Multi-Agent War Simulation using Reinforcement Learning

A browser-based visualization tool for multi-agent reinforcement learning in a war simulation environment.

Features

• Simulate interactions between multiple agents with different learning strategies
• Configure agent health, number of agents, and mutual animosity
• Visualize simulation results with interactive charts and animations
• Train RL agents and observe learning behavior
• Modern, responsive UI for a better user experience

Technologies Used

• Backend: Flask, Python
• ML/AI: PyTorch, NumPy
• Frontend: HTML5, CSS3, JavaScript, Bootstrap 5
• Visualization: Matplotlib, Plotly

Getting Started

Local Development

1. Clone the repository

   git clone https://github.com/Subramanyam6/Multi-Agent-War-Simulation-using-Reinforcement-Learning.git
   cd Multi-Agent-War-Simulation-using-Reinforcement-Learning

2. Create and activate a virtual environment

   python -m venv .venv
   # On Windows
   .venv\Scripts\activate
   # On macOS/Linux
   source .venv/bin/activate

3. Install dependencies

   pip install -r requirements.txt

4. Run the application

   python app.py

5. Open your browser and navigate to http://localhost:5000

Configuration Options

Agent Types

• RL: Reinforcement Learning agents that learn and adapt over time
• Heuristic: Rule-based agents with predefined strategies
• Random: Agents that take random actions

Health Configurations

• Full Health: All agents start at maximum health
• Low Health: All agents start with low health
• Random Health: Agents start with random health values
• Half-Half: 50% of agents have high health, 50% have low health

Reinforcement Learning Settings

• Discount Factor (β): Determines how much the agent values future rewards
• Learning Rate (α): Controls how quickly the agent updates its knowledge
• Target Update Frequency: How often the target network is updated
• Replay Buffer Size: Size of the experience replay buffer
• Batch Size: Number of samples processed together during training
• Epsilon Parameters: Control exploration vs. exploitation behavior

License

This project is licensed under the MIT License - see the LICENSE file for details.

Author

Bala Subramanyam