The purpose of this study is understanding Q-learning through the theory that structures Reinforcement Learning and implementing Q-learning in a Third Person Shooter game. Markov Decision Processes are used as frameworks to solve Reinforcement Learning problems. The methods that are covered in this research are categorized into Dynamic Programming, Monte Carlo, or Temporal Difference. The three methods are used to derive the structure of Q-learning. The Third Person Shooter game is created using the Unity Game Engine. The game is modified to fit the Q-learning method and train the game enemy to play against the player. The training was performed over 6 weeks and results showed a satisfactory increase in the AI performance over the weeks. The results demonstrated that Q-learning successfully improved the enemy AI through a better approximation of the optimal policy.
Aguero Quinteros, Juan Ramon, "Reinforcement Learning: Q-learning in Third Person Shooter Games" (2019). Senior Independent Study Theses. Paper 8454.
Q-learning, Unity, Reinforcement Learning, Third Person Shooter
Bachelor of Arts
Senior Independent Study Thesis
© Copyright 2019 Juan Ramon Aguero Quinteros