State space optimization using plan recognition and reinforcement learning on RTS game

  • Authors:

  • Jaeyong Lee;Bonjung Koo;Kyungwhan Oh

  • Affiliations:

  • Department of Computer Science and Engineering, Sogang University, Seoul, South Korea;Department of Computer Science and Engineering, Sogang University, Seoul, South Korea;Department of Computer Science and Engineering, Sogang University, Seoul, South Korea

  • Venue:
  • AIKED'08 Proceedings of the 7th WSEAS International Conference on Artificial intelligence, knowledge engineering and data bases
  • Year:
  • 2008

Quantified Score

Hi-index 0.00

Visualization

Abstract

Real Time Strategy (RTS) Game has the same problem which has to be solved in decision making in the real world. These problems are on real-time performance, high complexity caused by the large state space and multi-agent, insufficient information and on-line learning. AI which has been applied to RTS Game is somewhat limited and has poor performance due to these problems. Recent research to apply AI to RTS Game has proposed Dynamic Scripting. This method is generating rule-based game script by using reinforcement learning. Dynamic Scripting, however, assuming arbitrary and limited state space, is not able to fully reflect characteristics of real-time environment. This paper suggests a method to optimize state space as a solution. For state space optimization Case-Based Plan Recognition (CBPR) is utilized to model behavior pattern of opponent agent, and abstract state space which was used in CBPR is also utilized as well. And it leads to simplify state space to decrease high complexity and enable RTS game to apply to real-time environment. This method is used in distributed RTS Game, WARGUS in this paper.