Implementasi Convolutional Neural Network Untuk Sistem Kendali Kemudi Menggunakan Donkeycar Simulator

Kuncahyojati, Anindriyo (2023) Implementasi Convolutional Neural Network Untuk Sistem Kendali Kemudi Menggunakan Donkeycar Simulator. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5009201077-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only until 1 April 2025. Download (2MB) | Request a copy

Abstract

Meningkatnya jumlah kendaraan semakin meningkat. Kecelakaan lalu lintas dapat disebabkan oleh faktor kesalahan manusia, faktor pengemudi, faktor jalan, faktor kendaraan bermotor, dan faktor alam. Oleh karena itu perlu adanya solusi teknologi seperti Autonomous Car untuk meningkatkan keamanan dalam lalu lintas kendaraan serta mampu mengurangi kemacetan dan memberikan faktor kenyamanan pada penumpang berkendara dengan berbasis AI. Metode yang digunakan yaitu Behavior Cloning menggunakan Convolutional Neural Networks (CNN) dengan merujuk pada arsitektur yang dikembangkan oleh Nvidia untuk Autonomous Drivin Car dengan Donkeycar Simulator. Metode ini bekerja dengan cara melajukan kendaraan secara manual kemudian direkam pergerakannya untuk dijadikan dataset yang berupa Gambar dan rekaman sudut steer. Gambar tersebut dimasukkan ke dalam CNN yang kemudian diolah untuk mendapatkan perintah kemudi yang diusulkan (θcmd). Perintah yang diusulkan dibandingkan dengan perintah yang diinginkan berdasarkan rekaman sudut steer (θm), dan bobot CNN disesuaikan untuk mendekatkan keluaran CNN ke keluaran yang diinginkan. Penyesuaian bobot dilakukan menggunakan backpropagation. Hasil yang didapatkan yaitu nilai variable loss 0,1, total nilai kesalahan lateral 3,45% dan nilai Mean Square Error (MSE) hasil dari perbandingan antara truth label dan prediksi model adalah 0.041. Behavioral cloning dengan menggunakan CNN menunjukkan sudah dapat mengikuti lajur tetapi masih memiliki kekurangan. Hal ini berdasarkan pengujian kesalahan lateral mobil masih mengalami keluar jalur tetapi mampu me-recovery apabila mobil sudah melewati garis batas.
===================================================================================================
The number of vehicles is increasing. Traffic accidents can be caused by human error factors, driver factors, road factors, motor vehicle factors, and natural factors. Therefore, there is a need for technological solutions such as Autonomous Car to improve safety in vehicle traffic and be able to reduce congestion and provide a comfort factor for driving passengers based on AI. The method used is Behavior Cloning Using Convolutional Neural Networks (CNN) with reference to the architecture developed by Nvidia for Autonomous Driving Car with Donkeycar Simulator. This method works by manually driving the vehicle and then recording its movement to be used as a dataset in the form of images and steer angle recordings. The image is fed into a CNN which is then processed to get the proposed steering command (θcmd). The proposed command is compared with the desired command based on the recorded steer angle (θm), and the CNN weights are adjusted to bring the CNN output closer to the desired output. The weight adjustment is performed using backpropagation. The results obtained are the variable loss value of 0.1, the total lateral error value of 3.45% and the Mean Square Error (MSE) value resulting from the comparison between the truth label and the model prediction is 0.041. Behavioral cloning using CNN shows that it can follow the lane but still has shortcomings. This is based on lateral error testing, the car still gets out of the lane but is able to recover when the car has crossed the boundary line.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	behavior clonning, variable loss, mean square error, kesalahan lateral
Subjects:	Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science) T Technology > TL Motor vehicles. Aeronautics. Astronautics > TL152.8 Vehicles, Remotely piloted. Autonomous vehicles.
Divisions:	Faculty of Industrial Technology > Physics Engineering > 30201-(S1) Undergraduate Thesis Faculty of Industrial Technology and Systems Engineering (INDSYS) > Physics Engineering > 30201-(S1) Undergraduate Thesis
Depositing User:	Anindriyo Kuncahyojati
Date Deposited:	15 Feb 2023 15:10
Last Modified:	15 Feb 2023 15:10
URI:	http://repository.its.ac.id/id/eprint/97399

Actions (login required)

View Item