Since Friend-Q learning assumes the same value function for both players, the player will help the opponent score and take the ball into the opponent’s goal. Although the Q-value converges, as ...
Recreation of Correlated Q-Learning paper by Greenwald/Hall - GitHub - furuolan/soccergame: Recreation of Correlated Q-Learning paper by Greenwald/Hall
agent Q-learning, and show how CE-Q, Nash-Q, and FF-Q are all special cases of this generic algorithm. Next, we compare CE-Q learning with Q-learning and FF-Q in grid games. In the following section, we ex-periment with the same set of algorithms in a soccer-like game. Overall, we demonstrate that CE-Q learning
Correlated Equilibria Q-Learning. This project reporduces the results of the "Correlated Q-Learning" by _Amy Greenwald.. In this project I've implemented and compared 4 Q-Learning algorithms in application to Markov game "Soccer" similarly to the paper approach.
Soccer Game environment. This repository can be used to simulate the Soccer Game environment as described in the following paper: Greenwald, A., Hall, K., & Serrano, R. (2003, August). Correlated Q-learning.
Description. The purpose of this project was to test the convergence of four different types of learning algorithms in a simple zero sum markov game (4x2 grid soccer game with 2 players). Each algorithm simulates 1,000,000 turns and checks if the Q-value of a particular state converges. The algorithms tested are: Q-Learning, Friend Q-Learning, Foe Q-Learning, and Correlated Q-Learning.
SOCCER SIMULATOR. This package contains a environment simulator for the Soccer toy game as shown in: Michael L Littman. "Friend-or-foe Q-learning in general-sum Games" 2001. Amy Greenwald and Keith Hall. "Correlated Q-Learning" 2003.
Keyboard controls in our soccer games are player-friendly and meant to turn you into an all-star in no time! Within minutes, you’ll be putting heavy spin on free kicks and dancing around your opponents with ease. Many of our soccer challenges feature in-game tutorials which will help you learn controls and allow you to practice before playing a real game. But, if you want to just launch into soccer action right away, then go ahead!
Nash Q-learning than with a single-agent Q-learning method. When at least one agent adopts Nash Q-learning, the performance of both agents is better than using single-agent Q-learning. We have also implemented an online version of Nash Q-learning that balances exploration with exploitation, yielding improved performance.