Рет қаралды 39,911
Using UE4 blueprints, machine learning and genetic algorithms, I created a car which learns to drive a track.
It uses a neural network with 6 input nodes, 2 hidden layers with 6 nodes each, and 2 output nodes. Each hidden node and output node also has a bias.
A population of 19 cars run simultaneously (but do not interact with each other). After each round, the top 5 cars are chosen based on their fitness score. Their neural networks are mutated randomly by 10%, and they form the next population.