Boid

- This is a Neuroevolution simulation that combines both Neural Networks and Genetic Algorithms to play the game Flappy Bird.
- It takes about 20 generations for the boid to become good at the game, and about 50 to master it.
- The simulation speed and birds per generation can be adjusted using the sliders at the bottom left of the page.
- Well performing boids can be saved and loaded into the game.

Link: Boid

---

Implement a Flappy Bird clone

• Boid class:

  • Gravity
  • Y-velocity
  • Jump on Key Pressed
  • Die on collision
  • Boid image
  • Change head pointing position while flying

• Pipe class:

  • Two Rectangles
  • X-velocity
  • Spacing
  • Pipe Image

• Create Environment:

  • Create World
  • Create Boid
  • Create Pipes
  • Handle collisions
  • Background Image
  • Make a Scoreboard
  • Track generations

Implement a Neural Network

• Design NN Architecture (10/10/2 vs 6/10/2)
• Create Layers
• Randomize weights
• Add activation functions
• Copy Neural Network function
• Predict / Think function
• Save NN of boid with Best Fitness in JSON format
• Load NN from JSON and assign it to a boid

Implement a Genetic Algorithm

• Create a Generation of Boids
• Implement Mutate function
• Implement a Crossover function
• Copy NN with best fitness
• Create a generation of mutated boids from best NN
• Create a slider for learning speed
• Store boids with best Fitness from every generation

NOTE: It is recommended to have at least 100 boids per generation for a faster training.