Find Optimal Rush Attacks in Real Time Strategy Games

1
Find Optimal Rush Attacks in Real Time Strategy
Games

• Li wenjie
• Surpervior ole christoffer
• University of adger,2008

2
Introduction

• Real time strategy games (RTS)
• --several players struggle in a resource
  scattered map. Player commands the game units to
  developing economy, climbing technology, crafting
  new troop. The aim is crafting a powerful troop
  and guiding them into battle. Afterwards, destroy
  you opponents base and troop.
• RTS brings high competition and intensity
  attention to players. However, current AI is so
  weak and stable.
• --RTS game is so complex for AI to handle.
• --commercial games close its source.
• --lower competition in AI research.
• Open real time strategy game design environment.

3
General Goal

• Analyses the standard game information in ORTS.
  Afterwards, we design our AI package with
  rush-attack strategy.
• Design our learning method.
• -- learning automaton
• --decision tree
• --point system (our design)

4
AI mode and General Process
scout
Resource collection
server
client
5
Learning method

• It is a quintuple P, F, S, E, M .
• P the point of units.
• F the unchangeable game information, like hit
  point, attack power and so on.
• S game states which may influence the point.
• E Experience, it the key structure in this
  system, its a probability that denotes the
  performance of units in real conflict..
• M method to learning, in a word how to make the
  experience more precise and efficient.

P F ?(ES) Enew M (E)
6
Our implementation (simulation in Java)
7
Testing
Probability of win
8
Conclusion

• We find a new learning theory for AI in RTS
  games. Advanced Point system
• After testing, the system works well and it
  would increase the probability of win and minimal
  the time to start attack.
• There are other elements influence the point
  calculation. These would be future works.

9
Thanks! Question?

• Contact
• Wenjie Li
• jevons.lee_lwj_at_hotmail.com