Programmer. Hiker. Cook. Always looking for interesting problems to solve.
I’ve been trying to perfect e-puck’s line following ability and I think I’ve succeeded:
Behaviour was evolved using feedforward neural network and genetic algorithm.
Summary of parameters:
Inputs: 3 (Floor colour sensors)
Outputs: 2 (motor actions)
Number of hidden layers: 1
Number of neurones in a hidden layer: 12
Population size: 30
Selection: Roulette wheel.
Crossover probability: 50%.
Mutation probability: 10%.
Quite good behaviour was evolved after just 10 generations. I’m pretty amazed with how well NN can be trained!