AMBIENCE & COLLABORATION

Embedded Agents in a Human-centered World

M. J. O’Grady, G. M. P. O’Hare, R. Tynan

CLARITY- Centre for Sensor Web Technologies, School of Computer Science & Informatics, University College Dublin

Belfield, Dublin 4, Ireland

R. W. Collier, C. Muldoon

School of Computer Science & Informatics, University College Dublin, Belfield, Dublin 4, Ireland

Keywords: Ambient Intelligence, Embedded agents, Human-centered computing.

Abstract: Supporting people in the pursuit of their everyday activities is a laudable objective and one which

researchers in various disciplines including computing, actively seek to accomplish. The dynamic nature of

the end-user community, the environments in which they operate, and the multiplicity of tasks in which they

engage in, all seem to conspire against the desired objective of providing services to the end-user

community in a transparent, intuitive and context-aware fashion. Indeed, this inherent complexity raises

fundamental problems for software engineers as they frequently lack the tools to effectively model the

various scenarios that dynamic user behaviour give rise to. This difficulty is not limited to exotic

applications or services; rather, it is characteristic of situations where a number of factors must be identified,

interpreted, and reconciled such that an accurate model of the prevailing situation at a given moment in time

can be constructed. Only in this way, can services be delivered that take into account the prevailing human,

social, environmental and technological conditions. Constructing such services calls for a software solution

that exhibits, amongst others, diffusion, autonomy, cooperation and intelligence. In this paper, the potential

of embedded agents for realising such solutions is explored.

1 INTRODUCTION

Computing is a multi-faceted concept and is

perceived in a number of diverse ways. Quite how it

is perceived is influenced by a number of factors but

one prominent issue concerns the role and

prominence that computing plays in a persons'

everyday life. For some, designing, implementing

and administrating computing systems is their

everyday task. For others, computing forms an

indispensable tool in the fulfilment of their work.

However, for a significant number, software

applications and services have a peripheral role in

their lives. Why this is the case must remain a matter

of conjecture. However, it does indicate that a

significant opportunity exist for harnessing

computational resources for the benefit of people in

the pursuit of their everyday lives – Ambient

Assisted Living (Nehmer at al, 2006) being a

particular case in point. To avail of this opportunity

it beholds software professionals to demonstrate

clear and tangible benefits if their technological

solutions are to be adopted. Financial considerations,

though important, are not paramount. Moreover, the

learning curve must be short and the services

themselves easy and intuitive to use. Achieving

these objectives is, of course, a challenging and

formidable task.

In focusing on the technological, there is a

significant risk that the desires and requirements of

people may be neglected. For example, people may

be perceived as black-boxes - objects about which

very little is known about, and the circumstances in

which they operate inscrutable. Alternatively, people

may be treated as a homogenous entity and ascribed

a worldview remarkably similar to the designers of a

piece of hardware or software. In either case, people

lose. Clearly, focusing solely on the technology is a

deficient approach, and to remedy this, it is

necessary to obtain a broader picture of the

361

J. O’Grady M., M.P. O’Hare G., Tynan R., W. Collier R. and Muldoon C. (2009).

AMBIENCE & COLLABORATION - Embedded Agents in a Human-centered World.

In Proceedings of the 4th International Conference on Software and Data Technologies, pages 361-364

DOI: 10.5220/0002281503610364

 SciTePress

prevailing situation at the time people are actively

using applications and services or, in other words,

their context (Greenberg, 2001).

2 A QUESTION OF CONTEXT

Before a context can be incorporated into the design

of a service, it is necessary to articulate the elements

of context that the services in question require. It

must then be ascertained how the necessary

contextual elements can be captured. In some cases,

this may be accomplished by asking the user in

question. In other cases, the application may be able

to dynamically determine the device parameters

under which it is operating and adapt accordingly,

displaying some limited autonomic behaviour in the

process (Kephart and Chess, 2003). Finally,

individual contextual elements may be determined

using suitably equipped sensors and Wireless Sensor

Networks (WSNs). Assuming that the required

contextual elements are available, software

designers may need to consider if the cost, both

computationally and financially, of capturing

context is worth the effort. Of particular relevance to

this discussion is the case where individual

contextual elements must be explicitly and

continuously monitored such that a model of user

behaviour may constructed that would enable the

identification of scenarios where proactive

intervention would aid the user in the

accomplishment of the task at hand. Such an

intervention would only be possible after a

significant quantity of context-related data had been

captured, filtered and characterised. Identifying

patterns in individual context data streams, and

reconciling those with patterns identified from other

context data streams, as well as patterns from

previous behaviour models, is a computationally-

intensive task and suggests the need for harnessing

intelligent techniques.

3 STRATEGIES FOR REALISING

INTELLIGENCE

An array of Artificial Intelligence (AI) techniques

exists, and depending on the requirements of the

services in question, any individual or combination

of techniques may be harnessed as appropriate.

However, the use of such techniques incurs a cost

which may be measured in at least two ways. Firstly,

AI techniques are notoriously computationally

intensive. Thus the immediate availability of

powerful workstations is a prerequisite. However, if

an interconnecting network that incorporate a

wireless component is utilised, there may be

problems due to latency and poor data throughput.

This leads directly to the second cost penalty which

concerns the perceived response time and,

ultimately, the usability of the service in question.

Recall that the primary objective is to incorporate

proactive and anticipatory behaviour for aiding users

in the fulfilment of their everyday tasks. A small

window of opportunity exists where such behaviour

can be fruitfully activated. Hence, the response time

to changes in the various contextual elements must

be such that these opportunities can be availed of in

a timely fashion.

It is interesting to reflect at this juncture that the

Ambient Intelligence (AmI) (Vasilakos and Pedrycz,

2006) initiative, as its name suggests, envisages the

incorporation of intelligent techniques. To recall:

AmI was formulated in response to anticipated

difficulties arising in scenarios where ubiquitous

computing environments were deployed. Such

environments would incorporate a significant

number of artefacts or objects that had been

augmented with computational technology. As such

artefacts increasingly proliferate, a scenario was

envisaged arising where competition for the user's

attention would be at a premium. The net result

being that such environments would be unusable and

inhospitable to users - a most unfortunate state of

affairs when the original objectives are considered.

Thus AmI anticipates the use of intelligent

techniques, especially Intelligent User Interfaces

(Maybury and Wahlster, 1998), as a means of

managing interaction within the environment and

minimising the need for explicit intervention. For

the purposes of this discussion, we are interested in

the potential of AI for identifying opportunities

where the user may be opportunistically aided in the

fulfilment of their tasks. AmI complements this

objective but does not seek to fulfil it.

To develop and maintain an accurate model of

user behaviour, it is essential to harness as much

contextual data as possible, using an array of sensors

and other devices. Ideally, these data streams would

be processed as near to their source as possible,

implying the need to employ a Distributed Artificial

Intelligence (DAI) or agent-based approach. In the

later case, one practical implementation for

resource-bounded devices is that of embedded

agents.

ICSOFT 2009 - 4th International Conference on Software and Data Technologies

362

3.1 Embedded Agents

Historically, agents have been deployed on fixed

networked nodes where access to computational

resources is not a problem. However, developments

in mobile and embedded technologies, both

hardware and software related, are extending the

reach of agent technologies towards the extreme

periphery of the network. Indeed, examples of

embedded agents for AmI applications have been

documented (Keegan et al, 2008; O’Grady et al,

2005; Hagras et al, 2004). The current range of

PDAs and smartphones are capable of running

multi-tasking and multi-threaded applications.

Admittedly, the screen size of ¼ VGA is one

obvious limitation but it is a significant

improvement over the screens that supported a few

lines of text that were state-of-the art just a few short

years ago. In addition, a number of popular

communications protocols are supported. However,

it is developments in tools and languages for

designing and developing software for mobile

devices that has had the most impact. In particular, a

significant number of mobile devices comprise a

Java Virtual Machine (JVM) compliant with the

Java ME specification.

Embedding agents (Figure 1) on sensor devices

may appear impractical initially but recent

developments suggest otherwise. A number of

sophisticated sensor platforms are commercially

available whose specifications are similar to many

PDAs and mobile phones, albeit not from a GUI

perspective. A JVM is available for such platforms

thus enabling them to support relatively

sophisticated applications, either in isolation or,

more likely, as a node in a distributed application or

service. In a standard Wireless Sensor Network

(WSN) topology, sophisticated sensors could act as

base stations, collecting data from nearby leaf nodes,

sharing it with other nodes and collaborating with

them to make sense of the data.

From a software platform perspective, a number

of embedded agent platforms have been developed

in the past number of years. O’Hare et al (2006)

provides a useful survey of some common

platforms. One particular example of a system that

demonstrates the feasibility of agents on WSNs is

Agilla (Kok et al, 2005). This is a middleware

solution that adopts mobile agents for coordination

and migration in the fulfilment of WSN specific

tasks.

Figure 1: Embedded agents can be deployed on suitably

equipped devices ranging from workstations to smart

phones to embedded sensors in everyday objects. In this

way, the necessary contextual elements can be captured

and interpreted, enabling the construction of sophisticated

behaviour models, and facilitating the identification of

opportunities where proactive and anticipatory activities

by applications and services can be enabled.

4 REALISING AN AMBIENT

SOLUTION

Having discussed the broad nature of embedded

agents, it is now appropriate to revisit the original

motivation for this discussion, namely how to

monitor users so as to anticipate their needs and

proactively help them in the course of their everyday

activities. The solution to this problem does not

solely manifest itself either in an exclusively

hardware or software solution. Both have an

indispensable role to play in fulfilling this objective.

WSN technologies represent a significant step

forward in the pursuit of the ubiquitous computing

vision. However, it must be acknowledged that there

are significant technical obstacles outstanding before

the transparent integration of computational

elements into everyday objects becomes a reality.

Power management is one critical issue that springs

to mind.

Traditionally, it has been the software element

that has been lagging behind hardware

developments. However, the increasing

incorporation of runtime environments such as Java

ME into both mobile devices and embedded

technologies is giving rise to a situation where, for

the first time, a common platform is available to

software developers. Again, it must be

acknowledged that there are limitations and

incompatibly issues amongst others; nevertheless,

the vision of various heterogeneous hardware

elements all sharing a common software element is

beginning to crystallise. Thus the foundations are

AMBIENCE & COLLABORATION - Embedded Agents in a Human-centered World

363

laid for realising software services that are

inherently distributed yet adaptable to the prevailing

context at the time of invocation.

Now that both the hardware and software

foundations are in place to begin constructing

services that can harness and interpret various

aspects of the user's context, the question arises as to

how best to engineer such services. Based on the

previous discussions, it can be seem that the

embedded agent paradigm is a particularly apt one

as agents incorporate a significant number of

features that can be fruitfully harvested to deliver the

necessary adaptivity. This is not to say that it is the

only approach or the best approach. It is just an

acknowledgment that, at this moment in time at

least, embedded agents singularly possess some of

the essential characteristics necessary for collecting

and interpreting the necessary contextual elements

essential to the provision of human-centered

services.

To reflect further on some of the more pertinent

agent characteristics: autonomy and reactivity are

essential to the continuous monitoring of contextual

cues. Collaboration is essential for the integration of

the contextual cues and the construction of a model

of the user's world. Intelligence is necessary for

interpreting the meaning of the collated contextual

cues and the construction of models of past

behaviour that can be used to predict likely future

actions. Agents can then proactively use those

models to anticipate and pre-empt user requests.

Finally, agents are inherently distributed software

entities. This makes them ideal for implementing

solutions that must harness data from numerous

diverse sources, interpreted it in an intelligent and

collaborative manner, and collate the results such

that an accurate model of the prevailing context at

any given time may be constructed. Only in this way

can sophisticated behaviour models be constructed,

patterns of behaviour identified and future activities

predicted, paving the way for the delivery of truly

adaptive human centered applications and services.

5 CONCLUSIONS

Embedded agents offer one vision of how disparate

data sources may be captured and interpreted to

realize services in a range of applications that are

human-centric. Their inherent characteristics make

them a particularly apt solution for modelling such

services. Ongoing developments in WSN

technologies are continuously extended the number

of platforms on which such agents can be

realistically deployed. Yet many challenges remain,

including identifying effective strategies for data and

decision fusion such that contexts and tasks can be

recognised within a time frame that allows effective

responses and interventions.

ACKNOWLEDGEMENTS

This material is based upon works supported by the

Science Foundation Ireland (SFI) under Grant No.

(Grant No. 07/CE/I1147).

REFERENCES

Fok, C., Roman, G., and Lu, C. (2005) Mobile agent

middleware for sensor networks: an application case

study. In: Proceedings of the 4th international

Symposium on information Processing in Sensor

Networks, Los Angeles, California, April 2005.

Piscataway: IEEE Press, pp. 382- 387.

Greenberg, S. (2001) Context as a dynamic construct.

Hum.-Comput. Interact. 16 (2): 257-268.

Hagras, H., Callaghan, V., Colley, M., Clarke, G., Pounds-

Cornish, A., and Duman, H. (2004) Creating an

Ambient-Intelligence Environment Using Embedded

Agents. IEEE Intelligent Systems 19 (6):12-20.

Keegan, S. O'Hare, G.M.P., O'Grady, M.J., (2008)

EasiShop: Ambient Intelligence Assists Everyday

Shopping. Information Sciences, 178 (3):588-611

Kephart, J. O. and Chess, D. M. (2003) ‘The Vision of

Autonomic Computing’. Computer 36 (1): 41-50.

Maybury, M.T. and Wahlster, W. (Eds.). (1998) Readings

in Intelligent User Interfaces. San Francisco: Morgan

Kaufmann Publishers Inc.

Nehmer, J., Becker, M., Karshmer, A., and Lamm, R.

(2006) Living assistance systems: an ambient

intelligence approach. In: Proceedings of the 28th

international Conference on Software Engineering,

Shanghai, China, May 2006. New York: ACM, pp. 43-

50.

O'Hare, G. M.P., O'Grady, M. J., Muldoon, C., and

Bradley, J. F. (2006) Embedded agents: a paradigm for

mobile services. Int. J. Web Grid Serv. 2 (4): 379-

405.Vasilakos, A. and Pedrycz, W. (Eds.) (2006)

Ambient Intelligence, Wireless Networking, and

Ubiquitous Computing. Norwood: Artech House, Inc.

O’Grady, M.J., O’Hare, G.M.P. Sas, C., (2005) Mobile

Agents for Mobile Tourists: A User Evaluation of

Gulliver's Genie. Interacting with Computers,

17(4):343-366

O'Hare, G. M. and O'Grady, M. J. (2002) Addressing

Mobile HCI Needs through Agents. In Proceedings of

the 4th international Symposium on Mobile Human-

Computer interaction, Pisa, Italy September 2002.

London: Springer-Verlag, pp. 311-314.

ICSOFT 2009 - 4th International Conference on Software and Data Technologies

364