creatures avoiding planks - 大トロ

creatures avoiding planks – 大トロ

  Recently I came across a video of a simulation that demonstrates the use of evolutionary techniques to train agents to avoid moving obstacles.  The methodology used seems to employ a variation of the NEAT algorithm used to evolve the topology of neural networks so that it can perform certain tasks correctly.  It was written to be used in the Unity 5 game engine to be incorporated to general game AIs which interested me. The results are really cool and the dynamics seemed quite elegant, so I wanted to try to implement a similar demo in javascript that can run inside a browser session.  After playing around with a basic implementation, I found out that for obstacle avoiding problems, it turns out that a very simple neural network can be used effectively, and that even a fully connected neural network can be overkill.  I also realised in the end that that even a simple perceptron -like network works quite well,

Read full article from creatures avoiding planks – 大トロ