Self-Driving Car Simulation

A neural network-based self-driving car simulation built with pure JavaScript, demonstrating artificial intelligence and machine learning concepts in a real-time driving environment.

Project Overview

This simulation implements a self-driving car that navigates through traffic using a custom neural network. The car uses sensor inputs to make driving decisions, providing a practical demonstration of AI and machine learning principles.

Features

• Neural Network AI: Custom-built neural network for autonomous driving decisions
• Sensor System: Ray-casting sensors for obstacle and road boundary detection
• Real-Time Visualization: Live visualization of neural network decision processes
• Traffic Simulation: Multiple AI-controlled vehicles for realistic driving scenarios
• Training System: Save and load functionality for the best-performing neural networks

Project Structure

self-driving-car/
├── index.html          # Main HTML file
├── style.css           # Application styling
├── main.js             # Core application logic and animation
├── car.js              # Car physics and AI controls
├── road.js             # Road generation and management
├── sensor.js           # Sensor system implementation
├── network.js          # Neural network architecture
├── visualizer.js       # Neural network visualization
├── controls.js         # Control system for driving
├── utils.js            # Utility functions
├── car.png             # Car sprite image
└── README.md           # Project documentation

Neural Network Architecture

The neural network uses a feedforward architecture with:

• Input Layer: 5 sensor rays (configurable)
• Hidden Layer: 6 neurons
• Output Layer: 4 neurons (forward, left, right, reverse controls)

Usage

1. Open index.html in a web browser to start the simulation
2. The AI car will automatically begin navigating through traffic
3. Interface controls:
   • Save Button: Save the current best neural network to localStorage
   • Discard Button: Remove saved network and reset simulation

Technical Implementation

Car Physics System

• Acceleration and friction modeling
• Collision detection with road borders and vehicles
• Damage assessment system
• Turning mechanics based on speed

Sensor System

• Ray-casting implementation for environmental scanning
• Multiple sensor configuration for comprehensive detection
• Distance measurement to obstacles

Neural Network

• Custom implementation without external dependencies
• Feedforward propagation
• Mutation capabilities for training optimization
• Local storage integration for model persistence

Getting Started

To run the simulation:

1. Download or clone the project files
2. Open index.html in any modern web browser
3. The simulation will start automatically

Training Methodology

The AI learning process follows these steps:

1. Generate multiple cars with varied neural network configurations
2. Evaluate performance based on distance traveled
3. Select and save the best-performing neural network
4. Use mutation to explore new behavioral patterns
5. Repeat the process for continuous improvement

Customization Options

The project supports various modifications:

• Sensor configuration in sensor.js
• Neural network architecture in network.js
• Traffic patterns in main.js
• Physics parameters in car.js

Educational Value

This project demonstrates:

• Neural network fundamentals and implementation
• Genetic algorithms and evolutionary strategies
• Computer vision concepts through ray-casting
• Real-time physics simulation
• Canvas-based graphics rendering