prikaz prve stranice dokumenta Razvoj algoritma autonomne vožnje zasnovanog na procjeni kuta zakreta upravljača vozila i testiranje performansi algoritma u različitim simulatorima
  Download
PDF 15.94 MB
master's thesis
Razvoj algoritma autonomne vožnje zasnovanog na procjeni kuta zakreta upravljača vozila i testiranje performansi algoritma u različitim simulatorima
2022. urn:nbn:hr:200:594925
Benja, Borna
Josip Juraj Strossmayer University of Osijek
Faculty of Electrical Engineering, Computer Science and Information Technology Osijek
Department of Communications
Chair of Multimedia Systems and Digital Television

Cite this document

Benja, B. (2022). Razvoj algoritma autonomne vožnje zasnovanog na procjeni kuta zakreta upravljača vozila i testiranje performansi algoritma u različitim simulatorima (Master's thesis). Retrieved from https://urn.nsk.hr/urn:nbn:hr:200:594925

Benja, Borna. "Razvoj algoritma autonomne vožnje zasnovanog na procjeni kuta zakreta upravljača vozila i testiranje performansi algoritma u različitim simulatorima." Master's thesis, Josip Juraj Strossmayer University of Osijek, Faculty of Electrical Engineering, Computer Science and Information Technology Osijek, 2022. https://urn.nsk.hr/urn:nbn:hr:200:594925

Benja, Borna. "Razvoj algoritma autonomne vožnje zasnovanog na procjeni kuta zakreta upravljača vozila i testiranje performansi algoritma u različitim simulatorima." Master's thesis, Josip Juraj Strossmayer University of Osijek, Faculty of Electrical Engineering, Computer Science and Information Technology Osijek, 2022. https://urn.nsk.hr/urn:nbn:hr:200:594925

Benja, B. (2022). 'Razvoj algoritma autonomne vožnje zasnovanog na procjeni kuta zakreta upravljača vozila i testiranje performansi algoritma u različitim simulatorima', Master's thesis, Josip Juraj Strossmayer University of Osijek, Faculty of Electrical Engineering, Computer Science and Information Technology Osijek, accessed 24 March 2023, https://urn.nsk.hr/urn:nbn:hr:200:594925

Benja B. Razvoj algoritma autonomne vožnje zasnovanog na procjeni kuta zakreta upravljača vozila i testiranje performansi algoritma u različitim simulatorima [Master's thesis]. Osijek: Josip Juraj Strossmayer University of Osijek, Faculty of Electrical Engineering, Computer Science and Information Technology Osijek; 2022 [cited 2023 March 24] Available at: https://urn.nsk.hr/urn:nbn:hr:200:594925

B. Benja, "Razvoj algoritma autonomne vožnje zasnovanog na procjeni kuta zakreta upravljača vozila i testiranje performansi algoritma u različitim simulatorima", Master's thesis, Josip Juraj Strossmayer University of Osijek, Faculty of Electrical Engineering, Computer Science and Information Technology Osijek, Osijek, 2022. Available at: https://urn.nsk.hr/urn:nbn:hr:200:594925

Please login to the repository to save this object to your list.
Metadata
TitleRazvoj algoritma autonomne vožnje zasnovanog na procjeni kuta zakreta upravljača vozila i testiranje performansi algoritma u različitim simulatorima
Title (english)Development of the autonomous driving algorithm based on vehicle steering angle prediction and algorithm performance testing in different simulators
AuthorBorna Benja
MentorMario Vranješ (mentor)
Committee memberMario Vranješ (predsjednik povjerenstva)
Committee memberRatko Grbić (član povjerenstva)
Committee memberMarijan Herceg (član povjerenstva)
GranterJosip Juraj Strossmayer University of Osijek
Faculty of Electrical Engineering, Computer Science and Information Technology Osijek
(Department of Communications)
(Chair of Multimedia Systems and Digital Television)
Osijek
Defense date and country2022-09-08, Croatia
Scientific / art field,
discipline and subdiscipline		TECHNICAL SCIENCES
Computing
Artificial Intelligence
Abstract
U ovom radu dan je pregled postoje´cih algoritama za autonomnu voˇznju zasnovanih na procjeni kuta zakreta upravljaˇca vozila. Izradeni su novi vlastiti algoritmi za autonomnu voˇznju zasnovani na Swin transformerskoj neuronskoj mreˇzi. Implementirane su 4 neuronske mreˇze (Swin1, Swin2, Swin3 i Swin4) koje se razlikuju po broju potpuno povezanih slojeva i broju neurona u pojedinom sloju. Implementirane neuronske mreˇze trenirane su i testirane na skupovima podataka prikupljenim u CARLA i MetaDrive simulatorima, a dobiveni rezultati pribliˇzno su jednaki rezultatima PilotNet neuronske mreˇze i vlastite LSTM neuronske mreˇze. Takoder, promatrano je i vrijeme inferencije tj. brzina obrade podataka razliˇcitih neuronskih mreˇza. Najve´cu brzinu obrade postigla je Swin1 neuronska mreˇza s 16.04 FPS. Uz testiranje na prikupljenim podacima, modeli neuronskih mreˇza testirani su u virtualnim okruˇzenjima obaju simulatora te su postigli visoke performanse. U CARLA simulatoru, Swin2 model ostvario je bolje rezultate nego algoritmi iz literature na ve´cini testova. U MetaDrive simulatoru, Swin2 model ostvario je pribliˇzno jednake rezultate kao i PilotNet neuronska mreˇza. Modeli neuronskih mreˇza koji su postigli najbolje rezultate dotrenirani su na trening skupu podataka iz drugog simulatora metodom prijenosnog uˇcenja, no dobiveni rezultati nisu bili znaˇcajno bolji u odnosu na rezultate dobivene treniranjem modela na svakom skupu zasebno. Zakljuˇceno je da neuronske mreˇze zasnovane na Swin transformerskoj neuronskoj mreˇzi mogu biti osnova algoritma autonomne voˇznje zasnovanog na procjeni kuta zakreta upravljaˇca vozila uz dovoljno velik i raznovrstan skup podataka za treniranje.
Abstract (english)
This paper provides an overview of the existing autonomous driving algorithms based on steering angle prediction. Proprietary autonomous driving algorithms based on the Swin transformer neural network were developed. Four neural network models were implemented (Swin1, Swin2, Swin3 and Swin4) differing in the number of fully connected layers and the number of neurons in each layer. The implemented neural networks were trained and tested on datasets collected in CARLA and MetaDrive simulators, and the obtained results are comparable to the results of the PilotNet neural network and the proprietary LSTM neural network. Also, the inference time i.e. the speed of data processing of different neural networks was observed. The highest processing speed was achieved by the Swin1 neural network with 16.04 FPS. In addition to testing on collected datasets, neural network models were tested in the virtual environments of both simulators and achieved high performance. In CARLA simulator, the Swin2 model achieved better results than contemporary algorithms on most tests. In MetaDrive simulator, the Swin2 model achieved results comparable to those of the PilotNet neural network. The models that achieved the best results were trained on the training dataset from another simulator using the transfer learning method but the results obtained were not significantly better than the results obtained by training neural network models on each dataset separately. It was concluded that neural networks based on the Swin transformer neural network can provide a platform for an autonomous driving algorithm based on steering angle prediction given a sufficiently large and diverse training dataset.
Keywords
Keywords (english)
Languagecroatian
URN:NBNurn:nbn:hr:200:594925
Study programmeTitle: Graduate University Study Programme in Computer Engineering
Study programme type: university
Study level: graduate
Academic / professional title: magistar/magistra inženjer/inženjerka računarstva (magistar/magistra inženjer/inženjerka računarstva)
Type of resourceText
File originBorn digital
Access conditionsOpen access
Terms of useLicence In copyright icon
Created on2022-10-19 07:59:28