Designing Adaptive Enemy AI Using Utility System And Behavior Tree Methods In Turnbased RPG

Zhafir, Nafilah Yuga (2026) Designing Adaptive Enemy AI Using Utility System And Behavior Tree Methods In Turnbased RPG. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5025221108-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (6MB)

Abstract

Penelitian ini merancang dan mengimplementasikan kecerdasan buatan (AI) musuh yang adaptif pada gim strategi berbasis giliran (turn-based) dengan peta grid menggunakan Arsitektur Hibrida yang menggabungkan Utility System dan Behavior Tree. Permasalahan yang diangkat adalah perilaku musuh yang cenderung statis dan mudah ditebak pada implementasi Behavior Tree tanpa lapisan pengambilan keputusan. Solusi yang diterapkan memisahkan logika AI menjadi dua lapisan: lapisan pengambilan keputusan (The Brain) berupa Utility System yang menghitung skor agresif dan defensif secara real-time dari lima variabel kontekstual, serta lapisan eksekusi aksi (The Body) berupa Behavior Tree yang dibangun secara native di Unity 6. Pengujian dilakukan melalui 360 pertempuran otomatis yang menghasilkan 9.574 keputusan AI, dengan membandingkan mode Hibrida terhadap mode baseline sebagai kelompok kontrol pada tiga stage dengan karakteristik medan yang berbeda. Hasil penelitian menunjukkan bahwa mode Hibrida meningkatkan durasi bertahan hidup unit sebesar 21,8% dan persentase sisa HP sebesar 28,6% dibandingkan baseline, dengan tingkat kemenangan yang secara konsisten lebih tinggi (78,9% berbanding 70,0%). Sistem ini juga terbukti merespons secara proporsional terhadap gaya bermain pemain yang berbeda, dengan gradasi keputusan defensif mulai dari 49,0% untuk pemain agresif hingga 1,1% untuk pemain pasif. Hasil-hasil tersebut menunjukkan bahwa Arsitektur Hibrida mampu menghasilkan perilaku adaptif yang terukur tanpa memerlukan metode machine learning.
==========================================================================================================================================
This research designs and implements adaptive enemy artificial intelligence (AI) in a grid-based turn-based strategy game using a Hybrid Architecture that combines a Utility System and a Behavior Tree. The problem addressed is enemy behavior that tends to be static and easily predictable in a Behavior Tree implementation without a decision-making layer. The solution applied separates the AI logic into two layers: a decision-making layer (The Brain) in the form of a Utility System that computes aggressive and defensive scores in real-time from five contextual variables, and an action-execution layer (The Body) in the form of a Behavior Tree built natively in Unity 6. Testing was carried out through 360 automated battles that produced 9,574 AI decisions, comparing the Hybrid mode against a baseline mode as a control group across three stages with different terrain characteristics. Results show that the Hybrid mode increases unit survival duration by 21.8% and remaining HP percentage by 28.6% compared to baseline, with a consistently higher win rate (78.9% versus 70.0%). The system is also shown to respond proportionally to different player playstyles, with a gradient of defensive decisions ranging from 49.0% for aggressive players down to 1.1% for passive players. These results demonstrate that the Hybrid Architecture can produce measurable adaptive behavior without requiring machine learning methods.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Adaptive AI, Hybrid Architecture, Utility System, Behavior Tree, Turn-Based RPG, Unity.
Subjects:	T Technology > T Technology (General) > T57.84 Heuristic algorithms. T Technology > T Technology (General) > T58.62 Decision support systems
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55201-(S1) Undergraduate Thesis
Depositing User:	Nafilah Yuga Zhafir
Date Deposited:	25 Jul 2026 07:28
Last Modified:	25 Jul 2026 07:28
URI:	http://repository.its.ac.id/id/eprint/138102

Actions (login required)

View Item