• Mateusz Mazurkiewicz

Neuroevolution for control

A triangle is to follow a point in two dimensional space, having control of its second derivative of rotation and position (angular and forward acceleration). Air resistance applies. This simple control theory problem is solved with neuroevolution.

A neural network with 6 inputs and 4 neurons in the hidden layer is controlling the two. To arrive at neural weights, 600 triangles were simulated, with the worst-performing ones being deleted and replaced by randomly changed (mutated) copies of best-performing ones.