daniberg.com

Genetic Programming in Practice

I was playing with the browser canvas to render a few pixels and wrote a simple 2d game inspired by breakout. I thought that would be a good excuse to try the genetic programming code from the previous post.

The idea was to evolve one or more expressions to control the enemy's paddle. And so, I've ported the kotlin code to javascript, extended the expressions to include a few trigonometry expressions and let the computer play against itself overnight.

It did work. The resulting expression is not as graceful as a human playing and the movements have a bias towards the left x position of the paddle but it does the job.

In the next iterations I'll expand the variables fed to the expressions. Currently, only the ball and paddle x positions are exposed. Adding the blocks and opponent positions can in theory yield more strategic expressions.

In order to make it more entertaining I've added a table with scores and the expressions from the player and the opponent in use.

In a real world use case no rendering is necessary and the matches can be parallelized. Nonetheless, it's entertaing to watch the matches and follow the evolution of the expressions.