Please use this identifier to cite or link to this item: http://hdl.handle.net/10609/119886
Title: Agente Sonic. Deep reinforcement learning
Author: Alemán de León, Cristóbal Daniel
Tutor: Nuñez Do Rio, Joan Manuel
Others: Ventura, Carles  
Abstract: Reinforcement learning is a branch of artificial intelligence that studies algorithms capable of making the systems learn to do tasks automatically without using traditional algorithms. They are based on an achievement system in which the right actions are positively rewarded. Within these algorithms, we can find Deep Q-Network, which uses profound neural networks for complex environments such as video games.The purpose of this project is the creation of a DQN agent that learns to overcome different levels of a video game based on the challenge proposed by the team OpenIA in 2018. In this challenge, the creation of agents able to overcome different levels than the ones used for training them is suggested. Using the Gym Retro library, OpenIA provides us with the tools needed to carry this challenge out. These tools consist of observations, actions, and rewards for completing levels of the game Sonic the Hedgehog¿. In the end, the agent developed will be able to take actions that allow it to obtain a larger horizontal movement within each level. The environments where we evaluate the agent are different from the training environment. This way, we check the results of the generalization made by the algorithm of Deep Learning in an unknown environment.
Keywords: machine learning
artificial neural network
deep learning
Document type: info:eu-repo/semantics/bachelorThesis
Issue Date: Jun-2020
Publication license: http://creativecommons.org/licenses/by-nc-nd/3.0/es/  
Appears in Collections:Bachelor thesis, research projects, etc.

Files in This Item:
File Description SizeFormat 
calemandTFG0620memoria.pdfMemoria del TFG450,39 kBAdobe PDFThumbnail
View/Open