The Neural Network is learning how to pilot a drone and the goal is to stay alive for as long as possible (stays within the screen boundaries).
Using genetic algorithm to alter neural network weights after some generations (kind of training?) we obtain a decent drone pilot. For example, under a specific configuration, genetic algorithm discoverd a set of weights (NN) that could pilot the drone for more than 90 seconds after only 10 mins of searching, piloting.
This is my way of implementing drone physics, movements, forces, neural networks (weights, links and forward pass), genetic algorithm for searching an optimal set of weights (NN).
Feel free to check out the code, fork the repo and own it.