Modeling Physical Variability for Synthetic MOUT Agents

Generating behavioral variability is an important prerequisite in the development of synthetic MOUT (Military Operations in Urban Terrain) agents for military simulations. Agents that lack variability are predictable and ineffective as opponents and teammates for human trainees. Along with cognitive differences, physical differences contribute towards behavioral variability. In this paper, we describe a novel method for modeling physical variability in MOUT soldiers using motion capture data acquired from human subjects. Motion capture data is commonly used to create animated characters since it retains the nuances of the original human movement. We build a cost model over the space of agent actions by creating and stochastically sampling motion graphs constructed from human data. Our results demonstrate how different cost models can induce variable behavior that remains consistent with military doctrine.