Learning and Generating Cooperative Behavior Based on Multi-Agent Symbol Emergence

Fusion of Control as Inference and Metropolis Naming Game

Abstract

This paper proposes a generative probabilistic model (PGM) of emergent communication for multi-step cooperative tasks performed by two agents. The agents plan their actions by probabilistic inference, called control as inference, and messages communicated between two agents are latent variables and estimated based on the planned actions. Through these messages, each agent can send information about its own actions and know information about the actions of another agent. Therefore, the agents change their actions according to the estimated messages to achieve cooperative tasks. This inference of messages can be considered as communication, and this procedure can be formulated by the Metropolis naming game. Through experiments in the grid world environment, we show that the proposed PGM can infer meaningful messages to achieve the cooperative task.