Project for the Reinforcement learning course by Erwan LE PENNEC.
The aim of the project is to solve a hard exploration problem using the technique from https://arxiv.org/abs/1901.10995
We created our own "game" to simplify and better vizualise their results.
In the file Projet_RL.pdf, you will find our report that explains everything that was done in this project.
In the notebook Go-Explore.ipynb you will find the code used in this project. It contains the class Phase1 to run the "phase 1" of the Go-Explore algorithm and an Environment class to train an agent on our game using RL.
In the folder Figures and Animations, you will find .gif files that illustrate our game and the results.
The file best_run.csv contains the "ground truth" that was obtained from phase 1. This is used for imitation learning.