Extraordinary MarI/O – Machine Learning for Video Games






MarI/O – Machine Learning for Video Games

MarI/O is a program made of neural networks and genetic algorithms that kicks butt at Super Mario World.
Source Code:
“NEAT” Paper:
Some relevant Wikipedia links:

BizHawk Emulator:

SethBling Twitter:
SethBling Twitch:
SethBling Facebook:
SethBling Website:
SethBling Shirts:
Suggest Ideas:

Music at the end is Cipher by Kevin MacLeod
MarI/O – Machine Learning for Video Games

45 Replies to “Extraordinary MarI/O – Machine Learning for Video Games”

  1. How are you displaying the state of MarI/O's brain? Is that debug display built into NEAT (in some way), or is it something you made for this video?
    Also, is there any chance the stuff shown in this video was recorded on a VM or something and is still on your hard drive somewhere? If so, I encourage you to upload it somewhere so we can see it for ourselves, and possibly even reverse engineer it to understand it better.

  2. Hey guys, sethbling here. Remember when we made MarI/O? I made it in Minecraft using command blocks and armor stands.

  3. could you make this work in a 3D environment? or is the power requirement to much ?

    really smart video, very interesting.

  4. Is it possible to train these neural networks to simulate the reaction time of a real human (e.g. a visual stimulus only gets processed into inputs after a 6-frame delay)?

  5. Just a question. Does it learn the exact sequence of button presses, or does it learn the behavior of items. I mean, will it succeed at another different level with the same items but arranged differently?

Leave a Reply

Your email address will not be published. Required fields are marked *