Evolutionary programming for automatically generating strategies in real-time strategy games

Ch'ng, Siong Huat (2010) Evolutionary programming for automatically generating strategies in real-time strategy games. Masters thesis, Universiti Malaysia Sabah.


Download (2MB) | Preview


The video game industry is currently estimated to be worth USD 41 billion annually and growing every year. In this industry, the developments in game Artificial Intelligence (AI) typically use traditional techniques such as minimax, alpha-beta and A* search algorithms in their implementations. More advanced techniques such as evolutionary algorithms are very rarely used. Therefore, this thesis will focus specifically on the implementation of evolutionary algorithms for the automatic generation of military strategies. WARGUS, as the open source RTS game engine, is chosen to be the research platform due to its extendibility and possibility for integration with an external AI system. This research encompasses three main research questions: (i) can Evolutionary Programming (EP) be adapted successfully as the game AI into an RTS in the form of WARGUS?; (ii) how well can the EP system perform in terms of scoring capability and learning speed using modified EP systems implemented specifically to cope with the RTS environment?; and (iii) will this EP system be suitable for a full game scenario in WARGUS? The experimental results of this series of investigations have showed that: (i) the EP proved itself to be highly useful when incorporated as the game AI into WARGUS where it outperformed a benchmark random strategy system as well as a hill-climbing system; (ii) after modifying the standard EP specifically for the RTS game environment, a new modified EP was successfully created which is able to obtain high scoring capabilities in a comparatively shorter amount of time; and (iii) the modified EP also performed successfully when plugged into a full RTS game scenario where highly encouraging results as high as 80%, 90% and 100% winning rates were achieved against three state-of-the-art WARGUS full-game strategies crafted by expert human players which is later refined to full 100% winning rates against all three strategies using an incremental seeding methodology

Item Type: Thesis (Masters)
Uncontrolled Keywords: video game industry, evolutionary algorithm, EP system, WARGUS, AI system
Subjects: Q Science > QA Mathematics > QA76 Computer software
Divisions: SCHOOL > School of Engineering and Information Technology
Date Deposited: 03 Dec 2013 03:45
Last Modified: 12 Oct 2017 03:07
URI: http://eprints.ums.edu.my/id/eprint/7703

Actions (login required)

View Item View Item