Home
A
Simulation
Real Time Genetic Algorithm
B
Best Neural Network
C
Algorithm Parameters
Generation
0
Best Fitness
0.00
Avg Fitness
0.00
Progress
0%
Training Progress
D
Training History
Interactive demonstration of a genetic algorithm training a neural network to play Flappy Bird. (A) Real-time simulation canvas showing the current generation of agents navigating procedurally generated pipe obstacles with adaptive difficulty scaling. (B) Weight and activation visualisation of the best-performing neural network (architecture: 8 inputs → 12 hidden → 1 output); connection thickness encodes weight magnitude; node connection colour describes the signed activation level. Green indicates positive activation, red indicates negative activation. (C) Configurable algorithm parameters: population size, mutation rate/strength, elitism, and maximum generations. (D) Per-generation training log recording best and average fitness scores. Fitness indicates the number of frames an agent successfully navigated before colliding with a pipe obstacle. Performance fluctuations are due to improvement of bird flight, and the adverserial neural network generation of increasingly difficult pipe obstacles.