A neural-network based behavioral theory of tank commanders

Abstract

Based on the presumption that data observed in high-tech and fast changing battles do have some intrinsic richness to them that synthetic modelling fails to capture, we contend that data induction techniques can be successfully used to generalize combat behaviors. This paper reports the use of neural networks as a computer-based adaptive induction algorithm to understand and uncover ground combat behaviors. Experiments with neural networks using tank movement data from the National Training Center (NTC), Fort Irwin, demonstrate that a two-dimensional cognitive map of closely task organized units can be derived. The findings seem to confirm our behavioral theory that tank commanders (i) are mission-driven, (if) act as an integral part of their platoon, (iii) perform sequential decision making to determine their next moves, and (iv) when isolated, extemporaneous behaviors may take precedence over normative gorup behavior. Once trained, a neural-network based model of closely task organized units can be used to predict the itinerary sequences of a tank given its initial geographic position.