Infinite Mario CPU AI

You can see the path it plans to go as a red line, which updates when it detects new obstacles at the right screen border. It uses only information visible on screen.

At the "close call" situation: In this version of Mario, when you're jumping while sliding on a wall, you jump backwards and upwards away from it. That's what the AI did twice to get out of the hole.

For the technical inclined:
It uses a self-implemented A* algorithm (check http://en.wikipedia.org/wiki/A*_search_algorithm), and runs in real-time. If you're taking part in the competition, it's level difficulty 10 with a seed of 30.
arghnesssays...

This Mario AI implementation sort of takes advantage of the rules of the Mario AI competition, apparently.

The rules stated that the AIs shouldn't use information taken from the Infinite Mario engine classes (this would effectively let the AI know what was coming up in the future). Instead, the AIs were meant to use an "observation" class that broke the screen in to a grid and said what was inside each square. This grid was not very fine grained (only ~22*22), and so it didn't give accurate information as to the exact location of enemies on screen, so the problem was a lot harder using this technique.

Robin (the creator of the AI here) worked around the rules a little by almost doing a copy/paste of the engine classes in to his own code, then using those to obtain very exact information.

Valid within the exact confines of the rules, but a bit cheeky and not really in the spirit of the competition!

Send this Article to a Friend



Separate multiple emails with a comma (,); limit 5 recipients






Your email has been sent successfully!

Manage this Video in Your Playlists




notify when someone comments
X

This website uses cookies.

This website uses cookies to improve user experience. By using this website you consent to all cookies in accordance with our Privacy Policy.

I agree
  
Learn More