Predicting Opponent Position and Modeling Uncertainty
Abstract
Current combat simulation software developments for automated planning do not account for fog-of-war in their methods. This makes their outputs less realistic, as it is not reasonable to have the exact enemy positions in real-world planning. An artificial intelligence-controlled force should be able to operate without information that is not available to a human in the same situation. This dissertation presents a method for AI agents to predict and assess possible opposing force positions given typical intelligence products. We also present a method to aggregate the risk implications of these positions. We demonstrate the techniques in a combat simulation environment and evaluate their performance in multiple battle scenarios. The results show the importance of uncertainty in combat simulations and illustrate that our method of risk aggregation can be effective.
Document Details
- Document Type
- Technical Report
- Publication Date
- Sep 01, 2020
- Accession Number
- AD1126513
Entities
People
- Kenneth J. Maroon
Organizations
- Naval Postgraduate School