GAIG Game AI Research Group @ QMUL

Hyperstate Space Graphs for Automated Game Analysis

2019
Cook, Michael and Raad, Azalea

Abstract

Automatically analysing games is an important challenge for automated game design, general game playing, and co-creative game design tools. However, understanding the nature of an unseen game is extremely difficult due to the lack of a priori design knowledge and heuristics. In this paper we formally define hyperstate space graphs, a compressed form of state space graphs which can be constructed without any prior design knowledge about a game. We show how hyperstate space graphs produce compact representations of games which closely relate to the heuristics designed by hand for search-based AI agents; we show how hyperstate space graphs also relate to modern ideas about game design; and we point towards future applications for hyperstates across game AI research.
URL: https://ieeexplore.ieee.org/document/8848026

Cite this work

@inproceedings{cook2019hyperstate,
author= {Cook, Michael and Raad, Azalea},
title= {{Hyperstate Space Graphs for Automated Game Analysis}},
year= {2019},
booktitle= {{IEEE Conference on Games (CoG)}},
pages= {1--8},
url= {https://ieeexplore.ieee.org/document/8848026},
abstract= {Automatically analysing games is an important challenge for automated game design, general game playing, and co-creative game design tools. However, understanding the nature of an unseen game is extremely difficult due to the lack of a priori design knowledge and heuristics. In this paper we formally define hyperstate space graphs, a compressed form of state space graphs which can be constructed without any prior design knowledge about a game. We show how hyperstate space graphs produce compact representations of games which closely relate to the heuristics designed by hand for search-based AI agents; we show how hyperstate space graphs also relate to modern ideas about game design; and we point towards future applications for hyperstates across game AI research.},
}

Comments

Content