Autonomous Car Simulation with Genetic Algorithm

This project simulates autonomous cars learning to navigate a track using neural networks and a genetic algorithm. The cars learn to avoid walls and dynamic obstacles through evolutionary training.

Overview

The simulation features:

• Cars with neural network "brains" that control their movement
• A procedurally generated racetrack with inner and outer boundaries
• Dynamic obstacles that move along the track
• Genetic algorithm for evolving better-performing cars over generations
• Real-time visualization of car sensors (LiDAR-like rays)
• 3D-like visualization of the car's perspective
• New: A dedicated "About" panel providing detailed information about the simulation, its features, and controls.
• New: A clear information bar above the simulation canvas displaying real-time generation, speed, and distance from obstacles.
• New: Comprehensive training analytics dashboard with real-time charts and performance metrics.

Files Structure

• index.html - Main HTML file
• js/ - Directory containing all JavaScript files
  • nn.js - Neural network implementation using TensorFlow.js
  • ga.js - Genetic algorithm functions
  • particle.js - Car/Agent class with neural network brain
  • ray.js - Raycasting for sensors
  • boundary.js - Track boundary and obstacle classes
  • sketch.js - Main p5.js sketch with setup and draw functions
  • loadbinary.js - Binary file loading utilities
  • p5.collide2d.js - Collision detection library
• css/ - Directory containing stylesheet files
  • style.css - Main stylesheet for the application
• ml-data/ - Pre-trained models
• model.json - Base model structure
• git-commands.sh - Git helper script

Prerequisites

• A modern web browser (Chrome, Firefox, Edge, Safari)
• Internet connection (for loading external libraries)

Instructions to Run

1. Open index.html in a web browser
2. The simulation will start automatically
3. Watch as cars evolve over generations to better navigate the track

Controls

• Dynamic/Static Toggle: Switch obstacle movement on/off
• Obstacle Count: Set number of obstacles (enter number and click "Done")
• Speed Slider: Control simulation speed
• Save Best: Save the best performing car's neural network
• Load Model: Load a pre-trained model
• Settings: Toggle visibility of the simulation settings panel.
• About: Toggle visibility of the about information panel.
• Analytics: Open the comprehensive training dashboard with real-time charts and performance metrics.
• Toggle Theme: Switch between light and dark UI themes.

Version Control

This project uses Git for version control. To set up version control (already done):

git init
git add .
git commit -m "Initial commit: Autonomous car simulation with genetic algorithm and neural networks"

For future development:

# Check status
git status

# Add changes
git add .

# Commit changes
git commit -m "Description of changes"

# View commit history
git log

Git Helper Script

A helper script git-commands.sh is included to make common Git operations easier:

• Run ./git-commands.sh and follow the prompts
• Options include checking status, committing changes, viewing logs, and switching branches

Branching Strategy

• main branch: Stable, working versions of the code
• development branch: Active development branch for new features

When working on new features:

1. Switch to the development branch: git checkout development
2. Make your changes
3. Commit your changes
4. When ready to release, merge to main: git checkout main && git merge development

Technical Details

Neural Network Architecture

• Input layer: 13 neurons (sensor readings)
• Hidden layer: 26 neurons with sigmoid activation
• Output layer: 2 neurons with sigmoid activation (steering angle and speed)

Genetic Algorithm Parameters

• Population size: 100 cars
• Mutation rate: 20%
• Lifespan: 30 frames
• Sensor range: 80 pixels

Sensors

Cars have 13 sensors (rays) spread across a 130-degree field of view to detect:

• Distance to track boundaries
• Distance to dynamic obstacles

Code Improvements

The code has been refactored to improve:

• Modularity: Each functionality is separated into its own class or function
• Readability: Added comprehensive comments and consistent formatting
• Documentation: Detailed JSDoc comments for all functions and classes
• Maintainability: Improved code organization and structure

Training Analytics Dashboard

The simulation now includes a comprehensive analytics dashboard that provides:

Real-time Charts

• Fitness Evolution: Tracks best and average fitness scores across generations
• Lap Completion: Monitors progress in track completion over time
• Population Survival: Shows percentage of agents surviving each generation
• Speed Performance: Displays average speed metrics throughout training

Performance Metrics

• Current Generation: Live generation counter
• Best Fitness: Highest fitness score achieved
• Best Laps: Maximum laps completed by any agent
• Average Survival: Population survival percentage
• Training Time: Total time spent in training

Advanced Analytics

• Improvement Rate: Measures fitness improvement over recent generations
• Convergence Score: Indicates how close the population is to optimal performance
• Diversity Index: Measures genetic diversity within the population
• Success Rate: Percentage of agents completing at least one lap

Data Export

• Export training data as JSON for analysis
• Reset statistics and start fresh training sessions
• Historical data tracking with configurable retention

Refactored Files

1. nn.js: Neural network implementation with detailed documentation

2. ga.js: Genetic algorithm functions with clear explanations

3. particle.js: Particle (car) class with comprehensive sensor and control logic

4. ray.js: Ray casting for obstacle detection with improved comments

5. boundary.js: Track boundary and obstacle classes with better structure

6. sketch.js: Main simulation logic with organized setup and draw functions

7. dashboard.js: New comprehensive training analytics system with Chart.js integration

Development Roadmap

For detailed development tasks, feature requests, and implementation priorities, see TODO.md.

The roadmap includes:

• Performance optimizations and memory management
• Enhanced track generation and curriculum learning
• Advanced neural network features and training methods
• Comprehensive testing and validation framework
• UI/UX improvements and accessibility features

Libraries Used

• p5.js - Creative coding framework
• TensorFlow.js - Machine learning library
• p5.collide2d - Collision detection library

License

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