Sarsamax implementation with OpenAI Gym and Unity engine

Abstract

This paper is focused on presenting a Q-Learning solution to a simple reinforcement learning problem that has been implemented in Unity Game Engine. The method has been written as a python script which communicates with the unity executable with the help of Unity ML-agents package through an interface that was first implemented by OpenAI Gym. The paper begins with familiarizing the reader with reinforcement learning; more specifically, the aspects of it that are necessary to have a clear understanding of the proposed solution. Please note that reinforcement learning is an ever-growing field and not all relating topics are covered in the aforementioned description. Then a brief description of OpenAI, Gym, Unity Game Engine and Unity ML-Agents is given in order to provide more context on how the solution works and how said parts are integrated. After that the titular solution is explained and effects of experiments are shown. The paper presents the solution implemented using Python and Unity ML-Agents on Unity-made environment as easy to do and straightforward. It is proposing that the framework brings significant value by considerably quickening the modeling process of reinforcement learning solutions.