Article Preview
TopIntroduction
Nowadays, the software engineering of Ambient Intelligence (AmI) systems is widely investigated. The need comes from new applications that aim at taking profit from the ubiquitous computing, often involving the human life in its all, like in the development of smart and assisting environments (Nakashima, Aghajan, & Augusto, 2010) (Poland, Nugent, Wang, & Chen, 2009). Actually, such systems have more and more impact over people's lifestyles (Augusto, 2007), (Preuveneers & Novais, 2012) (Dingli, Attard, & Mamo, 2012). As some major challenges, the building of intelligent systems requires the adaptation and sustainability to the context, with regard to the open, non-deterministic and uncertain environment. To deal with these features, different frameworks and systems providing context-awareness pervasive services are widely investigated within multimodal abilities (Meier & Lee, 2009) (Lee, Lunney, Curran, & Santos, 2009).
However, context-awareness requires specific models and powerful representation of context information. In (Chen & Kotz, 2000), four types qualifying the context are defined: (1) computational context – available resources, network quality and related information; (2) user context – profile of the user, people nearby, social situation; (3) physical context – lighting, temperature, traffic conditions, noise levels, etc., and (4) time context – time of day, date of the year. As two other types of contexts, the work of (Henricksen, Karen & Indulska, Jadwiga, 2006) introduces the activity of some agent-assisted users and the work of (Chen & Kotz, 2000) deals with history context.
On another side, several modeling approaches are already proposed to assist the designer of AmI systems (Hani Hagras, 2004) (Hegarty, Lunney, Curran, & Mulvenna, 2010) (Najjar, Courtemanche, Hamam, Dion, & Bauchet, 2009). In fact, the major problem for the system entities consists in recognizing environmental context, including location modeling, resource management, real-time requirements, planning within time limited, and discovery of neighbors e.g. (Cook, Youngblood, & Das, 2006), (de Silva, Dekker, & Harland, 2007). In addition, most context modeling approaches only focus on a subset of the above types of context, depending on the requirements of specific applications e.g. (Guivarch, Camps, & Péninou, 2014) (Domnitcheva, 2000), (Hu & Lee, 2004).
Usually, multi-agent systems (MAS) approaches offer interesting and powerful framework for the design of AmI systems, since their agents are considered as autonomous, intelligent, proactive and flexible (Bordini, et al., 2006). Thanks to their mental attitudes, the Belief-Desire-Intention (BDI) agents of (Rao & Georgeff, 1995) are designed to take rational decisions as practical reasoning. At the operational level, intentions correspond to plans of actions to be performed. However, this is not enough when conditions are highly dynamic like in AmI systems. This assumes an efficient context-awareness ability, dealing with unexpected situations and changes of context which define an ambient agent.
Ambient Agents are dynamic entities changing their intentions and then their plans, throughout their evolutions. Then, to recognize the environment contexts, agents are equipped with efficient context-awareness mechanism. In addition, planning modules are embedded in agents for self-adaption of behavior, in respect to changes of the environment context even user preferences (e.g. (Nunes & Luck, 2014)).