Reinforcement Learning Project

This repository contains the code and documentation for the Reinforcement Learning project, conducted as part of RL - Apprentissage par renforcement - CS - PARIS - SACLAY (2023-2024) course. The project aims to implement and experiment with various Reinforcement Learning methods on the Highway-env collection of environments.

Project Overview

The project is divided into three main parts:

• Part 1: Highway Environment

  • Implementation of a DQN from scratch to solve the Highway environment.
  • Evaluation of training performance and agent behavior.

• Part 2: Continuous Actions

  • Configuration of an environment for continuous actions.
  • Implementation of an algorithm of choice, possibly using code from lab sessions.
  • Comparison of results with Highway environment using discrete actions.

• Part 3: Stable Baselines Reference Implementations

  • Utilization of the StableBaselines library to train existing algorithms on a chosen environment.
  • Experimentation to study various aspects of the task and algorithm, such as generalization, hyperparameter impact, safety, etc.

Project Structure

The repository is organized into three main folders:

• part1: Contains code and documentation for Part 1 of the project, focusing on the Highway environment.

• part2: Contains code and documentation for Part 2 of the project, exploring continuous action environments.

• part3: Contains code and documentation for Part 3 of the project, utilizing Stable Baselines for reference implementations.

Usage

Each part folder contains the necessary code and documentation to replicate the experiments and results described in the project. Detailed instructions for running and understanding the code are provided within each folder.

Group Members

• Romain SENHADJI
• Baptiste LEMAIRE
• Matthias PICARD
• Paul CANAL

Report

The project report, along with individual reports from each group member, can be found on edunao website.