It randomly generates 16 neural networks at the beginning... After each generation, the ones that move furthest from the center get to populate the next generation along with mutants and breeds... The cycle continues...
The neural network has 6 input neurons (1 for the elevation of each of the hips, 1 const, and 1 for the sin(timer*const)) and has a hidden layer of 8 neurons and then the output layer controls each of the 8 muscles attached to the legs. The furthest I've gotten it is 29.7 units for the fitness (the amount it has traveled from the center) after 20 generations. https://scratch.mit.edu/projects/179164859/
