AI-Controlled Mario and Level Generation

Last year, the IEEE Symposium on Computational Intelligence And Games hosted a competition where participants were asked to develop an AI that can play mario.  Here's an example of a winning AI player:



This year they've added a level-generation aspect to the competition.

The level generation track of the competition is about creating procedural level generators for Infinite Mario Bros. Competitors will submit Java code that takes as input desired level characteristics, and output a fun level implementing these particular characteristics. The winner will be decided through live play tests.

 For more information, visit the 2010 Mario AI Championship page.
They also have competitions based on Ms. PacMan, StarCraft, and others.

Natural Motion for CG Characters

New algorithms from NaturalMotion allow digital characters to dynamically and realistically respond to changes in their environment.  What is most interesting about this work is that the methods that they use are not hard-coded - rather than completely and painstakingly modeling the motion of walking characters by hand, they use a mixture of physics modeling and evolutionary algorithms to allow the system to 'learn' how to react to the environment.  This enables the characters first to learn about walking, then dynamically adapt to perturbations like pushes and hits from objects and other characters.  This results in very robust and realistic motion.