Solve a maze using Reinforcement Learning

Overview

This repository contains a C++ implementation of different Reinforceent Learning algorthms which allow to solve the maze problem. The goal is reaching a specified state in a gridworld scenario, starting from any random position.

Algorithms

Four algorithms have been implemented to solve the maze problem:

• SARSA
• Q-learning
• Double Q-learning
• QV(λ)-learning

Both the ε-greedy and Boltzmann exploration strategies are available.

Basic Usage

Clone this repository:

$ git clone https://github.com/nicoleorzan/RL-maze-solver

Compile the code using the Makefile:

$ make

Run the main code, which executes the four learning algorithms on the example maze in figure:

When finished, clean up:

$ make clean

Personalized Usage

You can define the maze you want to solve in the maze_definition file. He you should insert:

• the maze size N, an integer number
• the non-visitable states, defined as integer numbers between 0 adn N-1

References

• Richard S. Sutton and Andrew G. Barto. 1998. Introduction to Reinforcement Learning (1st. ed.). MIT Press, Cambridge, MA, USA.

• Wiering, Marco. (2005). QV(λ)-learning: A New On-policy Reinforcement Learning Algrithm.