The use of artificial intelligence for flying a drone

Ovaj rad bavi se problematikom autonomne bespilotne letjelice. Na početku opisujemo vrstu letjelice koju ćemo koristiti u implementaciji I način na koji ista leti. Zatim su ukratko opisane umjetne neuronske mreže i potporno učenje, dva pristupa umjetne inteligencije ključna za razumjevanje dubokog potpornog učenja. Opisujemo PPO, metodu dubokog potpornog učenja, odabranu za učenje željenog ponašanja naše letjelice. U Unity Editor-u izrađujemo model letjelice i pomoću ML- Agents alata pripremamo okolinu za učenje. Prevodimo naš problem u sustav nagrada kojima letjelicu navodimo na željeno ponašanje. Nakon završenog učenja grafički prikazujemo dobivene rezultate.The topic of this essay is the problematics of autonomous unmanned aircrafts. In the beginning, we described a type of aircraft that is used in implementation, along with its manner of flying. What follows is a brief description of artificial neural networks and reinforced learning, two approaches in artificial intelligence that are key to understanding deep reinforced learning. We also described PPO, a method of deep reinforced learning, chosen for the purpose of teaching our aircraft a desired behavior. In the Unity Editor, we made a model of the aircraft, and with the ML-Agents tool we prepared an environment for learning. We translated our problem into a reward system, which is used to lead the aircraft onto desired behavior. After the learning process is done, the obtained results are displayed graphically