𐰆𐰺𐰴𐰣

Neuroevolution Boids Simulation

Intelligent Swarm Behaviors

This project presents an advanced version of the classic Boids algorithm that simulates the collective movement of bird flocks and fish schools in nature, enhanced with artificial neural networks and evolutionary learning techniques. Unlike traditional Boids, each individual (boid) learns how to respond to its environment through its own neural network, and these networks evolve over generations through genetic algorithms.

Key Features

• Intelligent Decision Mechanism: Each boid is controlled by a neural network that takes 14 inputs (velocity, neighbor position, edge distances etc.) and produces 4 outputs (direction change, speed adjustment etc.)
• Real-time Evolution: The top 20% of each generation forms the next generation, with continuous improvement through mutation and crossover
• Multiple Themes: Switch between bird, fish, drone, and arrow visual themes
• Edge Behaviors: Different boundary responses including bounce, wrap, and avoid
• Comprehensive Statistics: Generation count, alive individuals, best score, average score, and collision count
• Parameter Control: Adjust population size, elitism rate, mutation rate, and mutation range via intuitive UI

Evolution Mechanism

Boids earn points based on their survival time. After a maximum of 5000 frames, the individuals with the longest survival are selected to create the next generation. The goal is to develop intelligent swarm behaviors that can survive longer and avoid collisions with each successive generation.