prikaz prve stranice dokumenta Izgradnja algoritama za upravljanje vozilom u simulatoru primjenom podržanog učenja
  Download
PDF 2.21 MB
master's thesis
Izgradnja algoritama za upravljanje vozilom u simulatoru primjenom podržanog učenja
Osijek: Josip Juraj Strossmayer University of Osijek, Faculty of Electrical Engineering, Computer Science and Information Technology Osijek, 2023. urn:nbn:hr:200:732951
Činčurak, Dorian
Josip Juraj Strossmayer University of Osijek
Faculty of Electrical Engineering, Computer Science and Information Technology Osijek
Department of Computer Engineering and Automation
Chair of Automation and Robotics

Institutional repository: Faculty of Electrical Engineering, Computer Science and Information Technology Osijek

Cite this document

Činčurak, D. (2023). Izgradnja algoritama za upravljanje vozilom u simulatoru primjenom podržanog učenja (Master's thesis). Osijek: Josip Juraj Strossmayer University of Osijek, Faculty of Electrical Engineering, Computer Science and Information Technology Osijek. Retrieved from https://urn.nsk.hr/urn:nbn:hr:200:732951

Činčurak, Dorian. "Izgradnja algoritama za upravljanje vozilom u simulatoru primjenom podržanog učenja." Master's thesis, Josip Juraj Strossmayer University of Osijek, Faculty of Electrical Engineering, Computer Science and Information Technology Osijek, 2023. https://urn.nsk.hr/urn:nbn:hr:200:732951

Činčurak, Dorian. "Izgradnja algoritama za upravljanje vozilom u simulatoru primjenom podržanog učenja." Master's thesis, Josip Juraj Strossmayer University of Osijek, Faculty of Electrical Engineering, Computer Science and Information Technology Osijek, 2023. https://urn.nsk.hr/urn:nbn:hr:200:732951

Činčurak, D. (2023). 'Izgradnja algoritama za upravljanje vozilom u simulatoru primjenom podržanog učenja', Master's thesis, Josip Juraj Strossmayer University of Osijek, Faculty of Electrical Engineering, Computer Science and Information Technology Osijek, accessed 05 November 2024, https://urn.nsk.hr/urn:nbn:hr:200:732951

Činčurak D. Izgradnja algoritama za upravljanje vozilom u simulatoru primjenom podržanog učenja [Master's thesis]. Osijek: Josip Juraj Strossmayer University of Osijek, Faculty of Electrical Engineering, Computer Science and Information Technology Osijek; 2023 [cited 2024 November 05] Available at: https://urn.nsk.hr/urn:nbn:hr:200:732951

D. Činčurak, "Izgradnja algoritama za upravljanje vozilom u simulatoru primjenom podržanog učenja", Master's thesis, Josip Juraj Strossmayer University of Osijek, Faculty of Electrical Engineering, Computer Science and Information Technology Osijek, Osijek, 2023. Available at: https://urn.nsk.hr/urn:nbn:hr:200:732951

Metadata
TitleIzgradnja algoritama za upravljanje vozilom u simulatoru primjenom podržanog učenja
Title (english)Algorithms for driving a vehicle in a simulator using reinforcement learning
AuthorDorian Činčurak
MentorRatko Grbić (mentor)
Committee memberRatko Grbić (predsjednik povjerenstva)
Committee memberMarijan Herceg (član povjerenstva)
Committee memberMario Vranješ (član povjerenstva)
GranterJosip Juraj Strossmayer University of Osijek
Faculty of Electrical Engineering, Computer Science and Information Technology Osijek
(Department of Computer Engineering and Automation)
(Chair of Automation and Robotics)
Osijek
Defense date and country2023-09-25, Croatia
Scientific / art field,
discipline and subdiscipline		TECHNICAL SCIENCES
Computing
Artificial Intelligence
Abstract
U radu je detaljnije opisano podržano učenje i određeni postojeći algoritmi podržanog učenja temeljeni na dubokom učenju. Ideja rada je primijeniti opisane algoritme na problem upravljanja autonomnim vozilima u prometu unutar simulatora vožnje. U okviru ovog rada su izrađena dva rješenja za autonomno upavljanje vozilom unutar simulatora: prvo je zasnovano na algoritmu dubokog Q-učenja (DQN), a drugo na algoritmu dubokog determinističkog gradijenta politike (DDPG). Agent dubokog učenja šalje brojčane vrijednosti za gas i zakret volana preko CARLA Python API-a do CARLA simulatora koji zatim primijenjuje primljene podatake na vozilo u simulatoru. Pritisak papučice gasa se održava u specifičnim granicama, a neuronske mreže predviđaju kut zakreta volana iz početne slike dobivene preko semantičke kamere postavljene na prednju stranu vozila. Vlastiti algoritmi su izrađeni u Python programskom jeziku, te su treniranja i testiranja provedena unutar CARLA simulatora na različitim scenarijima vožnje. Evaluacija izgrađenih algoritama provedena je pomoću predloženih metrika: prosječno prijeđen put, prosječno ukupno vrijeme koje je vozilo provelo vozeći kroz epizodu, prosječna ukupna kumulativna nagrada dobivena kroz epizodu, prosječan broj pogrešaka koje agent učini tijekom jedne epizode, prosječno prijeđen put i prosječno vrijeme provedeno u susjednim trakama kada ta akcija nije optimalna.
Abstract (english)
This master’s thesis describes reinforcement learning in more detail and certain existing reinforcement learning algorithms based on deep learning. The idea of this thesis is to apply the described algorithms to the problem of autonomous vehicle control in traffic within a driving simulator. As part of this thesis, two solutions were developed for autonomous vehicle driving within the simulator: one based on the Deep Q-Learning (DQN) algorithm and the other on the Deep Deterministic Policy Gradient (DDPG) algorithm. The deep learning agent sends numerical values for throttle and steering through the CARLA Python API to the CARLA simulator, which then applies the received data to the vehicle in the simulator. The amount of throttle is maintained within specific limits, and neural networks predict the angle of rotation of the steering wheel from the initial image obtained through the semantic camera placed on the front of the vehicle. Algorithms were developed in the Python programming language, and training and testing were carried out in the CARLA simulator on different driving scenarios. The evaluation of the developed algorithms was carried out using the proposed metrics: the average distance traveled, the average total time the vehicle spent driving throughout the episode, the average total cumulative reward obtained throughout the episode, the average number of errors made by the agent during one episode, the average distance traveled and the average time spent in adjacent lanes when that action is not optimal.
Keywords
Keywords (english)
Languagecroatian
URN:NBNurn:nbn:hr:200:732951
Study programmeTitle: UNIVERSITY STUDY AUTOMOBILE COMPUTER AND COMMUNICATION
Study programme type: university
Study level: graduate
Academic / professional title: magistar/magistra inženjer/inženjerka automobilskog računarstva i komunikacija (magistar/magistra inženjer/inženjerka automobilskog računarstva i komunikacija)
Type of resourceText
File originBorn digital
Access conditionsOpen access
Terms of useLicence In copyright icon
Created on2023-10-10 11:03:10