Integration of distributed autonomous systems through a multi-layer, IoT-based hierachical approach

The huge number of IoT devices used nowadays for the various applications and environments introduces the problem of interoperability of heterogeneous platforms and software integration. Furthermore, the integration of heterogeneous platforms is complex and time consuming. In this disseration I tackle this problem by presenting the design of an approach aimed at providing a software platform called JarvSis. JarvSis is a distributed scheduler capable to automate the execution of multiple heterogeneous tasks on di erent applications by means of a modular and adaptable software architecture based on micro-services and nodes. JarvSis is capable to interact by design with any devices, from a complex robot platform to a simple sensor, that expose heterogeneous remote interfaces, e.g. web-api or MQTT message capabilities. The proposed approach is capable to automate the con guration and deploying of work ows of IoT related activities, by composing heterogeneous tasks organized in clusters, from the Cloud, to the Fog, to the smart devices running on the \ground". The tasks are organized in a hierarchical network on which Fog/Edge resources are used as a bridge between the computational resources hosted in the Cloud, and the devices operating on the \ground". In this con guration the top layers will provide control and coordination, while the bottom one is distributed among the Fog/Edge resources. As a proof-of-concept of Jarv- Sis and its features, two detailed examples are provided in this dissertation: the former is related to the integration of IoT devices, the latter is related to robotic domain applications. In order to simulate the di erent scenarios and the devices, JarvSis has been coupled with a multi-agent simulator called AgentSimJs. AgentSimJs is a JavaScript simulation framework aimed at designing software simulations that can directly run on a Web browser. AgentSimJs is composed by a modular architecture made by a number of di erent modules to provide a set of exible primitives to write all the parts of an agent behavior. Through AgentSimJs a simple and fast implementation of the aspects related to communication, motion, and group formation can be performed. Finally AgentSimJs includes the required capabilities to render a 3D scene with objects and agents, as well as to distribute the simulation among di erent machines and/or di erent threads.