An Influence Diagram framework for acting under influence by agents with unknown goals
Nicolaj Sønderberg-Madsen, Finn Verner Jensen · VBN Forskningsportal (Aalborg Universitet) · 2008
We consider the situation where two agents try to solve each their own task in a common environment. We present a general framework for representing that kind of scenario based on Influence Diagrams (IDs). The framework is used to model the analysis depth and time horizon of the opponent agent and to determine an optimal policy under various assumptions on analysis depth of the opponent. Not surprisingly, the framework turns out to have severe complexity problems even in simple scenarios due to the size of the relevant past. We propose an algorithm based on Limited Memory Influence Diagrams (LIMIDs) in which we convert the ID into a Bayesian network and perform single policy update. Empirical results are presented using a simple board game.