نبذة مختصرة : Cilj ovog završnog rada je pokazati mogućnosti koje donose strojno i duboko učenje kada ih se kombinira s moćnim alatima za stvaranje virtualnih simulacija. Ovo je omogućeno kroz Unity game engine koji se koristi za izradu simulacija i razvoj 2D i 3D video igara te ML-Agents dodatak koji sadrži algoritme za treniranje inteligentnih agenata. Opisani su osnovni principi i algoritmi dubokog učenja uz pomoć kojih agent percipira okolinu oko sebe i uči rješavati određeni zadatak. U sklopu rada izrađena je simulacija u kojoj agent pokušava pronaći skriveni predmet u određenom vremenu koji se postavi na novo nasumično generirano mjesto nakon pronalaska. Agent uči korištenjem strojnog učenja s potporom. Primjenjujući odgovarajuću implementaciju i parametre agent je uspješno naučio svladati dani problem. ; Purpose of this bachelor’s thesis is to show possibilities of machine learning and deep learning in combination with powerful tools which are used in making virtual simulations. This is enabled through Unity game engine, a software which is used for developing 2D and 3D video games and simulations, and ML-agents plugin which contains algorithms for training intelligent agents. Basic principles and algorithms of deep learning are described. Agent uses those methods for perception of environment and solving certain tasks. Within thesis, simulation is created in which agent is trying to find hidden item by using reinforcement machine learning. If agent finds item, new place for item is randomly generated. Also, he must complete task in certain time. With the right implementation and parameters, agent solved the task successfully.
Processing Request
No Comments.