Desain Sistem Penghindaran Rintangan Berbasis Deep Deterministic Policy Gradient untuk Navigasi Mobile Robot

Pratama, Eryan Bima (2025) Desain Sistem Penghindaran Rintangan Berbasis Deep Deterministic Policy Gradient untuk Navigasi Mobile Robot. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5022211093-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (4MB) | Request a copy

Abstract

Penghindaran rintangan merupakan aspek penting dalam navigasi mobile robot, terutama dalam lingkungan yang dinamis dan kompleks. Penelitian sebelumnya telah menerapkan algoritma Deep Deterministic Policy Gradient (DDPG) dalam skenario navigasi robot dengan pendekatan standar. Pada penelitian ini, dilakukan pengembangan lebih lanjut terhadap algoritma DDPG melalui perluasan vektor observasi, arsitektur jaringan aktor-kritik yang lebih dalam, strategi eksplorasi aksi berdasarkan arah waypoint, serta penerapan curriculum learning yang disesuaikan secara bertahap. Efektivitas metode yang diusulkan divalidasi melalui perbandingan langsung terhadap algoritma Q-Learning yang telah diimplementasikan dalam lingkungan yang sama. Hasil menunjukkan bahwa DDPG yang dikembangkan mampu meningkatkan reward rata-rata sebesar 19,80%, menurunkan waktu tempuh hingga 30,37%, serta mengurangi error pencapaian waypoint sebesar 81,21% dibandingkan dengan algoritma Q-Learning. Hal ini membuktikan bahwa pendekatan DDPG yang dikembangkan lebih efisien, adaptif, dan akurat dalam penghindaran rintangan mobile robot.
=======================================================================================================================================
Obstacle avoidance is an important aspect in mobile robot navigation, especially in dynamic and complex environments. Previous research has applied the Deep Deterministic Policy Gradient (DDPG) algorithm in robot navigation scenarios with a standardized approach. In this study, further development of the DDPG algorithm is carried out through the expansion of observation vectors, deeper actor-critic network architecture, action exploration strategy based on waypoint direction, and the application of incrementally adjusted curriculum learning. The effectiveness of the proposed method is validated through direct comparison against the Q-Learning algorithm that has been implemented in the same environment. The results show that the developed DDPG is able to increase the average reward by 19.80%, decrease the travel time by 30.37%, and reduce the waypoint achievement error by 81.21% compared to the Q-Learning algorithm. This proves that the developed DDPG approach is more efficient, adaptive, and accurate in mobile robot obstacle avoidance.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Mobile robot, Penghindaran rintangan, Reinforcement Learning, Neural Network, Deep Deterministic Policy Gradient, Q-Learning, Mobile robot, Obstacle avoidance, Reinforcement Learning, Neural Network, Deep Deterministic Policy Gradient, Q-Learning.
Subjects:	T Technology > TJ Mechanical engineering and machinery > TJ211.415 Mobile robots
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Electrical Engineering > 20201-(S1) Undergraduate Thesis
Depositing User:	Eryan Bima Pratama
Date Deposited:	25 Jul 2025 03:08
Last Modified:	25 Jul 2025 03:08
URI:	http://repository.its.ac.id/id/eprint/121213

Actions (login required)

View Item