Shaymus22
This was made for my Artificial Intelligence class at UNH using A* heuristic search using Julian Togelius and Sergey Karakovskiy’s Mario AI Competition API http://julian.togelius.com/mariocompetition2009/
Source
Similar Posts
6 thoughts on “Infinite Mario AI Progression”
Comments are closed.
Thanks! I'm using this API (I just added the link to the description) and feeding the AI the map directly. As for heuristics, I developed several different values that are added or subtracted based on the state of each iteration of the world. The interesting thing is that due to the nature of Mario's movements, the controls that are enabled or disabled play a major role in the prediction of future states.
I iterate through all given possible configurations of the controls and generate as many different states as I can before I run out of time, since it is running about 8-12 times per second (based on the generated plan length). AI is interesting, but I find it very challenging at times…the difficult part for this, I found, was the physics simulations for predicting each state. It was very calculation-intensive, which is why I had to vary my calculation times so much.
Hello where i can find Mario game with the API so i can control Mario using programing as you do ? i want to make the same idea as you and put it in a project for my university final project, if possible to help me i will be grateful!
This looks like it'd be fun to play as a normal game…can it become a non-AI-controlled Mario game?
What are the red lines coming from mario that run along the ground? :
I would like the game even better if it had the efficiency to add more enemies (Lakitu, Hammer Bro, Flying Koopa Paratroopa, etc.), more power-ups (Star, Super Leaf, Hammer Suit, etc.) and more themes (Underwater, Desert, Sky/Athletic, etc.). I wonder if that'll ever happen. ?