Maulana, Azzam Wildan (2026) Autonomous Vehicle Navigation System using Machine Learning and Graph Based SLAM. Masters thesis, Institut Teknologi Sepuluh Nopember.
![[thumbnail of 6022241101-Master_Thesis.pdf]](http://repository.its.ac.id/style/images/fileicons/text.png) |
Text
6022241101-Master_Thesis.pdf Restricted to Repository staff only Download (27MB) | Request a copy |
Abstract
Icar is an autonomous vehicle developed by Institut Teknologi Sepuluh Nopember (ITS) to navigate the campus environment autonomously. Its previous navigation system relied on pose estimation through sensor fusion between odometry and the Global Navigation Satellite System (GNSS). However, GNSS signals are prone to degradation in environments with obstacles such as tall buildings and trees, while odometry suffers from wheel slippage, encoder drift, and inertial sensor errors. These limitations lead to unreliable localization, potentially causing Icar to deviate from its intended trajectory. This research proposes an enhanced navigation system by integrating a camera and LiDAR. Road detection is performed using a deep-learning-based semantic segmentation model, enabling Icar to distinguish road surfaces from non-road areas in real time. Pose refinement is achieved through graph-based optimization within a SLAM framework, incorporating depth camera and LiDAR data aligned using the Iterative Closest Point (ICP) algorithm. The Graph-Based SLAM significantly improves pose estimation accuracy, reducing the maximum error from 2.24 meters to 0.10 meters and improving precision from 0.69 meters to 0.03 meters. Furthermore, the Road Segmentation module achieves a best IoU of 0.7928, with temporal filtering implemented through GRU embedding and frame- by-frame low-pass filtering to enhance segmentation stability from 2.57% probablition of False Positive Hole detected to 1.26% probablition of False Positive Hole detected. Overall, the integration of SLAM and machine learning provides a more robust and GNSS independent navigation solution, improving Icar’s reliability when operating in semi-structured or dynamic environments.
================================================================================================================================
Icar adalah kendaraan otonom yang dikembangkan oleh Institut Teknologi Sepuluh Nopember (ITS) untuk menavigasi lingkungan kampus secara otonom. Sistem navigasi sebelumnya mengandalkan estimasi pose melalui fusi sensor antara odometri dan Sistem Satelit Navigasi Global (GNSS). Namun, sinyal GNSS rentan terhadap degradasi di lingkungan dengan hambatan seperti gedung tinggi dan pepohonan, sementara odometri mengalami selip roda, drift enkoder, dan kesalahan sensor inersia. Keterbatasan ini menyebabkan lokalisasi yang tidak dapat diandalkan, yang berpotensi menyebabkan Icar menyimpang dari lintasan yang dibuat sebelumnya. Penelitian ini mengusulkan sistem navigasi yang ditingkatkan dengan mengintegrasikan kamera dan LiDAR. Deteksi jalan dilakukan menggunakan model segmentasi semantik berbasis Deep Learning, memungkinkan Icar membedakan jalan dari area non-jalan secara real time. Koreksi posedidapat melalui optimisasi berbasis grafik dalam kerangka SLAM, yang menggabungkan data kamera dan LiDAR yang diselaraskan menggunakan algoritma Iterative Closest Point (ICP). SLAM Berbasis Grafik secara signifikan meningkatkan akurasi estimasi pose, mengurangi kesalahan maksimum yang awalnya 2,24 meter menjadi 0,10 meter dan meningkatkan presisi yang awalnya 0,69 meter menjadi 0,03 meter. Selain itu, modul Segmentasi Jalan mencapai IoU terbaik sebesar 0,7928, dengan filter temporal yang diimplementasikan melalui GRU embedding dan low-pass filter frame-by-frame untuk meningkatkan stabilitas segmentasi dari 2,57% kemungkinan False Positive Hole terdeteksi menjadi 1,26% kemungkinan False Positive Hole terdeteksi. Secara keseluruhan, integrasi SLAM dan Machine Learning memberikan solusi navigasi yang lebih tangguh dan independen dari GNSS, meningkatkan keandalan Icar saat beroperasi di lingkungan semi-terstruktur atau dinamis.
| Item Type: | Thesis (Masters) |
|---|---|
| Uncontrolled Keywords: | Autonomous vehicle, Icar, GPS, Odometry, Pose Estimation, Camera, Semantic Segmentation, SLAM, ICP, Graph-Based Optimization, Deep Learning, Kamera |
| Subjects: | T Technology > TA Engineering (General). Civil engineering (General) > TA1637 Image processing--Digital techniques. Image analysis--Data processing. T Technology > TJ Mechanical engineering and machinery > TJ211 Robotics. T Technology > TJ Mechanical engineering and machinery > TJ211.4 Robot motion T Technology > TJ Mechanical engineering and machinery > TJ211.415 Mobile robots T Technology > TJ Mechanical engineering and machinery > TJ213 Automatic control. |
| Divisions: | Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Electrical Engineering > 20101-(S2) Master Thesis |
| Depositing User: | Azzam Wildan Maulana |
| Date Deposited: | 22 Jan 2026 23:59 |
| Last Modified: | 22 Jan 2026 23:59 |
| URI: | http://repository.its.ac.id/id/eprint/130099 |
Available Versions of this Item
- Autonomous Vehicle Navigation System using Machine Learning and Graph Based SLAM. (deposited 22 Jan 2026 23:59) [Currently Displayed]
Actions (login required)
 |
View Item |