Automatic Power-up Items Placement on Shooter Game using Convolutional Neural Network

Alvin Satria Nugraha, Abas Setiawan, Wijanarto Wijanarto

Abstract


- A shooter game is a popular game genre with various components. To make a shooter game more attractive, some power-ups items can support players to achieve their goals. Power-ups items provide more power to players, some of which include ammo, extra lives, and invulnerability. The location of power-ups items should be in a special place so that it neither too easy to find nor too difficult to find. Item placement could be done manually by a human or a technical artist. It will need a relatively long time and high cost. In this paper, we try to mimic technical artist vision when placing an item. Visual images have been collected by scanning spatially the forest terrain by using a virtual camera on top. Each image data comply with the item placement rules according to the Tomb Raider and Uncharted 4 games. Convolutional Neural Network (CNN) is used to find out which images can be occupied by power-up items or not. From several experimental scenarios, the use of the Global Average Pooling layer is proven to produce a model that is not overfitting. The best CNN models are developed and got an accuracy of 90.5% with an architecture that includes the Global Average Pooling layer. That model is applied to the new forest terrain so that power-up items can automatically be placed in an appropriate location.


Full Text:

PDF


DOI: https://doi.org/10.33633/jais.v5i1.4213

Article Metrics

Abstract view : 236 times
PDF - 219 times

Refbacks

  • There are currently no refbacks.


Flag Counter

 

 

 

 

Journal of Applied Intelligent System (e-ISSN : 2502-9401p-ISSN : 2503-0493) is published by Department of Informatics Universitas Dian Nuswantoro Semarang and IndoCEISS.

  

 

Journal of Applied Intelligent System indexed by :


This journal is under licensed of Creative Commons Attribution 4.0 International License.

Visitor Stats