ECE 470: Incremental Genetic Algorithm

Demo for the ECE 470 Final Project Submission (University of Victoria)

This demo shows an example of an Incremental Genetic Algorithm. The algorithm will learn to jump across the platforms to make it to the finish line. With the given settings below the algorithm takes approximately 200-250 generations to complete.

Visual Context: 

Corgi's jumping across the screen that are highlighted red are the best agents from the previous generation. Clear / neutral color'd corgis are the new offspring agents with slight mutations.

Incremental Genetic Algorithm Settings:

• Population Size: 50
• Survivors: 10
• Starting Frame Limit: 30
• Increase Frame Limit By: 10 every 3 Generations

Termination Condition:

When an agent first reaches the finish tile, a countdown starts that gives the genetic algorithm 25 more generations to try to improve the best fitness score. If an agent improves the best fitness score within the countdown the timer is reset, and the genetic algorithm is given another 25 generations to try to improve again. Otherwise, it is returned as the overall best fitness score, and the chromosome of the agent that achieved it is returned as the optimal solution.

Credit:

• Tristan Lucas
• Stefan Tomanik
• Tyler Lin