Entropy to Control Planning in Video-Games

We consider the problem of controlling the online planning of the behaviors of non-player characters (NPCs) in video-games. We present an urn model of this online planning of the behaviors of NPCs where urns are NPCs, the plans correspond to balls in an urn, and the type of a plan correspond to the color of a ball, to compute the information entropy of the planning activity. We compare this model with the planning entropy measurements from six commercial First-Person Shooters: our model predicts successive variations in entropy and sometimes closely adjusts to experimental values; we thus provide macroscopic parameters and combine them in a simple ratio to control the planning of behaviors of NPCs.