Mario AI built with Machine Learning - Difficulty 2

This is a Mario AI which I created for my Game AI class. It uses the competition code from http://julian.togelius.com/mariocompetition2009/. Our assignment required that the agent be built with machine learning; I used a k-Nearest Neighbors algorithm and trained my agent with Robin Baumgarten's agent, who won the 2009 competition. The agent entirely learned all of its behavior; the only logic it uses is to recognize features in the world.

In this video, Mario makes his way through a level with Difficulty 2. In this difficulty, you can see Mario navigate over pits. This also demonstrates how the algorithm was not that effective at handling different numbers of enemies; in general, supervised learning is much more difficult when the number of observable features can vary.

Category:

People & Blogs

Tags:

• marioDiff2

[j7512]

As a machine learning enthusiast, I found that very impressive!

[mutehero7]

I wish I could just plug learning AI's into any video game. I'd be entertained for life.