Analisis Perbandingan Inverse Kinematics Metode Analitis Dengan Artificial Neural Network Pada Robot Arm 4 DOF Berdasarkan Akurasi Posisi End Effector

Naian, Aliffitrah Paksy (2026) Analisis Perbandingan Inverse Kinematics Metode Analitis Dengan Artificial Neural Network Pada Robot Arm 4 DOF Berdasarkan Akurasi Posisi End Effector. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 2040221110-Undergraduate_Thesis.pdf Restricted to Repository staff only Download (6MB) | Request a copy

Abstract

Robot arm merupakan salah satu teknologi otomasi yang banyak diterapkan di industri, di mana ketepatan posisi end-effector menjadi parameter utama keberhasilan tugasnya. Prototipe robot arm 4-DOF di Laboratorium Electric Drive and Power Electronic System (ELDRIVE) sebelumnya masih dikendalikan berdasarkan perintah sudut secara langsung, dan pengujian yang telah dilakukan hanya berfokus pada evaluasi duty cycle servo dengan nilai MAPE sebesar 0,91%, serta baru memiliki pemodelan forward kinematics tanpa penerapan inverse kinematics maupun pengujian akurasi posisi end-effector. Penelitian ini bertujuan untuk menerapkan dan membandingkan dua metode inverse kinematics, yaitu metode analitis dan Artificial Neural Network (ANN), berdasarkan akurasi posisi end-effector yang dihasilkan. Metode analitis diselesaikan melalui pendekatan hubungan geometris antar-link robot dengan orientasi end-effector dikunci konstan, sedangkan metode ANN dilatih menggunakan dataset hasil pemodelan forward kinematics dan diimplementasikan pada mikrokontroler ESP32-S3, dengan konfigurasi terbaik berupa arsitektur hidden layer 128-128-64, learning rate 0,001, batch size 32, dan rasio dataset 80:10:10 yang diperoleh melalui pengujian empiris bertahap. Pengujian akurasi dilakukan dengan memberikan 15 titik koordinat target yang identik pada kedua metode, kemudian sudut joint hasil perhitungan diimplementasikan pada robot fisik dan posisi end-effector aktual diukur untuk dibandingkan dengan posisi target. Hasil pengujian menunjukkan bahwa metode analitis menghasilkan rata-rata error posisi sebesar 3,21% (1,93% sumbu X, 3,59% sumbu Y, 4,10% sumbu Z), jauh lebih rendah dibandingkan metode ANN sebesar 25,15% (19,16% sumbu X, 35,83% sumbu Y, 20,47% sumbu Z), dengan metode analitis unggul konsisten pada seluruh sumbu dan titik uji (0,85–7,71%), sedangkan ANN berfluktuasi tajam hingga 49,14%. Evaluasi model ANN menunjukkan R² menurun monoton dari 0,997 (θ1) hingga 0,033 (θ4), mengindikasikan keterbatasan inheren model dalam memetakan joint distal akibat sifat inverse kinematics yang redundan, di samping indikasi kesalahan sistematis pada integrasi model ke mikrokontroler. Dengan demikian, metode analitis dinilai lebih sesuai diterapkan pada sistem robot arm yang membutuhkan presisi dan konsistensi tinggi.
===========================================================================================================================================
Robotic arms are one of the automation technologies widely applied in industry, where end-effector positioning accuracy is the main parameter of task success. The 4-DOF robotic arm prototype at the Electric Drive and Power Electronic System (ELDRIVE) Laboratory was previously controlled through direct angle commands, with prior testing limited to servo duty cycle evaluation (MAPE of 0.91%), and only possessed a forward kinematics model without inverse kinematics implementation or end-effector accuracy testing. This research aims to implement and compare two inverse kinematics methods, namely the analytical method and Artificial Neural Network (ANN), based on the resulting end-effector position accuracy. The analytical method was solved through a geometric relationship approach between links with a constant end-effector orientation, while the ANN method was trained using a dataset generated from the forward kinematics model and implemented on an ESP32-S3 microcontroller, with the best configuration consisting of a 128-128-64 hidden layer architecture, learning rate of 0.001, batch size of 32, and dataset split ratio of 80:10:10 obtained through sequential empirical testing. Accuracy testing was conducted using 15 identical target coordinates for both methods, with resulting joint angles implemented on the physical robot and actual end-effector positions measured for comparison. Results show the analytical method produced an average positional error of 3.21% (1.93% X-axis, 3.59% Y-axis, 4.10% Z-axis), significantly lower than the ANN method's 25.15% (19.16% X-axis, 35.83% Y-axis, 20.47% Z-axis), with the analytical method consistently outperforming across all axes and test points (0.85–7.71%), while ANN fluctuated sharply up to 49.14%. ANN model evaluation revealed R² decreasing monotonically from 0.997 (θ1) to 0.033 (θ4), indicating inherent model limitations in mapping distal joints due to the redundant nature of inverse kinematics, alongside indications of systematic error during model-to-microcontroller integration. Therefore, the analytical method is considered more suitable for robotic arm systems requiring high precision and consistency.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Robot arm, Artificial Neural Network , Inverse Kinematic, 4 DOF, akurasi end effector, Robotic Arm, Artificial Neural Network, Inverse kinematics, 4 DOF, end effector accuracy.
Subjects:	T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK3070 Automatic control
Divisions:	Faculty of Vocational > 36304-Automation Electronic Engineering
Depositing User:	Aliffitrah Paksy Naian
Date Deposited:	05 Aug 2026 02:37
Last Modified:	05 Aug 2026 02:37
URI:	http://repository.its.ac.id/id/eprint/143274

Actions (login required)

View Item