Reinforcement Learning application on Gamification

Machine Learning, a component of Artificial Intelligence, has gained increasing importance and contribution to our society. This paper presents one of many machine learning methods, Reinforcement Learning, through gamification of a simple maze game. The purpose of this paper is to visualize the basic concept of reinforcement learning and strategy which the reader may be able to apply. The paper compares two model-free learning algorithms: on-policy and off-policy approaches using SARSA and Q-Learning models. Our experiments will demonstrate the difference between on-policy and off-policy by training the agent on mazes of different scales and complexity. Moreover, this paper will also illustrate the effect of Epsilon Greedy Policy. The maze is generated and visualized using Pygame. The result is presented in a table and graph plot for comparisons.