#Experiments with evolving simulated robots
Every experiment is a simulation of robot interacting with its environment.
A robot is controlled by a controller which may be any algorithm that transforms observation vectors into action vectors. Currently implemented controllers include Recurrent Neural Network and Linear Perceptron.
Environment also defines a fitness measure that outputs score at the end of simulation. This score could be used as a guide to evolve increasingly competent robot controllers with a blackbox optimization method such as Genetic Algorithm.
##Experiment 0: "avoid"
In this experiment a puck-shaped robot with 9 distance sensors tries to control its wheels' speeds to avoid driving over balls randomly spawned in its 10x10 area.
To run this experiment type node experiment.js
You can speed up the experiment by running it in clustered mode: node experiment.js cluster 3, tune the number to N-1 of cores available on your machine.