Machine learning example that plays mario !

Kobusvdwalt9

I found this while looking at some machine learning examples and I think this is super cool. Ive managed to replicate it with a canabalt clone but I am still learning this as I go. Perhaps I will post a simple tutorial once I have something that works properly. Has anyone done anything similar?

Here is the video :


PS :
What is the forums attitude towards posting content like this ? Just linking to another thing. Is it generally seen as OK ?

mattbenic

That was fantastic, thanks for sharing it :)

garethf

Very cool. We did something similar in Uni with evolving neural nets and swarms of maze solving agents, was cool to watch them learn. And a bit freaky, there was this distinct tipping point where they went from bumbling around to very quickly solving the maze with confidence that.

When SC is done, I wanna play around with AI again, it's so much fun.

dislekcia

I've always enjoyed AI, at university I'd always overengineer every project - even did a genetic algorithm approach to back-training neural nets to play a game called Bao. This NEAT approach looks really interesting, I like how it modifies the NN structure instead of creating a fixed number of neurons and fitting signals back to inputs. Imma read me that paper :)

dislekcia

Here's an HTML5 genetic algorithm implementation that tries to evolve a stable walking system: http://rednuht.org/genetic_walkers

Popped up in my twitter feed today at random :)

raithza

Are there any games that are centered around setting up selection criteria to optimise a genetic algorithm? I feel like that could potentially be interesting. It seemed like mew-genics may have been leaning towards that direction.

Maybe it's a good theme for a jam...?

Kobusvdwalt9

@raithza Interesting.. So you would have to breed for example the best frog for accomplishing a task like jumping over obsticals or racing other frogs ?