Affan, Lazuardi Y. (2018) Penentuan Fungsi Utility Terbaik Berbasis Genetic Algorithm untuk Perilaku NPC Menggunakan Utility-Based AI dalam Fighting Game. Undergraduate thesis, Institut Teknologi Sepuluh Nopember.
Preview |
Text
07211440000016-Undergraduate_Theses.pdf - Accepted Version Download (11MB) | Preview |
Abstract
Dalam fighting game, fitur bermain dengan Non-Player Character (NPC) merupakan suatu keharusan. Salah satu metode desain AI untuk NPC pada game adalah utility-based AI. Utility-based AI bekerja dengan cara mengambil pilihan aksi terbaik melalui pemberian skor kebergunaan pada setiap aksi berdasarkan keadaan tertentu, yang disebut dengan nilai utility. Hal ini menyebabkan AI menjadi lebih mudah didesain, dan pengambilan keputusan juga lebih variatif, karena pengevaluasian pilihan dihitung dengan fungsi yang bersifat kontinu. Namun, karena nilai utility dihitung dengan fungsi yang ditentukan secara bebas, maka akan terdapat banyak kemungkinan bentuk fungsi, sehingga pada penelitian ini digunakanlah metode genetic algorithm (GA) untuk menentukan fungsifungsi utility terbaik pada setiap aksi suatu NPC, yang setiap kromosomnya tersusun atas gen-gen berupa fungsi-fungsi utility. Fungsi fitness yang digunakan adalah dengan perhitungan ELO Ratings. Setelah dilakukan pelatihan dengan GA, AI cenderung bertambah kuat seiring bertambahnya generasi. Dari hasil pengujian terhadap manusia, rata-rata tingkat kesulitan dan tingkat kepuasan cenderung naik seiring bertambahnya generasi, yang diikuti dengan turunnya rata-rata status kemenangan. Dengan perhitungan koefisien korelasi Pearson, status kemenangan dan tingkat kesulitan memiliki korelasi -0,504, status kemenangan dan tingkat kepuasan memiliki korelasi -0,037, sedangkan tingkat kesulitan dan tingkat kepuasan memiliki korelasi 0,426.
=========================================================
In fighting games, a feature to play with Non-Player Character (NPC) is required, that the NPC would be more challenging when its behaviour is more unpredictable. One of the game AI designing methods is utility-based AI. Utility-based AI works by identifying action options and then choosing the best option through usefulness scoring of each action based on certain conditions, which is called utility value. This causes AI to be easier to design, and enabling more various decision making, because the decision evaluation is calculated by continuous function. But, since the utility calculation is based on functions those are defined manually by the AI designer, there will be too many possibilities of functions used. So, in this final project, genetic algorithm (GA) was implemented to determine the best utility functions on each action of the NPC, where each chromosome consists of utility functions as its genes. The fitness function is obtained from the ELO Ratings calculation. As the result of the GA implementation after training, the AIs tend to be stronger as the generation iterates. As the result of testing against human players, the mean of the difficulty level and satisfaction level of the AIs tend to increase as the generation iterates, followed by the decreasing of the mean of victory status. Using Pearson’s correlation coefficient, victory status and difficulty level have correlation value of -0.504, victory status and satisfaction level have correlation value of -0.037, while difficulty level and satisfaction level have correlation value of 0.426.
| Item Type: | Thesis (Undergraduate) |
|---|---|
| Uncontrolled Keywords: | Kecerdasan Buatan, Utility-Based AI, Genetic Algorithm, ELO Ratings, Koefisien Korelasi Pearson |
| Subjects: | G Geography. Anthropology. Recreation > GV Recreation Leisure > GV1469.2 Computer games Q Science > QA Mathematics > QA336 Artificial Intelligence Q Science > QA Mathematics > QA402.5 Genetic algorithms. Interior-point methods. |
| Divisions: | Faculty of Electrical Technology > Computer Engineering > 90243-(S1) Undergraduate Thesis |
| Depositing User: | Lazuardi Ya'qub Affan |
| Date Deposited: | 18 Jun 2021 12:50 |
| Last Modified: | 18 Jun 2021 12:50 |
| URI: | http://repository.its.ac.id/id/eprint/58718 |
Actions (login required)
 |
View Item |