A Real-time Strategy Agent Framework and Strategy Classifier for Computer Generated Forces
Abstract
This research effort is concerned with the advancement of computer generated forces AI for Department of Defense (DoD) military training and education. The vision of this work is agents capable of perceiving and intelligently responding to opponent strategies in real-time. Our research goal is to lay the foundations for such an agent. Six research objectives are defined: 1) Formulate a strategy definition schema effective in defining a range of RTS strategies. 2) Create eight strategy definitions via the schema. 3) Design a real-time agent framework that plays the game according to the given strategy definition. 4) Generate an RTS data set. 5) Create an accurate and fast executing strategy classifier. 6) Find the best counterstrategies for each strategy definition. The agent framework is used to play the eight strategies against each other and generate a data set of game observations. To classify the data, we first perform feature reduction using principal component analysis or linear discriminant analysis. Two classifier techniques are employed, k-means clustering with k-nearest neighbor and support vector machine. The resulting classifier is 94.1% accurate with an average classification execution speed of 7.14 us. Our research effort has successfully laid the foundations for a dynamic strategy agent.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jun 01, 2012
- Accession Number
- ADA563243
Entities
People
- Lyall J. Di Trapani
Organizations
- Air Force Institute of Technology