Generativne neuronske mreže za sintezu robotskih putanja

master's thesis

• Title (english): Generative neural networks for synthesis of robot paths
• Author: Tin Janković
• Mentor: Petar Ćurković (mentor)
• Granter: University of Zagreb; Faculty of Mechanical Engineering and Naval Architecture; Zagreb
• Defense date and country: 2022-07-21, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing; Artificial Intelligence

Abstract

U ovom radu proučavat će se mogućnost računalnog generiranja sadržaja pomoću Generativnih
kontradiktornih neuronskih mreža (engl. GAN) s ciljem poboljšanja navigiranja robota. Mreža
je vrlo složene strukture jer sadrži dvije manje neuronske mreže koje se međusobno
sukobljavaju. Prva manja mreža je generator koji želi kreirati takav sadržaj koji druga mreža
diskriminator neće moći razlikovati od stvarnog seta podataka.
U prvom dijelu rada objasnit će se osnove rada GAN mreža i objasniti njihova struktura. Zatim
će se opisati korišteni alati za razvoj mreže i upoznati sa ostalim potencijalnim primjenama
GAN mreža koje mogu olakšati svakodnevicu čovjeka. Predstavit će se jedno od mogućih
rješenja GAN mreže za generiranje trajektorija i navesti dobiveni rezultati te će se na kraju
primijeniti generirane trajektorije za vođenje simulacijskog modela robota.

Abstract (english)

In this paper, the possibility of computer-generated content generation using Generative adversarial neural networks (GAN) is investigated with the aim of improving robot navigation. The network has a very complex structure as it contains two smaller neural networks that collide with each other. The first smaller network is a generator that wants to produce such a content that the second network discriminator is not able to distinguish from the actual data set.
The first part of the paper will explain the basics of how GAN networks work and explains their structure. Then, the tools used for network development are described and other possible applications of GAN networks that can facilitate everyday life will be introduced. One of the possible solutions of the GAN network for trajectory generation is presented and the obtained results will be stated, and finally the generated trajectories are used to control the robot simulation model.

• Keywords: Generativne kontradiktorne mreže; GAN; CGAN; strojno učenje; neuronske mreže.
• Keywords (english): Generative adversarial networks; GAN; CGAN; machine learning; neural networks
• Language: croatian
• URN:NBN: urn:nbn:hr:235:947606
• Study programme: Title: Mechanical Engineering; Study programme type: university; Study level: graduate; Academic / professional title: magistar/magistra inženjer/inženjerka strojarstva (magistar/magistra inženjer/inženjerka strojarstva)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Open access
• Created on: 2022-07-05 23:10:43