Decentralized classification in societies of autonomous and heterogenous robots

This paper addresses the classification problem for a set of autonomous robots that interact with each other. The objective is to classify agents that "behave" in "different way", due to their own physical dynamics or to the interaction protocol they are obeying to, as belonging to different "species". This paper describes a technique that allows a decentralized classification system to be built in a systematic way, once the hybrid models describing the behavior of the different species are given. This technique is based on a decentralized identification mechanism, by which every agent classifies its neighbors using only local information. By endowing every agent with such a local classifier, the overall system is enhanced with the ability to run behaviors involving individuals of the same species as well as of different ones. The mechanism can also be used to measure the level of cooperativeness of neighbors and to discover possible intruders among them. General applicability of the proposed solution is shown through examples of multiagent systems from Biology and from Robotics.