Pengembangan Dead Reckoning Berbasis Multilayer Perceptron (MLP) Yang Diimplementasikan Di Raspberry Pi Pico

Satrio, Gagah Putra Haryo (2023) Pengembangan Dead Reckoning Berbasis Multilayer Perceptron (MLP) Yang Diimplementasikan Di Raspberry Pi Pico. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 07211940000009-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only until 1 September 2025. Download (5MB) | Request a copy

Abstract

Penentuan posisi sangatlah penting untuk pergerakan robot dalam ruangan tertutup (indoor). Salah satu penentuan posisi yang dapat digunakan adalah dead reckoning. Dead Reckoning adalah sistem navigasi berdasarkan lokasi awal relatif yang diketahui kemudian secara bertahap mengintegrasikan jarak yang ditempuh dan arah perjalanan untuk mengetahui titik lokasi selanjutnya. Karena termasuk sistem navigasi inersia yang menggunakan IMU (Inertial Measurement Unit), untuk melakukan prediksi posisi maka dilakukan perhitungan gabungan dari sensor giroskop, akseleromoter, dan magnetometer. Pada penelitian ini, akan menerapkan model deep learning yaitu Multilayer Perceptron. Sehingga dapat memberikan prediksi posisi berdasarkan data dari sensor IMU. Kemudian dari model yang telah dibuat akan dimasukkan ke mikrokontroler. Penggunaan mikrokontroler karena ukurannya yang kecil sehingga dapat diterapkan di robot kecil. Berbagai macam mikrokontroler saat ini, diantaranya Arduino nano, STM32F103C8T6, dan Raspberry Pi Pico. Pada penelitian ini memilih menggunakan mikrokontroler Raspberry Pi Pico karena memiliki beberapa keunggulan dari segi memori flash, SRAM dan kecepatan prosesornya dibandingkan mikrokontroler lainnya. Hal ini untuk menunjang penggunaan model Multilayer Perceptron ke dalam mikrokontroler. Model Multilayer Perceptron yang dibuat akan bervariasi untuk menemukan model yang ideal untuk sistem dead reckoning. Berdasarkan hasil dari pengujian model multilayer perceptron ke perangkat mikrokontroler Raspberry Pi Pico, maka dihasilkan model yang ideal yaitu model Multilayer Perceptron tanpa berbasis timeseries dengan 3 hidden layer dengan nodes masing-masing sebesar 250, 125, dan 30. Model ini dipilih karena menghasilkan nilai metrik R-square sebesar 0.7452684 pada pengujian di perangkat mikrokontrole Raspberry Pi Pico.
========================================================================================================================
============
Position determination is crucial for the movement of robots in enclosed environments (indoors). One of the position determination methods that can be used is dead reckoning. Dead reckoning is a navigation system based on the known initial location, which then gradually integrates the traveled distance and direction to determine the next location point. As it involves an inertial navigation system that uses an Inertial Measurement Unit (IMU), position prediction is achieved by combining data from the gyroscope, accelerometer, and magnetometer sensors. In this research, a deep learning model called Multilayer Perceptron will be implemented to provide position predictions based on IMU sensor data. Subsequently, the developed model will be incorporated into a microcontroller. The use of a microcontroller is preferred due to its small size, making it suitable for implementation in small robots. Various microcontrollers are available, such as Arduino Nano, STM32F103C8T6, and Raspberry Pi Pico. For this study, Raspberry Pi Pico is chosen because it offers several advantages in terms of flash memory, SRAM, and processor speed compared to other microcontrollers. This choice aims to support the integration of the Multilayer Perceptron model into the microcontroller. The Multilayer Perceptron model created will be varied to find the ideal model for the dead reckoning system. Based on the results of testing the Multilayer Perceptron model on the Raspberry Pi Pico microcontroller, an ideal model is obtained, namely a Multilayer Perceptron model without a time series basis, consisting of 3 hidden layers with nodes of 250, 125, and 30, respectively. This model is selected because it yields an R-square metric value of 0.7452684 in the testing on the Raspberry Pi Pico microcontroller.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Dead Reckoning, IMU, Multilayer Perceptron, Raspberry Pi Pico
Subjects:	Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines. Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science) T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK5105.546 Computer algorithms
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Computer Engineering > 90243-(S1) Undergraduate Thesis
Depositing User:	Gagah Putra Haryo Satrio
Date Deposited:	03 Aug 2023 15:31
Last Modified:	28 Aug 2023 08:45
URI:	http://repository.its.ac.id/id/eprint/101408

Actions (login required)

View Item