A communication model for interactive POMDPs

Communication is one of the central activities in multiagent systems. It enables the information sharing among agents and improves the qualities of agents´ policies. In this paper, we introduce a communication decision problem to the interactive partially observable Markov decision processes (I-POMDPs), which provide a framework for sequential decision-making in multi-agent settings. In our framework, agents can communicate to obtain others´ observations and treat the received observations as their own. We extend the traditional observation function in I-POMDPs and show the belief update processes. We also develop a new graphical model for our framework by integrating communicative acts into I-DIDs, which are the graphical representations of I-POMDPs. We illustrate our framework using a simple application domain and show examples of belief updates and solutions.