Pembangkitan Level Permainan Super Mario Bros berdasarkan Perilaku Pemain Menggunakan Integrasi Reinforcement Learning dan TOAD-GAN

Sanjaya, Mohammad Ovi (2026) Pembangkitan Level Permainan Super Mario Bros berdasarkan Perilaku Pemain Menggunakan Integrasi Reinforcement Learning dan TOAD-GAN. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6026232003-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (13MB) | Request a copy

Abstract

Penelitian ini bertujuan mengembangkan sistem pembangkitan level permainan adaptif dengan mengintegrasikan Procedural Content Generation (PCG) dan Dynamic Difficulty Adjustment (DDA). Pendekatan yang diusulkan mengombinasikan TOAD-GAN sebagai generator level berbasis token dengan Reinforcement Learning menggunakan algoritma Proximal Policy Optimization (PPO) untuk menyesuaikan tingkat kesulitan secara dinamis berdasarkan performa pemain. Evaluasi dilakukan melalui pengujian kinerja agen PPO, analisis struktur level menggunakan metrik TPKL-Div, serta pengujian pengalaman bermain. Hasil penelitian menunjukkan bahwa agen PPO menghasilkan proses pembelajaran yang lebih stabil, integrasi DDA tidak mengganggu orisinalitas level TOAD-GAN dengan rata-rata selisih TPKL-Div sebesar 0,15, serta meningkatkan tingkat playability level dari 65% menjadi 80% setelah perbaikan level. Selain itu, hasil pengujian menunjukkan bahwa sistem mampu menjaga keseimbangan antara tantangan dan pengalaman bermain pemain.
=====================================================================================================================================
This research aims to develop an adaptive game level generation system by integrating Procedural Content Generation (PCG) and Dynamic Difficulty Adjustment (DDA). The proposed approach combines TOAD-GAN as a token-based level generator with Reinforcement Learning using the Proximal Policy Optimization (PPO) algorithm to dynamically adjust the difficulty level based on player performance. The evaluation was conducted through PPO agent performance testing, level structure analysis using the TPKL-Div metric, and gameplay testing. The results of the study show that the PPO agent produces a more stable learning process, the integration of DDA does not interfere with the originality of the TOAD-GAN level with an difference average TPKL-Div value of 0.15, and increases the level of playability from 65% to 80% after level improvement. In addition, the test results show that the system is able to maintain a balance between challenge and player experience

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	Procedural Content Generation, Dynamic Difficulty Adjustment, Generative Advesarial Network, Reinforcement Learning = Procedural Content Generation, Dynamic Difficulty Adjustment, Generative Advesarial Network, Reinforcement Learning
Subjects:	T Technology > T Technology (General) > T58.62 Decision support systems
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Information System > 59101-(S2) Master Thesis
Depositing User:	Mohammad Ovi Sanjaya
Date Deposited:	29 Jan 2026 06:38
Last Modified:	29 Jan 2026 06:38
URI:	http://repository.its.ac.id/id/eprint/131023

Actions (login required)

View Item