Hierarchical Reinforcement Learning Model for Military Simulations

Majority of the actions in army are hierarchical and occur simultaneously with some other action. Mission of an echelon is sub-divided into sub-missions which are assigned to the lower echelon. These lower echelons pursue their missions simultaneously. To apply reinforcement learning to such highly concurrent actions' domain as military, we propose a concurrent options model for a set of temporally extended actions that may not terminate at the same time and trigger the next transition without any regard for the other sub-options. We provide formal representation of the model.