Perencanaan Gerakan pada Mobil Otonom di Jalan Raya Menggunakan Quantile Regression Deep Q Network

Putra, Rifky Yulianto (2021) Perencanaan Gerakan pada Mobil Otonom di Jalan Raya Menggunakan Quantile Regression Deep Q Network. Undergraduate thesis, Institut Teknologi Sepuluh Nopember.

[img] Text
07111740000034-Undergraduate_Thesis.pdf - Accepted Version
Restricted to Repository staff only until 1 April 2023.

Download (2MB) | Request a copy

Abstract

Dalam beberapa dekade terakhir, mobil otonom telah menjadi salah satu topik yang banyak dipelajari dan dikembangkan oleh para akademisi dan praktisi industri. Untuk memberikan kenyamanan dan keamanan dalam berkendara, perencanaan gerakan pada mobil otonom memiliki peranan penting dalam menghasilkan navigasi, gerakan, dan manuver yang mulus serta bebas hambatan menuju tempat yang dituju. Pada penelitian ini, perencanaan gerakan mobil otonom dirancang menggunakan algoritma Quantile Regression Deep Q Network (QR-DQN). Pembelajaran dan pengujian sistem perencanaan gerakan mobil otonom yang telah dirancang dilakukan dengan menggunakan Unity ML-Agents Highway Simulator. Simulator ini mengakomodasi kamera RGB dan sensor LIDAR sebagai masukan dari sistem perencanaan gerakan mobil otonom. Hasil pengujian menunjukkan bahwa sistem perencanaan gerakan mobil otonom dengan algoritma QR-DQN yang dirancang mampu melakukan akselerasi, menjaga jarak dengan kendaraan, dan melewati kendaraan lain dengan baik dengan tingkat keberhasilan gerakan mencapai 91.13%. Pada pengujian dengan noise, sistem perencanaan gerakan mobil otonom juga masih mampu berjalan dengan baik dengan tingkat keberhasilan gerakan sekitar 89.47% - 90.17%, meski terdapat sedikit penurunan dalam hal kecepatan dan jumlah kendaraan yang dilewati pada beberapa skenario berkendara. ==================================================================================================== In recent decades, autonomous cars have become one of the most widely studied and developed topics by academics and industry practitioners. To provide comfort and safety in driving, motion planning of autonomous cars has an important role by generating a smooth, dynamically-feasible, and collision-free trajectory towards the destination. In this study, the autonomous car motion planning system was proposed using Quantile Regression Deep Q Network (QR-DQN) algorithm. The training and testing stage of the proposed autonomous car motion planning system is simulated using Unity ML-Agents Highway Simulator. This simulator provides an RGB camera and a LIDAR sensor as input for the autonomous car motion planning system. The testing results show that the proposed autonomous car motion planning system with the QR-DQN algorithm can accelerate, maintain distance from front vehicles, and overtake other vehicles well with 91.13% of successfully generated action. In testing with noise, the autonomous car motion planning system was also able to run well with 89.47% - 90.17% of successfully generated actions, although there was a slight decrease in speed and the number of overtaken vehicles in several driving scenarios.

Item Type: Thesis (Undergraduate)
Uncontrolled Keywords: Mobil Otonom, Perencanaan Gerakan, Autonomous Car, Motion Planning, Quantile Regression Deep Q Network, Unity ML-Agents Highway Simulator
Subjects: Q Science > QA Mathematics > QA336 Artificial Intelligence
Q Science > QA Mathematics > QA9.58 Algorithms
T Technology > TL Motor vehicles. Aeronautics. Astronautics > TL152.8 Vehicles, Remotely piloted. Autonomous vehicles.
Divisions: Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Electrical Engineering > 20201-(S1) Undergraduate Thesis
Depositing User: Rifky Yulianto Putra
Date Deposited: 24 Feb 2021 06:31
Last Modified: 24 Feb 2021 06:31
URI: https://repository.its.ac.id/id/eprint/82739

Actions (login required)

View Item View Item