Distributing intelligent functionalities in the Internet of Things with agents and Digital Twins

Autonomous Agents (AAs) and Digital Twins (DTs) are two widely used abstractions in the literature about the engineering of “intelligent” Internet of Things (IoT) systems and applications. However, their role can be considered partially overlapping given the fragmented landscape of approaches emerging from the literature. There, in fact, either AAs or DTs, or their combination, are sometimes used for achieving the same goals. In this paper, we attempt to clarify similarities and differences of these abstractions and argue that the choice to use AAs or DTs (or an integration of the two) should stem from a principled analysis of the IoT system requirements. That is, by matching the desired intelligent functionalities with the properties of the two abstractions to find the most appropriate one. Accordingly, we (i) analyse the state-of-the-art approaches to identify how AAs and DTs are currently used to encapsulate and distribute intelligent functionalities across IoT system components; (ii) propose a set of principles to assist designers in choosing the most suitable abstraction for a given functionality; and (iii) discuss exemplary architectures that may arise from applying such principles. To conceptually validate our contribution, we analyse the practical case of an intelligent manufacturing system and show how following the outlined principles leads to interesting properties in the final system design.