Reinforcement learning and computer games: an approach focused on algorithms analysis

Authors

  • Diego B. da Costa Centro de Ciências Computacionais - Universidade Federal do Rio Grande (C3/FURG)
  • Giancarlo Lucca Programa de Pós-Graduação em Modelagem Computacional - Universidade Federal do Rio Grande (PPGMC/FURG)
  • Diana Francisca Adamatti Centro de Ciências ComputacionaisUniversidade Federal do Rio Grande https://orcid.org/0000-0003-3829-3075

DOI:

https://doi.org/10.5335/rbca.v14i2.12500

Keywords:

Computer Games, Machine Learning, Reinforcement Learning

Abstract

The gaming market moves billions of dollars a year and is growing exponentially. Reinforcement learning is a trial and error technique that is directly correlated with this market. Thus, the study of these techniques in popular games becomes relevant, as the case study of this project, the Pac-man. This work aims to use metrics to validate the results obtained from the simulation of reinforcement learning algorithms. The rewards earned by the agent, the exploration, its completeness and the time for each simulation will be validated. Several tests were performed with each algorithm and the results show that for environments with unpredictable behaviors, reinforcement learning tends to take a long time to converge.

Downloads

Download data is not yet available.

Published

2022-07-10

Issue

Section

Original Paper

How to Cite

[1]
2022. Reinforcement learning and computer games: an approach focused on algorithms analysis. Brazilian Journal of Applied Computing. 14, 2 (Jul. 2022), 26–34. DOI:https://doi.org/10.5335/rbca.v14i2.12500.