Maze Solver using Reinforcement Learning

A reinforcement learning project that teaches intelligent agents how to solve a maze through trial and error.

The agent starts with no prior knowledge of the environment and gradually learns the optimal path to the goal using reward-based learning techniques.

This project demonstrates how different Reinforcement Learning (RL) algorithms can be applied to maze navigation problems.

The environment is built using a custom OpenAI Gym Maze Environment, where agents interact with the maze, receive rewards, and improve their navigation strategy over time.

The project includes multiple RL agents and exploration policies for comparing learning performance and efficiency.

• Custom Maze Environment using OpenAI Gym
• Q-Learning Agent
• SARSA Agent
• Policy Gradient Agent
• A* Search Agent for comparison
• Greedy & Boltzmann Exploration Policies
• Interactive GUI Visualization
• Reward-based learning system
• Adjustable training parameters
• Real-time maze solving visualization

• Q-Learning
• SARSA
• Policy Gradient Methods
• Exploration vs Exploitation
• Reward Optimization
• State-Action Value Functions
• Policy-Based Learning

Maze-Solver/
│
├── gym_maze/                 # Custom maze environment
├── AStarAgent.py             # A* search algorithm agent
├── QLearningAgent.py         # Q-Learning implementation
├── SarsaAgent.py             # SARSA implementation
├── PolicyGradientAgent.py    # Policy Gradient implementation
├── GreedyPolicies.py         # Greedy exploration strategies
├── BoltzmannPolicies.py      # Boltzmann exploration strategies
├── gui.py                    # GUI visualization
├── main.py                   # Main project entry point
├── setup.py                  # Project setup
└── README.md

git clone https://github.com/AhmedYasser06/Maze-Solver.git
cd Maze-Solver

python -m venv venv

venv\Scripts\activate

source venv/bin/activate

pip install -e .

Run the main application:

python main.py

The GUI window will open and display the maze environment where the agent learns and navigates toward the goal.

This project uses:

• OpenAI Gym
• Custom Maze Environment

1. The agent starts at a random position in the maze.
2. It performs actions:
   • Up
   • Down
   • Left
   • Right
3. The environment provides rewards:
   • Positive reward for reaching the goal
   • Negative reward for invalid or inefficient moves
4. Over many episodes, the agent learns the optimal path.

• Random movement
• Frequent collisions
• Inefficient paths

• Faster navigation
• Optimal route discovery
• Improved reward accumulation

• Deep Q-Networks (DQN)
• Multiple maze difficulty levels
• Dynamic obstacles
• Saving/loading trained models
• Performance analytics graphs
• Multiplayer agent competition

Abdelrahman Hussieni

GitHub: https://github.com/abdulrahman-hussieni

This project is open-source and available under the MIT License.