Please use this identifier to cite or link to this item: https://repository.uksw.edu//handle/123456789/25849
Title: Rancang Bangun Robot Maze Navigation Menggunakan Convolutional Neural Network
Other Titles: An Embedded Computer Vision using Convolutional Neural Network for Maze Classification and Robot Navigation
Authors: Hadiyanto, Dinar Rahmat
Keywords: robot;autonomous;convolutional neural network
Issue Date: Jul-2022
Abstract: Pengenalan citra menjadi salah satu metode yang dapat digunakan dalam penyelesaian masalah. Robot pemecah labirin dapat menggunakan sensor ultrasonik untuk mendeteksi adanya ruang gerak di sekitar robot, tetapi robot tetap saja tidak dapat mengenali jenis rintangannya. Dalam penelitian ini pemanfaatan sistem pengenal citra yang dijalankan menggunakan komputer mini, dan kamera digunakan untuk membantu robot pemecah labirin mengenali jenis rintangan pada suatu labirin. Jenis rintangan yang akan dikenali terbagi menjadi beberapa jenis seperti simpang empat, simpang tiga, jalan buntu, belokan kanan, belokan kiri, zona mulai, zona selesai. Convolutional neural network digunakan untuk melatih robot untuk mengenali jenis-jenis citra berdasarkan set data yang telah disediakan. Hasil penelitian diperoleh dengan akurasi latih sebesar 82% dengan waktu latih 30 menit 15 detik, dan proses klasifikasi membutuhkan 0,5 detik pada setiap pengujiannya.
Image recognition is one of the methods that can be used in solving problems. The maze solving robot can use ultrasonic sensors to detect the presence of space around the robot, but the robot still cannot recognize the type of obstacle. In this study, the use of an image recognition system that is run using a mini computer, and a camera is used to help a maze-solving robot recognize the types of obstacles in a labyrinth. The types of obstacles that will be recognized are divided into several types such as four-way intersection, three-way intersection, dead end, turn right, turn left, start zone, finish zone. Convolutional neural network is used to train robots to recognize types of images based on the data set that has been provided. The results obtained with training accuracy of 82% with a training time of 30 minutes 15 seconds, and the classification process requires 0.5 seconds for each test.
URI: https://repository.uksw.edu/handle/123456789/25849
Appears in Collections:T1 - Electrical Engineering

Files in This Item:
File Description SizeFormat 
T1_612016031_Judul.pdf989.09 kBAdobe PDFView/Open
T1_612016031_Isi.pdf652.77 kBAdobe PDFView/Open
T1_612016031_Daftar Pustaka.pdf376.17 kBAdobe PDFView/Open
T1_612016031_Lisensi.pdf807.1 kBAdobe PDFView/Open


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.