They learn to use the orange cube to jump over the wall. This demo will be included in the @unity3d ML-Agents 0.3 release in about a month. #MachineLearning pic.twitter.com/MSWsHU6Bcy
— PunchesBears ㅎ㉨ㅎ (@punchesbears) February 7, 2018
You thought the virtual worm was impressive, watch this blue block "learn" how to use a tool to get over an obstacle. Created by a deep learning and procedural animation coding wizard called Punches Bears (because of course), the demo will be featured as a demonstration of Unity 3D machine learning code in an update that's coming in a few weeks:
"They use curriculum learning to learn this behavior," Mr. Bears explains. In the first lesson, "The wall is low enough to jump over on their own. Learn where the goal is." In the second lesson, "The wall is too high. Learn that the block can be used to jump higher. It may take a while, but I think video game AI will transition from the rule-based behaviors we painstakingly build now into letting the AI teach themselves like in this vid. One day Mario may learn it's just simpler to settle down with a goomba...forget about the princess." (Read the whole Twitter thread here.)
Maybe you're skeptical someone named Punching Bears is capable of creating agents that learn, but he also created a procedural animation demo in which a buck-naked Danny Devito rides a giant bear, so I think that pretty much settles that.
Comments