Solving Mazes using an Artificial Developmental Neuron
Gul Muhammad Khan, Julian Francis Miller · 2010
An agent controlled by a single computational neuron is used to solve maze problems. The neuron has activity and time-dependent computational and topological structure. The be-haviour of a neuron is controlled by a collection of seven evolved programs that are loosely analogous to aspects of bi-ological neuron (dendrites, soma, axons, synapses, electrical and developmental behaviour). The programs are represented using Cartesian Genetic Programming. Our aim is to show that it is possible to evolve programs that develop a single neuron so that it is able to learn how to solve maze problems purely by experience.