Maze Solving Using Reinforcement Learning

This project demonstrates various algorithms for solving mazes, combining classical techniques with reinforcement learning (RL) approaches such as Q-learning and Sarsa. It features a menu-driven interface to generate mazes, solve them using different strategies, and visualize the results for deeper analysis.

---

Features

• Maze generation and interactive visualization
• Wall Follower (classical) maze-solving algorithm
• Q-learning and Sarsa reinforcement learning agents
• Multi-agent Sarsa experiments
• Projectile-avoidance reinforcement learning scenarios
• Performance visualization using matplotlib

---

Requirements

• Python 3.9+
• numpy
• matplotlib

Install Dependencies

pip install numpy matplotlib

Usage

Run the main program:

python src/main.py

You’ll be presented with the following menu:

1 : Create another Maze layout 2 : Wall Follower Solving 3 : Run Q-learning 4 : Run Sarsa 5 : Run multi agent Sarsa 6 : Run Q-learning with projectile 7 : Run Sarsa with projectile 8 : Run Sarsa with projectile (multi-agent) 9 : Exit the application

Project Structure

src/ ├── main.py # Main entry point and experiment menu ├── Grid.py # Maze grid and environment logic ├── Agent.py # Agent behavior and learning ├── Cell.py # Maze cell structure ├── Wall.py # Maze wall logic ├── Figures/ # Output images and visualizations

How It Works

• The maze is procedurally generated and visualized.

• Classical and RL algorithms are used to solve the maze.

• RL agents learn optimal paths through repeated exploration episodes.

• Multi-agent and projectile-avoidance modes introduce advanced RL dynamics.

• All outcomes are visualized using matplotlib for performance and policy analysis.