The results of the Mario Bros. AI competition have been made available. (Not really surprising that the top three entries used A*.)
Robin Baumgarten has won, and has made his source code available. AIGameDev.com also has an interview with Robin.
The results of the Mario Bros. AI competition have been made available. (Not really surprising that the top three entries used A*.)
Robin Baumgarten has won, and has made his source code available. AIGameDev.com also has an interview with Robin.
Julian Togelius and Sergey Karakovskiy have organized a competition to create an agent (or AI) that plays the video game Super Mario Bros. – or, more accurately, Infinite Mario Bros. a tribute game featuring random level generation.
The advantage of using Infinite Mario Bros. is the random level generation – which can let the agent learn more generalized playing tactics rather than tactics that are tailored to a static set of levels as in Ms. Pac-Man or Pitfall.
I look forward to seeing the results of the competition, and hope to see source code published as well.
The Netflix Prize has entered the 30-day notification period as a team has announced that they have achieved a 10.05% improvement over the original Cinematch algorithm.
Some further background on the contest can be found in a nice writeup in Wired from last year.
From Rutgers university comes a learning algorithm that they have applied to playing the Atari 2600 game “Pitfall!”.
An example video is on YouTube.
One of the research papers is apparently here (although the site isn’t being very responsive at the moment).
I’ll get around to posting on machine learning for Pac-Man/Ms. Pac-Man at some point as well.
(Spotted on Kotaku and GameSetWatch.)