep Learning methods, Deep Reinforcement Learning and Deep Q-Networks can be used with the libraries TensorFlow, OpenAI Gym, and TF-Agents to create an Artificial Intelligence (AI) that performs actions to learn from its interactions and maximize reward without being explicitly taught about its environment. Our research uses the newest version of TensorFlow, TensorFlow 2.1, to implement such an AI with OpenAI Gym to play the Atari game Space Invaders with the goals of achieving the highest score possible and showing how the high score changes depending on the Deep Q-Learning variants used, changes to the Deep Learning algorithm’s hyperparameters, and number of iterations.

• Matthew Cox graduated from Ware Shoals High School . He is currently a senior in computer information systems with a software development emphasis and a minor in cybersecurity. He has done extensive research on machine learning and machine learning algorithms and presented his findings at a previous Lander University College of Science & Mathematics Academic Symposium. He is currently planning to do a presentation at the South Carolina Upstate Research Symposium.