Implementasi Sistem Perilaku Adaptif Dan Pengambilan Keputusan Strategis Pada Enemy Gim Strategi Endless-run Berbasis Unity Ml-agents

Putra, Mohammad Hanif Furqan Aufa Putra (2026) Implementasi Sistem Perilaku Adaptif Dan Pengambilan Keputusan Strategis Pada Enemy Gim Strategi Endless-run Berbasis Unity Ml-agents. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5025221161-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (3MB) | Request a copy

Abstract

Kecerdasan buatan (Artificial Intelligence) memiliki peran penting dalam meningkatkan kualitas pengalaman bermain pada gim strategi. Pendekatan berbasis aturan seperti Finite State Machine (FSM) banyak digunakan karena mudah diimplementasikan dan dikendalikan, namun cenderung menghasilkan perilaku yang deterministik dan mudah diprediksi. Penelitian ini bertujuan merancang, mengimplementasikan, dan mengevaluasi sistem perilaku adaptif pada enemy non-player character (NPC) menggunakan Unity ML-Agents serta membandingkannya dengan pendekatan FSM. Lingkungan eksperimen berupa gim strategi endless-run yang dikembangkan menggunakan Unity. Musuh diimplementasikan menggunakan dua pendekatan, yaitu FSM dan Reinforcement Learning melalui Unity ML-Agents dengan algoritma Proximal Policy Optimization (PPO). Proses pelatihan dilakukan dalam lingkungan adversarial menggunakan agen simulasi pemain berbasis Goal-Oriented Action Planning (GOAP) yang dikombinasikan dengan Utility System. Evaluasi dilakukan melalui pengujian terhadap 31 partisipan menggunakan metode within-subject design, System Usability Scale (SUS), dan Player Experience Questionnaire (PX). Hasil penelitian menunjukkan bahwa versi Unity ML-Agents memperoleh skor SUS rata-rata 69,52, lebih tinggi dibandingkan FSM sebesar 64,52. Pada aspek pengalaman bermain, Unity ML-Agents menunjukkan peningkatan pada dimensi kenyamanan, adaptabilitas, dan tantangan. Selain itu, pengamatan visual menunjukkan munculnya perilaku emergent seperti pack hunting, tactical retreat, dan pengejaran adaptif yang tidak diprogram secara eksplisit. Hasil penelitian menunjukkan bahwa Unity ML-Agents mampu menghasilkan perilaku musuh yang lebih adaptif dan memberikan pengalaman bermain yang lebih baik dibandingkan pendekatan FSM.
====================================================================================================================================
Artificial intelligence in endless-run strategy games plays a crucial role in determining difficulty levels and player engagement. Rule-based approaches such as the Finite State Machine (FSM) are widely used due to their ease of implementation, but they tend to produce deterministic and predictable behaviors. This research aims to design, implement, and evaluate an adaptive enemy behavior system in an endless-run strategy game using Unity ML-Agents, and compare it with the FSM approach. The study developed a Unity-based endless-run strategy game as an experimental environment. The enemy AI system was implemented using two approaches: (1) FSM which manages behavioral transitions deterministically, and (2) Unity ML-Agents with the Proximal Policy Optimization (PPO) algorithm. Enemy agents were trained in an adversarial environment against a simulated player agent based on Goal-Oriented Action Planning (GOAP) combined with a Utility System. Evaluation was conducted through user testing with 31 participants using a within-subject design, measuring usability (System Usability Scale/SUS) and player experience (PX). The results showed that the game version with Unity ML-Agents-based enemies achieved an average SUS score of 69.52 in Good category, outperforming the FSM version which scored 64.52 in Marginal category. In terms of player experience, the Unity ML-Agents version consistently outperformed on the dimensions of Comfort (+0.10), Adaptability (+0.06), and Challenge (+0.04). Visual observations also confirmed the emergence of emergent behaviors in ML-Agents, such as pack hunting, tactical retreat, and adaptive pursuit maneuvers that were not explicitly programmed. This research concludes that the reinforcement learning approach through Unity ML-Agents, supported by a GOAP-based adversarial environment, is capable of producing enemy behaviors xiii that are more adaptive, challenging, and perceived as more intelligent compared to the traditional FSM approach.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Reinforcement Learning, Unity ML-Agents, Game AI, GOAP, Adaptive Behavior, Endless-Run Strategy Game, Unity, Proximal Policy Optimization
Subjects:	G Geography. Anthropology. Recreation > GV Recreation Leisure > GV1469.2 Computer games 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)
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55201-(S1) Undergraduate Thesis
Depositing User:	Mohammad Hanif Furqan Aufa Putra
Date Deposited:	24 Jul 2026 07:36
Last Modified:	24 Jul 2026 07:36
URI:	http://repository.its.ac.id/id/eprint/137152

Actions (login required)

View Item