Sistem Kontrol Drone Holybro Berbasis Hand Gesture Menggunakan Long Short-Term Memory (LSTM)

Nafiq, Zadun (2026) Sistem Kontrol Drone Holybro Berbasis Hand Gesture Menggunakan Long Short-Term Memory (LSTM). Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5024221018-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (37MB) | Request a copy

Abstract

Antarmuka kendali drone yang mudah dipelajari dan aman dibutuhkan untuk memperluas adopsi Unmanned Aerial Vehicle (UAV) dalam skenario sehari-hari. Penelitian ini merancang dan mengimplementasikan sistem kendali drone berbasis gestur satu tangan secara real-time menggunakan MediaPipe Hands dan model Long Short-Term Memory (LSTM). Sistem diintegrasikan dengan flight controller Pixhawk 6C pada drone Holybro S500 v2 yang menggunakan firmware PX4 melalui mode Offboard. Data visual ditangkap menggunakan kamera webcam 720p pada 30 fps dengan jarak 50 cm, di mana setiap frame diekstraksi menjadi 21 hand landmarks yang telah dinormalisasi. Fitur tersebut kemudian diklasifikasikan oleh model LSTM ke dalam 10 gestur, yaitu maju, mundur, kiri, kanan, naik, turun, putar kiri, putar kanan, hover, dan start/end. Stabilitas keputusan dijaga melalui confidence threshold, mekanisme majority voting berbasis prediction buffer, serta command cooldown. Label gestur dipetakan ke setpoint kecepatan linear dan laju yaw menggunakan protokol MAVLink. Sebagai fitur keselamatan, penelitian ini mengimplementasikan mekanisme Return-to-Launch (RTL) otomatis berbasis pemantauan komunikasi. Sistem dirancang untuk memicu mode RTL secara otomatis apabila drone kehilangan aliran data perintah dari ground station selama lebih dari tiga detik. Pendekatan ini memastikan drone dapat kembali ke titik awal secara mandiri ketika terjadi kegagalan sistem pengenalan atau terputusnya koneksi, sehingga meningkatkan tingkat keamanan operasi dibandingkan metode kendali manual sepenuhnya. Kontribusi utama penelitian ini adalah perancangan pipeline visi-ke-kendali yang ringan dengan protokol keselamatan terintegrasi pada platform drone kustom kelas menengah.
===============================================================================================================================
User-friendly and secure control interfaces are essential for expanding the adoption of Unmanned Aerial Vehicles (UAVs) in everyday scenarios. This study designs and implements a real-time, one-handed gesture-based drone control system utilizing MediaPipe Hands and a Long Short-Term Memory (LSTM) model. The system is integrated with a Pixhawk 6C flight controller on a Holybro S500 v2 drone running PX4 firmware via Offboard mode. Visual data are captured using a 720p webcam at 30 fps from a distance of 50 cm, where each frame is processed to extract 21 normalized hand landmarks. These features are classified by the LSTM model into ten distinct gestures: forward, backward, left, right, up, down, yaw left, yaw right, hover, and start/end. Decision stability is maintained through a confidence threshold, a buffer-based majority voting mechanism, and command cooldowns. Gesture labels are mapped to linear velocity and yaw rate setpoints using the MAVLink protocol. As a safety feature, an automatic Return-to-Launch (RTL) mechanism based on communication monitoring is implemented. The system automatically triggers RTL if the command stream from the ground station is interrupted for more than three seconds. This approach enables the drone to autonomously return to its home point during recognition failures or communication losses, thereby enhancing operational safety compared to fully manual control. The primary contribution of this research is a lightweight vision-to-control pipeline with integrated safety protocols for mid-range custom drone platforms.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Gestur tangan, Kendali Drone, LSTM, Pixhawk 6C, Holybro S500, PX4 Offboard, Hand gesture, Drone Control, LSTM, Pixhawk 6C, Holybro S500, PX4 Offboard
Subjects:	Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines. Q Science > QA Mathematics > QA336 Artificial Intelligence Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science) T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK2681.O85 Electric motors, Brushless.
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Computer Engineering > 90243-(S1) Undergraduate Thesis
Depositing User:	Zadun Nafiq
Date Deposited:	13 Jul 2026 04:20
Last Modified:	13 Jul 2026 04:20
URI:	http://repository.its.ac.id/id/eprint/134762

Actions (login required)

View Item