Neural network and Genetic Algorithms in Unity3D

Описание: Survival of the Fittest is a 3D racing game. It takes place on an endless track filled with obstacles. This game will be played by: a scripted AI, a player, and a learning neural network. When the game begins, every player starts positioned alongside of each other. They keep going forward avoiding obstacles in their way, they can collect yellow balls for extra points. Last player to crash wins.

The world takes a place on a never-ending track which sides are blocked off by walls leaving the only option to go forward. The world is made up of procedurally generated regions which contain obstacles (walls) and yellow balls, each of these are randomly generated. Some of the walls will be pre-set, but activated randomly.

Neural Network Setup:
• 7 inputs
o Ray cast in front of the AI.
o Ray cast 90 degrees on each side of the front.
o Ray cast 45 degrees on each side of the front.
o Ray cast 7.5 degrees on each side of the front.
o Each ray cast returns a reversed distance of an object it has hit.
• 2 outputs
o Steer Left.
o Steer Right.
o Each output (between 0-1) multiplied by Maximum Rotation.


Genetic Algorithm:
• Mutation rate 0.1.
• Crossover rate 1.
• Perpetuation is set to 0.3.
• Uses Elitism selection function.
• Keeps 2 alphas from previous generation.
• Picks the best 3 genomes used for mating.
• Total population is 11.


Github link:
https://github.com/StartkaCZ/Neural-N...

Research Documents Link:
https://drive.google.com/open?id=0Bw8...