Self-Driving Car Simulation

A JavaScript-based self-driving car simulation using neural networks and genetic algorithms. The project implements a complete AI system with no external libraries, featuring real-time visualization of both the car's behavior and its neural network decision-making process.

🚗 Features

• Neural Network AI: Cars learn to navigate traffic using a custom-built neural network
• Genetic Algorithm: Population of 1000 cars with mutation-based learning
• Real-time Visualization: Live display of neural network activity
• Sensor System: Ray-casting sensors detect road borders and traffic
• Collision Detection: Polygon-based collision detection system
• Persistent Training: Save and load the best-performing neural network
• Traffic System: Dynamic traffic obstacles with random colors

🎮 How It Works

Neural Network Architecture

• Input Layer: Sensor rays detecting obstacles
• Hidden Layer: 6 neurons processing sensor data
• Output Layer: 4 neurons controlling movement (↑, ←, →, ↓)

Learning Process

1. 1000 AI-controlled cars start simultaneously
2. Each car uses a neural network to make driving decisions
3. The best-performing car (travels farthest) is tracked
4. Save the best brain to continue training from that point
5. New generations mutate from the saved brain with 10% variation

🚀 Getting Started

Prerequisites

• A modern web browser (Chrome, Firefox, Safari, Edge)
• Local web server (optional, but recommended)

Installation

1. Clone or download the repository
2. Open index.html in your browser, or
3. Use a local server:

   # Using Python 3
   python -m http.server 8000
   
   # Using Node.js (with http-server)
   npx http-server

4. Navigate to http://localhost:8000

🎯 Usage

Controls

• 💾 Save Button: Save the best-performing neural network to browser storage
• 🗑️ Discard Button: Delete saved neural network and start fresh

Training Tips

1. Let the simulation run until most cars crash
2. Save the best brain using the save button
3. Refresh the page - the saved brain will load with mutations
4. Repeat the process to improve performance over time

📁 Project Structure

self_driver/
├── index.html          # Main HTML file
├── style.css           # Styling for canvas and UI
├── main.js             # Entry point, animation loop
├── car.js              # Car class with physics and rendering
├── controls.js         # Input handling (keyboard/AI)
├── sensor.js           # Ray-casting sensor system
├── network.js          # Neural network implementation
├── visualiser.js       # Neural network visualization
├── road.js             # Road rendering and boundaries
├── utils.js            # Utility functions (lerp, intersection, etc.)
└── car.png             # Car sprite image

🔧 Technical Details

Car Physics

• Acceleration: 0.2 units/frame
• Max speed: 3 units/frame (AI cars), 2 units/frame (traffic)
• Friction: 0.05 units/frame
• Steering: 0.03 radians per frame

Neural Network

• Type: Feedforward neural network
• Activation: Custom threshold-based activation
• Mutation Rate: 10% linear interpolation
• Training Method: Parallel genetic algorithm

Sensor System

• Ray count: Configurable (default in network architecture)
• Ray length: Detects road borders and traffic
• Ray spread: Full frontal arc coverage

🎨 Customization

Adjust Car Population

In main.js, change the value of N:

const N = 1000; // Change to desired number

Modify Mutation Rate

In main.js, adjust the mutation parameter:

NeuralNetwork.mutate(cars[i].brain, 0.1); // Change 0.1 to desired rate

Change Traffic Patterns

Edit the traffic array in main.js to add/remove obstacles:

const traffic = [
    new Car(road.getLaneCenter(1), -100, 30, 50, "DUMMY", 2, getRandomColor()),
    // Add more traffic cars here
];

🧠 How the Neural Network Works

1. Input: Sensor readings from ray-casting (distances to obstacles)
2. Processing:
   • Inputs are multiplied by weights
   • Biases are added
   • Activation function determines output
3. Output: Four control signals (forward, left, right, backward)
4. Learning: Genetic algorithm selects best performer and mutates it

📊 Performance

• Runs at 60 FPS on modern hardware
• Handles 1000+ simultaneous car instances
• Real-time neural network visualization
• Efficient collision detection using polygon intersection

📝 License

This project is open source and available for educational purposes.