Evolutionary optimization of mobile robot neural network

Razvoj umjetne inteligencije omogućava otkrivanje novih metoda za rješavanje dosadašnjih optimizacijskih problema. Primjena evolucijskih mehanizama u razvoju novih rješenja za razne probleme optimizacije oduvijek je bila zanimljiva tema kojom se bave i robotika i računarstvo. Cilj rada bio je pobliže proučiti područje evolucijske robotike te implementirati evolucijski algoritam koji optimira rad neuronske mreže u virtualnom okruženju na problemima praćenja ravne linije i slijeđenja izvora svjetlosti. Ovaj rad ispitao je kako se evolucijskim algoritmom evoluira neuronska mreža, koja služi za upravljanje Braitenbergovog modela mobilnog robota. Objašnjeni su glavni mehanizmi evolucije i princip rada neuronskih mreža. Izvršena je implementacija pomoću Neuro Evolution of Augmenting Topologies (NEAT) algoritma u programskom jeziku python, koji je korišten za rješavanje problema slijeđenja ravne linije i praćenja izvora svjetlosti. The development of artificial intelligence enables the discovery of new methods for solving existing problems. The application of evolutionary mechanisms in the development of new solutions to various problems has always been an interesting topic to both robotics and computer science. The aim of this study was to examine the field of evolutionary robotics in more detail, and to implement an evolutionary algorithm that optimizes the operation of a neural network in a virtual environment on the problems of tracking a straight line, and following a light source. This study investigated how the evolutionary algorithm evolves a neural network which is used to control Braitenberg's model of a mobile robot. The main mechanisms of evolution and the working principle of neural networks were explained. Implementation was performed using the Neuro Evolution of Augmenting Topologies (NEAT) algorithm in the python programming language which was used to solve the problems of following a straight line and tracking a light source.