Evolving Strategies For a Real-Time Strategy Game Using Genetic Algorithms

Student thesis: Master thesis (including HD thesis)

  • Steffan Bo Pallesen
  • Mikkel Graarup Jensen
  • Nikolaj Dam Larsen
4. term, Software, Master (Master Programme)
The demand for challenging AI in commercial
Real-Time Strategy games is increasing
by each generation. This project explores
how genetic algorithms might be used to
evolve more challenging AIs, by utilizing the
power of natural selection. The popular strategy
game Starcraft is used a testing platform
throughout the project, due to its maturity
as a research platform. Two approaches to
evolving a Starcraft AI is examined. One is
to use a traditional genetic algorithm, and
the other is to use a Estimation of Distribution
Genetic Algorithm. However, only the
traditional genetic algorithm is implemented
and tested. The results show that it is in
fact possible to evolve an AI that can successfully
beat an opponent playing the same
static strategy. More research is needed in order
to determine if genetic algorithms can be
used to evolve complete commercial grade
AIs.
LanguageEnglish
Publication date7 Jun 2013
Number of pages88
ID: 77336305