A Middleware for Ubiquitous Agents

There is a paucity of pervasive middleware that allow integrations between Artificial Intelligence (AI) of the kind supported by software agents and sensor/actuator networks (SAN) while simultaneously featuring all of the following characteristics: a) a systematic way to achieve the integrations, b) transparency, c) multiple MAS (Multi-Agent System) support and d) sensor, actuator and device heterogeneity. The thesis is a step forward to developing pervasive middleware for AmI (Ambient Intelligence) by presenting an approach with all the above characteristics. We propose a middleware that creates ubiquitous agents (UAs) by embedding software agents in the physical world as part of a ubiquitous computing environment. We use intelligent software agents residing in a multi-agent system (MAS) environment. UAs are built through linking the functionality of agents residing in MAS networked environments, to aggregations of sensors, actuators and devices in the physical world that we call avatars. The software agents consume services provided by physical sensors, actuators and SAN. The provider-consumer relationships enable agent functionality to access the data that is sensed by physical sensors and to also create effects in the physical world via physical actuators. Computationally expensive capabilities such as decision making and communication are performed by the agents in MAS platforms while the acting and the sensing in the physical world through their corresponding avatars. Our approach follows SOA principles to implement a message oriented middleware that architecturally consists of: a) a base-layer enabling the sensors and the actuators to register as service providers and the agents to register as service consumers using an API (Application Programming Interface) and b) a reflection layer that creates models of agents and avatars using registration metadata from the base-layer and uses these models to create and manage UA functionality. Furthermore, the UA framework uses the Z-notation for a detailed specification of every component of the middleware enabling researchers to implement it using different technologies. The eVATAR middleware that was developed in this thesis is such an implementation. eVATAR was applied on two scenarios for smart homes that use a MAS platform. Firstly we applied agent AI from the GOLEM and the JADE MAS on a miniature smart home containing real sensors and actuators in order to provide with evidence that the proposed approach is systematic, transparent and supports device heterogeneity and multiple MAS. Then we used a custom smart home simulation to illustrate the potential of a system using MAS agents, eVATAR and a sensor/actuator network embedded in a home context for becoming useful in confronting everyday lives problems. The thesis also includes a performance evaluation of eVATAR and discusses latency and how to reduce it. As future work we explore ways to improve our approach and to extend the scope of supported devices.