ARSITEKTUR MULTITASK LEARNING BERBASIS TRANSFORMER UNTUK PENINGKATAN AKURASI KENDARAAN OTONOM

Fachri Najm Noer Kartiman, - and Rasim, - and Yaya Wihardi, - (2025) ARSITEKTUR MULTITASK LEARNING BERBASIS TRANSFORMER UNTUK PENINGKATAN AKURASI KENDARAAN OTONOM. S1 thesis, Universitas Pendidikan Indonesia.

Abstract

Perkembangan teknologi kendaraan otonom menuntut sistem perencanaan jalur yang mampu memahami lingkungan secara menyeluruh dan menghasilkan keputusan mengemudi yang akurat dan aman. Model perencanaan tradisional berbasis CNN seringkali memiliki keterbatasan dalam menangkap konteks global dan hubungan antar elemen lingkungan secara efektif. Untuk mengatasi hal tersebut, penelitian ini mengusulkan arsitektur SKGE-Swin yang memanfaatkan Swin Transformer dengan mekanisme skip stage guna memperkuat representasi fitur di berbagai level jaringan. Pendekatan ini memungkinkan model untuk mempertahankan informasi penting dari tahap awal hingga akhir proses ekstraksi fitur, sehingga meningkatkan kemampuan dalam memahami pola kompleks di sekitar kendaraan. Hasil eksperimen menunjukkan bahwa arsitektur SKGE-Swin mampu memberikan performa yang lebih baik dengan Driving Score sebesar 37.10 dalam tugas prediksi waypoint dibandingkan dengan metode CNN, sehingga berkontribusi pada pengembangan sistem kendaraan otonom yang lebih handal dan aman. The advancement of autonomous vehicle technology demands a path planning system capable of comprehensively understanding the environment and generating accurate and safe driving decisions. Traditional CNN-based planning models often face limitations in effectively capturing global context and relationships between environmental elements. To address this issue, this study proposes the SKGE-Swin architecture, which leverages the Swin Transformer with a skip-stage mechanism to enhance feature representation across different levels of the network. This approach enables the model to retain crucial information from the early to the final stages of feature extraction, thereby improving its ability to understand complex patterns surrounding the vehicle. Experimental results demonstrate that the SKGE-Swin architecture achieves superior performance, attaining a Driving Score of 37.10 in waypoint prediction tasks, outperforming CNN-based methods. This contributes to the development of more reliable and safer autonomous driving systems.

	Text S_KOM_2106515_Title.pdf Download (629kB)
	Text S_KOM_2106515_Chapter1.pdf Download (512kB)
	Text S_KOM_2106515_Chapter2.pdf Restricted to Staf Perpustakaan Download (895kB)
	Text S_KOM_2106515_Chapter3.pdf Download (656kB)
	Text S_KOM_2106515_Chapter4.pdf Restricted to Staf Perpustakaan Download (12MB)
	Text S_KOM_2106515_Chapter5.pdf Download (67kB)

Item Type:	Thesis (S1)
Additional Information:	https://scholar.google.com/citations?hl=en&user=CzzuCxgAAAAJ SINTA ID Dosen Pembimbing: 5990962 RASIM 5994413 YAYA WIHARDI
Uncontrolled Keywords:	Multitask learning, transformer, Kendaraan otonom, perencanaan jalur, Swin Transformer, skip stage, prediksi waypoint, representasi fitur, model deep learning. Multitask Learning, Transformer, Autonomous Vehicle, Path Planning, Swin Transformer, Skip Stage, Waypoint Prediction, Feature Representation, Deep Learning Model.
Subjects:	Q Science > Q Science (General) Q Science > QA Mathematics Q Science > QA Mathematics > QA75 Electronic computers. Computer science Q Science > QA Mathematics > QA76 Computer software T Technology > T Technology (General) T Technology > TL Motor vehicles. Aeronautics. Astronautics
Divisions:	Fakultas Pendidikan Matematika dan Ilmu Pengetahuan Alam > Program Studi Ilmu Komputer
Depositing User:	Fachri Najm Noer Kartiman
Date Deposited:	30 Jul 2025 08:41
Last Modified:	30 Jul 2025 08:41
URI:	http://repository.upi.edu/id/eprint/134801

Actions (login required)

View Item