Statistical Planning: Building Models of Entropy of Centralized Planning for Multi-Agent Systems

Abstract

We present a statistical framework to build a family of models of the information entropy of the planning activity which is used to control the behaviors of agents in multi-agent systems such as commercial video-games. We discuss the (practical) hypotheses which are necessary to build this (theoretical) framework and present the probability distributions of the plans among the agents. We confront our entropy models with measures of the planning entropy in five commercial first-person shooters: our theoretical models either frame or else fit the experimental measures.