Agent-Based AmI System Case Study: The Easy Line + Project

8
Agent-Based AmI System case study: the Easy Line + Project Armando Roy Delgado, Rubén Blasco, Álvaro Marco, Diego Cirujano, Roberto Casas, Armando Roy Yarza, Vic Grout, Richard Picking Grupo Tecnodiscap, Universidad de Zaragoza, Spain {aroy; rblasco; amarco; ciruman; rcasas; armanroy}@unizar.es & Centre for Applied Internet Research (CAIR), Glyndwr University. North Wales {v.grout; r.picking}@glyndwr.ac.uk Abstract. Smart environments include context-aware computing to enhance system capa- bilities which affect user interaction positively. A context-aware application should collect contextual information through different ways depending on technology availability and handle this information adequately to execute actions or warnings, present information or modify the environment. Ambient intelligence (AmI) habitually means system complexity and heterogeneity. This paper describes an AmI system case-study, developed for the Easy Line + project, a decentralized intelligence carefully structured in agents grouped by roles in domains to improve system package interaction and integration. The main purposes for this implementation are: avoid complexity and inconsistence between the elements of the system and increase system´s performance. The Easy Line + project is primarily oriented to monitor and control white goods and to assist in emergency situations though it is deployed all over the home environment and can be extended, more easily thanks to the multi-agent approach, to control and monitor other devices or services at home. Keyworks Ambient intelligence, context-awareness, domain, agent. 1 Introduction The application of information technologies to home environments is mainly characterized by the integration of networked computational devices into the phys- ical context. The concept is also denominated with the term ubiquitous computing, which involves ubiquitous communications and user interfaces. AmI implies the existence of another abstraction level on top meaning adaptation, personalization, unobtrusiveness and anticipationexpressions that commonly cope with other AmI projects [1][2][3] also oriented to social inclusion. Context-sensitive systems must be flexible enough to perform according to the requirements independently from the architecture, location, user profile and the

Transcript of Agent-Based AmI System Case Study: The Easy Line + Project

Agent-Based AmI System case study: the Easy Line +

Project

Armando Roy Delgado, Rubén Blasco, Álvaro Marco, Diego Cirujano, Roberto

Casas, Armando Roy Yarza, Vic Grout, Richard Picking

Grupo Tecnodiscap, Universidad de Zaragoza, Spain

{aroy; rblasco; amarco; ciruman; rcasas; armanroy}@unizar.es

&

Centre for Applied Internet Research (CAIR), Glyndwr University. North Wales

{v.grout; r.picking}@glyndwr.ac.uk

Abstract. Smart environments include context-aware computing to enhance system capa-

bilities which affect user interaction positively. A context-aware application should collect

contextual information through different ways depending on technology availability and

handle this information adequately to execute actions or warnings, present information or

modify the environment. Ambient intelligence (AmI) habitually means system complexity

and heterogeneity. This paper describes an AmI system case-study, developed for the Easy

Line + project, a decentralized intelligence carefully structured in agents grouped by roles

in domains to improve system package interaction and integration. The main purposes for

this implementation are: avoid complexity and inconsistence between the elements of the

system and increase system´s performance. The Easy Line + project is primarily oriented to

monitor and control white goods and to assist in emergency situations though it is deployed

all over the home environment and can be extended, more easily thanks to the multi-agent

approach, to control and monitor other devices or services at home.

Keyworks Ambient intelligence, context-awareness, domain, agent.

1 Introduction

The application of information technologies to home environments is mainly

characterized by the integration of networked computational devices into the phys-

ical context. The concept is also denominated with the term ubiquitous computing,

which involves ubiquitous communications and user interfaces. AmI implies the

existence of another abstraction level on top meaning adaptation, personalization,

unobtrusiveness and anticipation—expressions that commonly cope with other

AmI projects [1][2][3] also oriented to social inclusion—.

Context-sensitive systems must be flexible enough to perform according to the

requirements independently from the architecture, location, user profile and the

2

technologies involved. Furthermore they should reconfigure dynamically to ac-

commodate users’ needs, taking into account a wide range of users and situations.

To develop a system with such characteristics there must be processing units

distributed all over the physical context, as well as wearable devices. Ubiquitous

computing enables these objects to communicate with each other and the user

transferring data from one point to another dynamically as required. A wide range

of devices can be used to computerize homes like cameras, location systems and a

variety of sensors, whose function is to collect contextual information. Aside from

data collecting devices AmI systems include a variety of human-machine interfac-

es (HMI) such as displays, touch screens, remote controls, TV, hand held devices

and so.

Finally networking devices have to be incorporated to put everything together

and some others to act upon control and coordinate everything—e.g. computes or

servers—. The type and number of these devices may vary according to the place

and needs of the user. Altogether, the integration of networked computational de-

vices into the targeted environment gives the possibility to reduce the issues that

nowadays arise about the risk of people losing independence at home and permits

to lower the levels of social exclusion in the society that are getting higher as pre-

dicted years ago [4].

To increase people autonomy at home is one of the main objectives of the Easy

Line + project [5]. Easy Line + project aims to allow elderly people to remain in

their homes living independently thanks to AmI, by facilitating realization of daily

tasks such as preparing foods or washing clothes, detecting risky or abnormal situ-

ations, and evaluating their quality of life in that scenario.

In this paper, we firstly make a constructive analysis about the advantages of

multi-agent systems in AmI. After that, there is a generic description of the ele-

ments of the system grouped by domains, explaining the main tasks of each do-

main and the interactions between them, and the enumeration of the agents in-

volved in each domain and their relationships. Finally, conclusions are presented.

2 Reasons to apply a muti-agent system approach in AmI

Distributing the intelligence makes the system easier to adapt and more scala-

ble. In other words AmI systems have to be able to add extra functionalities, per-

haps overlooked in the first instance, to cope with user requirements over time.

In contrast, a centralized intelligence in AmI is predisposed to fail because it

often implies risks that are not feasible for home environments where people

spend most of their time. There might be a lack of processing power, memory

bottlenecks or a loss of centralized data that may affect the performance of the

system making it unstable or unsuitable to fulfill user needs by that moment. In

addition, in a centralized system, a network failure or a faulty networked device

3

may collectively affect the rest of the network as some researchers stated like Park

et al. [6].

A Multi-agent system perspective has numerous advantages, almost certainly

the most important related to AmI systems are:

Monitoring and control versatility

Better resource allocation

Facilitate system design

Allow modularity and flexibility

Robust behavior on automated processes

Many communication protocols, network coordinators and a flexible device

manager make the system more versatile and therefore more capable to influence

the life of the user positively. In addition, having agents in diverse physical

locations requires less processing power and memory per device minimizing

bottleneck incidents. Furthermore if all agents have specific roles that can be

integrated with the rest of the system, the result is a modular architecture flexible

enough to accept changes in certain modules and/or to add new ones. On top of

that, the system can contain some reactive agents—automatons that receive an

input, processes it and produce an output—which perform better in certain

situations as they are unaffected by other system processes but are still integrated

with the rest of elements.

However developers have to bear in mind that the application of multi-agent

systems to AmI cannot be reduced to pursue those universal benefits as user

requirements tend to be more explicit. The following sections enter into more

detail about how Easy Line + is going to affect the lives of the users and which

multi-agent architecture is selected to adjust the system the user needs.

3 Systematization of the Easy Line + project into agent domains

A multi-agent system consists of some individual agents each of them oriented to

carry out specific tasks. Collectively, a set of agents are able to achieve tasks that

will be rather impossible to carry out in a centralized system having only a single

agent. The use of multi-agent systems has increased considerably during the past

few years, nowadays is a common term constantly used in context-awareness sys-

tems and service oriented architectures [7][8][9].

In the Easy Line + project, we seek to obtain the benefits of using a multi-agent

architecture explained in the previous section. In addition, we aim to accomplish

some other specific goals: context data mining; reusability and management of

data; user activity monitoring; non-intrusive system responses; learn user

capabilities; content adaptation [10][11]; user profile identification; and

accessible; personalized and coordinated HMI

To complete these tasks, the system has to incorporate many devices and

employ different network technologies; it also requires an assorted set of

4

software/hardware specifications/implementations and to customize or tailor the

information according to the user's preferences [11].

For that reason it is significant to classify agents in different groups or domains

according to their roles and characteristics to increase specialization and achieve a

better performance. Besides, the interoperability of all the domains has to be

consistent and fluent to minimize system latency. We have identified five domains

in the Easy Line + project as shown in figure 1.

Fig. 1. Agent Domains of the Easy Line + project.

HMI domain will be the one with closer contact with the user. It includes all

agents related to the interfaces: physical devices, personalization of interfaces,

human-machine interaction accessibility and HMI interface coordina-

tion/monitoring.

Context Management domain. Information about the context is important, the

more the system knows the better will acclimatize the environment to the user.

We include RFId information, environmental sensors, and the status of ap-

pliances, each white good must be able to inform about its status and also be

commanded remotely. RFId is integrated in the system to recognize food, gar-

ments and medicines in the environment (informing about food about expire,

recommended program for the laundry, etc.). Each technology needs to be inte-

grated involving a set of agents such as driver agents, communication agents,

context managing agents, networked devices (not HMI) etc.

Logic domain. There must be a set of agents controlling system´s logic. In a

way, this domain could be the glue which put the other domains together al-

5

though bearing in mind this is not entirely strict. Logic agents are in charge of

understanding predefined situations (or scenarios) and support the rest of the

system to act upon them.

Data Storage domain copes with distribution, classification and storage of da-

ta among the databases.

Quality of life evaluation domain (QoLE) is mainly oriented to evaluate us-

er´s behaviour, it involves neural networks, professionals to evaluate the results

of the neural networks, and system configurations according to the patterns en-

countered in the neural networks.

4 Agent specifications in the Easy Line + project

Once domains are sorted out and their roles established, each domain is broken

down into tasks and services. This part is critical in order to define the agents of

each domain. Each definition must include the processes each agent is capable of,

which data is going to be handled and to which other agents is going to be for-

warded to. In the end the resulted modularity has to be consistent and smooth, in

other words, agents do not have to wait long for other agents to finish their tasks

or do not have to be affected much about the errors other agents could have.

Moreover the agents have to be adjustable and/or extendible in case other agents

are introduced in the system and/or specifications are re-defined.

Fig. 2. Agents and their relationships in Easy Line +.

6

The system, as it is at the moment, has several definite agents most of them prede-

fined according to the system requirements established initially. The resulting hy-

brid architecture tries to combine deliberative and reactive aspects by combining

deliberative and reactive agents [12]. This architecture of agents is a good orienta-

tion to developers to find the proper programming model which, as Ricci et al.

said [13], is a key point to build a complex and modular system.

As there are too many agent specifications to make an extensive description of

them, we will enumerate and describe briefly some of them, emphasizing their key

characteristics and relationships with the other agents in the system (see figure 2).

HMI Device agents. Several types of clients can be used to manage the sys-

tem: mobile devices such as PDA, smart phone, wearable device, ultra-mobile

PC, touch screen device etc; and fixed devices such as computers, digital TV

and fixed screens that will be used as a centralized control.

HMI Controller agents integrate each HMI device with the system using a bi-

directional communication with the device and the HMI part of the database. It

does not connect directly with the logic domain agents, while this communica-

tion goes through the database and therefore remains registered to keep track of

timestamps and any data handled in all user interactions through any HMI.

Actions Driver agent processes data from the core agent regarding user inte-

ractions or automated controls, and forwards actions to the device manager

agent which will inform the driver about which tasks have to be carried out in

the network(s). It also keeps track of the changes by telling the database logger

to store environmental information in the context database.

Events Trigger agent is in charge of inform the logic domain about relevant

changes that take place in the environment. Afterwards the logic domain will

have to decide which ones have to be notified to the user and how.

Device manager agent creates a semantic representation of the devices: ap-

pliances, sensors, etc. This representation is done using the OSGi standard.

This approach will permit other implementations to include these devices and

make use of their features easily. OSGi framework is a good solution to man-

age the information provided for the devices. It makes easy to keep separate the

Hardware issues and requirements of the storage of data provided for them.

Driver agents are devoted to gateway control, device registration and device

publication. Published devices will be accessible by the logic domain an conse-

quently by the users. There are three driver agents developed at the moment to

fulfill the specific needs of the EasyLine+ system: PLC, ZigBee and Infrared

drivers.

The Core Agent will communicate with HMI and parse a predefined XML

message file. This module will be responsible to send the correct information to

the HMI controller agent according to the user profile and choices. Moreover it

will have to save user data into the HMI database and the QoLE database

agents, this last one to be useful to the neural network agents.

Scenarios handler agent analyzes the HMI database agent (every few millise-

conds) looking for situations or scenarios that require intervention from the sys-

7

tem. When a situation requires direct user interaction, for example when the

fridge’s door is open for more than a given time, when the washing machine

program is inappropriate for the garments in, etc, then it informs the core agent

which will communicate with the HMI Controller agent. These situations are

―fixed rules‖, which means that it is known what the system will do in a given

(identified) context.

Event handler agent. Any significant changes in appliances or sensors that the

Context Manager writes into the data base are sent, in real time, to the Event

handler agent. Then the core agent decides if the user must be informed

through the HMI domain or not.

The QoLE System agent analyzes logged data from the context looking for

patterns that might indicate relevant behavioral changes. For example, more

frequent oversights, capacities diminishing, etc. This is done with neural net-

works, and the rules are dynamically modified by the learning capability of the

neural network algorithms.

Information for QoLE (caretaker). Since system adaptation to user´s beha-

vior cannot be left to neural networks only, a professional has to consider if the

results are realistic. For example, there may be decisions such as: some people

are absent–minded and usually forget about closing the fridge’s door; this

doesn’t indicate low cognitive level, but if a person that never forgets about it,

starts forgetting closing the door, might start having cognitive problems.

Profiler agent: Users have complete profiles constructed with their capabilities

(cognitive, audible, visual, user level with new technologies, etc.). If there are

recognizable changes in the user´s profile this agent will update the profile and

inform the logic domain about it.

5 Conclusions and future work

The simplification of the system due to the applied multi-agent architecture

proved to be valuable. Nevertheless in Easy Line + the added value of applying

user profiles to adjust the system and the inclusion of neural networks to identify

human behavior patterns involves a higher complexity, a drawback that involves

more risks that has to be treated carefully to avoid intrusiveness and dangerous

situations.

Thanks to the agent classification by domains, all system elements could be

finely tuned according to their specific functionalities and requirements without

worrying too much about the influence on other elements. In addition, having dis-

tributed databases, processing power and memory greatly reduced the apparition

of bottlenecks allowing the system to perform tasks parallel in different domains;

making it less suitable to un-expected overloads.

To sum up, although the inclusion of several heterogeneous technologies makes

the system harder to develop, it is clear that the modularization of the system into

agents leads to the successful realization of the project.

8

Acknowledgments This work was partially supported by the EU under MonAMI

(IST-5-035147) and Easy Line + (IST-045515) projects and by the Spanish Minis-

try of Science and Technology under the AmbienNET Project (TIN2006-15617-

C03-02).

References

[1] Emiliani P. L., Stephanidis C., Universal access to ambient intelligence environments: Op-

portunities and challenges for people with disabilities. IBM Systems Journal 44 (3): 605-

620 (2005).

[2] Gonçalves D., ―Ubiquitous Computing and AI towards an Inclusive Society,‖ EC/NSF

Workshop on Universal Accessibility of Ubiquitous Computing: Providing for the Elderly,

Alcácer do Sal, Portugal (2001).

[3] Abascal J., Ambient Intelligence for people with disabilities and elderly people. SIGCHI

Workshop on Ambient Intelligence for Scientific Discovery (AISD) Vienna. (2004).

[4] Aichholzer G 2001, Report on changes and trends in service delivery. IST PRISMA project

(IST-1999-29088) Deliverable D2.1

[5] EASY LINE +: Low Cost Advanced White Goods for a Longer Independent Life of Elderly

People (2007-2010), [http://www.easylineplus.com/].

[6] Park S.B., Lee W.H., Park B.Y., Ahn S. J., Chung J.W., An Agent System for Recovering

Disabilities in the Ubiquitous Environment. International Journal of Computer Science and

Network Security Vol. 7, No. 8, pp. 128-132. (2007).

[7] Velasco, J. R.; Marsá-Maestre, I. Navarro, A.; Lopez, M.A,; Vicente, A.J.; De La Hoz,

E.;Paricio, A.; Machuca, M. ―Location-aware services and interfaces in smart homes using

multiagent systems‖. Proc. 2005 Int. Conferecence on Pervasive Systems and Computing.

PSC’05. Las Vegas, Nevada, USA June 27-30. ISBN 1-932415-52-1. pp 104-110. (2005).

[8] Neelima A. V., Sunitha B. R., and Aghila C. G. A Multi Agent, Service Reassembling Ar-

chitecture for Context-Aware Systems International Journal of Recent Trends in Engineer-

ing, Vol. 1, No. 1, May (2009).

[9] Gloss L., Mason B., McDowall J., "Multi-agent System Service and Ontology Design

Methods," cecandeee, pp.453-456, 2008 10th IEEE Conference on E-Commerce Technol-

ogy and the Fifth IEEE Conference on Enterprise Computing, E-Commerce and E-Services,

(2008)

[10] Moraïtis, P.; Petraki. E.; Spanoudakis, N. I. ―Engineering JADE Agents with the Gaia

Methodology.‖ Dep. of Computer Science, University of Cyprus, Springer Berlin,pp. 77-

91. (2003).

[11] Berhe, G.; Brunie, L.; Pierson, J. M. ―Modeling Service-Based Multimedia Content Adap-

tation in Pervasive Computing.‖ ACM CF’04, ISBN:1-58113-741-9, pp. 60-69, Italy, April.

(2004).

[12] Carrascosa C., Bajo J., Julian V., Corchado J.M. and Botti V.. Hybrid multi-agent architec-

ture as a real-time problem-solving model. Expert Systems with Applications, 34(1):2-17,

(2008).

[13] Ricci A., Buda C., and Zaghini N.. An agent-oriented programming model for SOA & web

services. In Proc.of 5th IEEE International Conference on Industrial Informatics

(INDIN'07), pages 1059-1064, (2007).