DeepQ learning in Atari Games

Abstract

The aim of this paper is to develop an AI agent with self-learning capabilities that is able to play classical Atari console games without human intervention and achieve next to human level performance. In order to achieve our goal we will use the OpenAI Gym library that will provide us with a simulated atari game enviorment where we are able to collect important information regarding the agent and it's enviorment. Initially our AI agent is expected to randomly explore the enviorment through the use of Monte Carlo Tree Search and build a Q-Table that will allow us to train a neural network to generalize past experiences and help the agent take the best possible acation given the current state of the game.