Conference PaperPDF Available

Abstract

This paper presents an agent-based approach for realizing a new generation of intelligent environments referred to as adaptive ambient ecologies. These are highly distributed systems, which require new ways of communication and collaboration in order to support the realization of people's tasks. We use three types of agents for the main functions of planning, adaptation and interaction. The knowledge repository of an ambient ecology is encoded in an ontology that is assembled on demand and is made accessible to the agents. In our approach, we introduce methodologies for extending the concept of ontologies beyond the storage of structured data to further include mechanisms for the exchange and manipulation of activity related information within ecologies.
1
Abstract—This paper presents an agent-based approach for
realizing a new generation of intelligent environments referred
to as adaptive ambient ecologies. These are highly distributed
systems, which require new ways of communication and
collaboration in order to support the realization of people’s
tasks. We use three types of agents for the main functions of
planning, adaptation and interaction. The knowledge
repository of an ambient ecology is encoded in an ontology that
is assembled on demand and is made accessible to the agents. In
our approach, we introduce methodologies for extending the
concept of ontologies beyond the storage of structured data to
further include mechanisms for the exchange and manipulation
of activity related information within ecologies.
I. INTRODUCTION
NTELLIGENT environments (IEs) are (usually closed)
spaces equipped with a variety of technical components
ranging from flat-screen monitors, centrally controlled
lights, radio receivers to HVAC systems and automated
window blinds. The strong heterogeneity of these
components makes their configuration and maintenance as
part of an integrated system a non-trivial task. Within the
European Commission funded project ATRACO (Adaptive
and Trusted Ambient eCOlogies) we try to lay the
foundation for the development of next generation IEs, so-
called Ambient Ecologies (AEs) [1]. An ambient ecology is
a set of devices located in close proximity (i.e. in the same
space) and services that can communicate, collaborate, and
interact with each other, the environment and most
importantly, the people inhabiting the IE.
At the basis of our approach is the assumption that the
end-user(s) want(s) to fulfil aims or goals such as feel
comfortable, go shopping or prepare dinner within their
AEs. For this reason we use the notion of activity spheres
describing the aim-dedicated conglomeration of entities and
the collective knowledge surrounding the user living in an
T. Heinroth is with the Institute of Information Technology, Ulm
University, Ulm, 89081 Germany (phone: 0049-731-50-26265; fax: 0049-
731-50-26259; e-mail: tobias.heinroth@ uni-ulm.de).
A. D. Kameas is an assistant professor with the Hellenic Open University
and the director of DAISy research unit at the Computer Technology
Institute, both in Patras, Hellas (phone: 0030-2610-367402, e-mail:
kameas@cti.gr).
C. Wagner is with the Computational Intelligence Centre, Department of
Computing and Electronic Systems, University of Essex, Colchester CO4
3SQ, U.K. (e-mail: chwagn @ essex.ac.uk).
Y. Bellik is with the National Center for Scientific Research (LIMSI-
CNRS) BP 133, 91403, Orsay cedex, France and also with the Paris-South
University (phone: 0033-169858110; fax: 0033-169858088; e-mail:
Yacine.Bellik@limsi.fr).
ecology. The formation of such activity spheres is supported
by a service-oriented ambient ecology architecture, which
includes APIs to interface with existing hardware modules
and communication protocols, ontologies and ontology
management modules, decision making mechanisms,
planning modules, negotiation and learning mechanisms,
intelligent agents, trust policies and privacy enforcement
mechanisms and compose-able interaction components, all
necessary for the formation of task-based transient
ubiquitous computing applications.
Compared to previous research, we adopt a unique
standpoint in that we assume that (a) various IEs are
available, each of them hosting an AE and (b) devices and
services of an AE (1) are inherently heterogeneous and (2)
as such contain heterogeneous descriptions of their
capabilities and services that can only be accessed from but
not modified by other components. Thus, in order to achieve
task-based collaboration amongst them, one has to deal with
this heterogeneity, while at the same time achieving
independence between a task description and its respective
realization within a specific AE. To this end, we employ
ontology alignment mechanisms, which are described in
more detail in Section IV. Moreover, the system aims to
establish only a transient collaboration between members of
an AE in order to support a specific user activity in changing
context. In order to achieve this, we use a set of intelligent
agents to support adaptive planning, task realisation and
enhanced human-machine interaction.
In this paper, we shall focus on the realization of the
ontology and the agents that support adaptive realization of
user tasks within changing heterogeneous AEs.
The remainder of this paper is organised as follows. In
Section III, the related work is presented. The conceptual
architectural design is introduced in Section III and the core
software modules Sphere Manager and Ontology Manager
are presented. Section IV describes the agents which are
necessary for the formation of the ecology. Section V
focuses on the realization of ambient ecologies concerning
information storage, exchange and manipulation through the
use of ontologies. An example realization of a task within an
AE is used throughout the paper to better illustrate our
concepts. Finally we draw conclusions and present some
aspects of our future work in Section VI.
II.
RELATED WORK
Ontology based pervasive systems have already been
presented in the literature. Among these:
A Proposal for Realizing Adaptive Ambient Ecologies through an
ontology-based Multi-Agent System
Tobias Heinroth, Achilles Kameas, Christian Wagner, and Yacine Bellik
I
- the CADO (Context-aware Applications with
Distributed Ontologies) framework [2] relies on
distributed ontologies that are shared and managed in
a peer-to-peer fashion, so as to ensure semantic
interoperability via the process of ontology merging;
- CoBrA (Context Broker Architecture), is a pervasive
context-aware computing infrastructure, that provides
knowledge sharing, context reasoning, and privacy
protection support for ubiquitous context-aware
systems, using a collection of ontologies, called
COBRA-ONT, for modelling the context in an
intelligent meeting room environment. These
ontologies expressed in the Web Ontology Language
(OWL), define typical concepts associated with
places, agents, and events and are mapped to the
emerging consensus ontologies that are relevant to
the development of smart spaces [3];
- GAIA is an infrastructure for Smart Spaces, which
are ubiquitous computing environments that
encompass physical spaces [4]. Ontologies are
introduced in this system, as an efficient way to
manage the diversity and complexity of describing
resources, that is, devices and services. Therefore,
these ontologies are beneficial for semantic
discovery, matchmaking, interoperability between
entities, and interaction between human users and
computers.
All these ontology-based systems use static heavyweight
domain ontologies to support ubiquitous computing
applications. These ontologies are used to represent,
manipulate, program and reason with context data and they
aim to solve particular ubiquitous computing problems, such
as policy management, context representation, service
modelling and composition, people description, location
modelling, etc. However, in the ubiquitous computing
Fig. 1. High-level view of one activity sphere including the main actors: agents, managers, devices, services, ontologies and the user.
domain, it is difficult for applications to share changing
context information, as they will have to constantly adapt to
the changes.
There have also been proposed agent-oriented
architectures for pervasive systems. For example, project
ASK-IT supports agent applications on a service-oriented
architecture, using agents as brokers between the needs of
the users and the server-side services [http://www.ask-
it.org]. Its successor, project OASIS, has adopted a full
fledged agent system, which consists of user agents, service
prioritisation agents, service provider agents and content
connector agents; a hyperontology is also used to deal with
heterogeneity [5]. This architecture is close to ATRACO,
but it approaches the problem of adaptation from a different
perspective – that of global co-ordination and optimization
of matching user requests with systems services based on
user preferences.
On the contrary, in ATRACO we propose an ontology-
based adaptation approach, which can be tailored to the task
description, applied by intelligent agents that realize task
execution, and maintained dynamically by re-calculating on-
demand the interfaces between the constituent local
ontologies.
In modelling ATRACO functionality, we adopt an
approach similar to the one described in project DynAMITE,
but we are proposing a distributed task-oriented resource
management and interaction scheme [6].
III. A
RCHITECTURE
In its traditional interpretation, the concept of IEs refers
mainly to the resources of a space and the services made
available to users or devices within it. We extend this notion
to define Ambient Ecologies, whereby we assume that all IE
resources (i.e. devices, services, data, etc.) are self-
described, accessible and have the capability for
communication. With the support of an AE, user tasks can
be realized. We introduce the notion of Activity Sphere (AS)
to model the set of resources and the software modules
necessary to support the realization of specific user goals
that can be expressed as inter-dependent tasks. Aim-related
activity spheres can represent the adaptive realization of a
user’s task in a changing environment concerning states,
events and all involved entities.
An AS is both the virtual description of the resources
(e.g., devices, services, agents, and user) required to achieve
a user’s single aim and its fulfilment in the context of a
specific ecology. Moreover, one AS may be owned by
multiple users, multiple AS may exist within the same AE,
while the same resource may be used in different AS. Thus,
the realization of AS based on AEs requires a new
architectural approach discussed in this section. In our
approach, an AS usually resides in an AE and in general
consists of ontologies as primary knowledge and information
repository, the user itself as the task owner (and possibly the
environment’s occupant) and further entities, which can be
categorised as passive and active.
Throughout this paper, we shall use the example of an AS
that supports the user’s goal “Feel comfortable after work”.
Fig. 1 shows a snapshot of the AS that supports it. The AS is
everything inside the main circle cloud. The sphere consists
of:
1) Ontologies:
The aim ontology acts as a description of
the user’s aim defined as a set of goals
each modelled within a task model (i.e.,
the domain).
Local ontologies are provided by both
active and passive entities and encode their
state, properties, capabilities and services.
The sphere ontology is an information
pool which is constantly evolving and
being updated. It serves as the core of the
sphere. Different ways of forming this
ontology are possible: Aligning, merging
and mapping (see Section V) of entities’
local ontologies. The two latter
possibilities result in a new ontology.
Aligning forms a transient quasi-ontology
which connects its constituting ontologies
though a network of reconfigurable
pointers.
2) Passive entities:
Devices such as interaction devices (touch
screens, speakers, microphones, etc.),
actuators and sensors including
televisions, radio receivers and HVACs,
etc.
Services such as remote or external web-
services (e.g., online banking) and local or
internal services (e.g., personal calendar).
3) Active entities:
Agents (Fuzzy Task Agent, Interaction
Agent and Planning Agent) are responsible
for automated adaptation, resolving
conflicts, interacting with the user,
establishing plans and in general realizing
the concrete tasks in the task model.
The ontology manager is responsible for
keeping the sphere ontology up-to-date
and for handling events within the sphere.
The sphere manager is responsible for the
formation, the integration into and
releasing from activity spheres of passive
entities as well as the spheres’ dissolving.
4) The user who as the occupant of the IE is at the
centre of each activity sphere.
The high-level view of the AS in Fig. 1
shows how the
realization of the aim Feel comfortable after work can be
supported. The sphere consists of a defined set of passive
and active entities, which participate in the different tasks
that realize the sphere goals. Each goal is decomposed into a
hierarchy of abstract tasks, where the description of an
abstract task does not depend on the AE at hand. Based on
this abstract description and the descriptions of the resources
of a specific AE, the Planning Agent derives the descriptions
of concrete tasks. Once the concrete tasks are computed,
they are mapped to the various entities in the AE. The
decomposition to concrete tasks and the mapping of entities
may initially be computed offline. As soon as the mapping is
complete, each active entity independently tries to realize its
function within the task by using passive entities or
interacting with other active entities.
The user and both passive and active entities except the
sphere and the ontology managers are maintaining their own
local ontology which stores their state, properties,
capabilities and services. Thus, all data concerning an agent,
a device, a service or a user is stored in their respective local
ontologies. Fig. 1 shows the local ontologies as small
ellipses. The lights, for example, may maintain the following
values within their local ontology:
Watts (W)
Amperage (A)
Durability (h)
Number of bulbs (integer)
Actual state (Boolean [on | off] or as linguistic
variable [bright | dim | dark)
The user’s local ontology is used as his profile. All active
entities can edit and add-in values regarding the user to this
ontology. Such values can be, for example, private
information, the user’s abilities and system-learned user-
behaviours or preferences.
The architecture presented in this section is based on the
principle of independence between AE and AS; thus an AS
may be realized in different AEs (in the general case, this
includes an AE changing its configuration because, for
example, a device has been switched off, or a new device
has arrived). Of course, this will require adaptation of the
AS, as two AEs cannot be expected to be identical. In our
architecture, we use three types of agents to achieve
adaptation: a Fuzzy Task Agent (FTA), a Planning Agent
(PA) and an Interaction Agent (IA) (see Section IV).
The PA re-computes the mapping from abstract to
concrete tasks when the AE changes. The FTA controls
directly the AE resources that have been mapped to a
specific task by implementing adaptive Fuzzy Logic based
control methods. The IA intervenes when a decision on
adaptation or task concretization cannot be reached, or when
the user wishes to express an explicit request to the system.
A. Sphere manager
The sphere manager (SM) module is responsible for
setting up, managing and dissolving an activity sphere. It
does so based on a goal description and on information
about the entities of the AE. SM then initializes the PA and
passes it the abstract task description. Subsequently, based
on the concrete task plan, it initializes the Ontology
Manager, which composes the sphere ontology.
The SM monitors the AE and “listens” to changes in its
composition. When such a change is detected (i.e., the SM
gets notified that a new device has entered, or some device
cannot be discovered any longer), it triggers the PA in order
to re-compute the concrete task plan, which is again passed
to the Ontology Manager.
Finally, the SM dissolves the sphere, once all tasks have
been accomplished and the goal attained, by asking the
agents to “shut-down” and relinquish the resources they’ve
been using, and by asking the Ontology Manager to delete
the ontology alignments.
B. Ontology manager
The Ontology Manager (OM) module creates and
manages the sphere ontology. It uses ontology alignment
algorithms to uncover alignment points between the
constituent ontologies, which it maintains until the sphere is
dissolved. The OM is able to answer queries in respect to the
knowledge stored in the ontology by using the alignments to
access the constituent ontologies, which in turn contain the
requested information.
IV. T
HE MULTI-AGENT SYSTEM
Agents are the most important active entities within the
activity sphere, as they directly affect and manipulate the
ecology surrounding the user. Three different types of agents
are necessary for the formation of an ambient ecology:
Planning Agents, Fuzzy Task Agents and Interaction Agents.
A. Planning Agent
The Planning Agent is able to solve combinatorial
problems that have a relative small number of consistent
plans (solutions) with respect to all possible plans. The PA’s
aim is to resolve all goals into concrete tasks given a specific
ecology [7], [8]. To start the planning process some initial
information (such as an abstract task model and a problem
description) is already provided regarding the ambient
ecology, but some information may also has to be retrieved
initially by communicating with the user through the
Interaction Agent.
During the reasoning process and during the evolution of
the sphere, a situation may occur, where the PA needs
further information that is either needed to complete the
planning or to direct search towards the solutions that the
user prefers. Direct user-machine interaction will then be
initiated by involving the Interaction Agent in those
situations.
B. Fuzzy Task Agent
The Fuzzy Task Agent (FTA) is responsible for the direct
control of any entities within the current ecology. More
specifically, an individual FTA is instantiated by the sphere
manager for the realization of each concrete task. During
instantiation, the FTA is associated with the input entities
(e.g. sensors) and output entities (e.g. actuators) which have
been identified by the PA in order to realize the current task.
If a specific task has previously occurred, the FTA will
incorporate any rule sets and fuzzy set definitions which
have previously been defined and stored in local ontologies
of the respective involved entities (e.g. the user).
If the task is considered novel, the FTA will start
monitoring the entities it has been associated with for a
specified amount of time, recording the combination of their
individual states as they change within the ecology (e.g.
heater setting modified by user, temperature sensor, etc.).
After the monitoring phase is complete, the recorded set
of input-output combinations is evaluated and a number of
fuzzy membership functions are created which model the
individual states of the entities [9]. The number of fuzzy
membership functions as well as the linguistic labels for
each of these functions is retrieved from the sphere
ontology. A heater for example could be represented by
three fuzzy logic sets labelled “cold”, “medium” and
“warm”.
Fig. 2. A schematic view of the Fuzzy Task Agents's adaptation process.
After the fuzzy sets have been created, a rulebase, which
incorporates the different input-output mappings from the
gathered data and applies them to the fuzzy sets, is generated
according to the principles outlined in [4]. The rules within
the rulebase have the following format which is standard in
Fuzzy Logic Control:
:






,…,

,
1,...,
,
(1)
where
to
are fuzzy input sets,
to
are fuzzy output
sets, P is the number of FLC inputs, Q the number of FLC
outputs and S is the number of rules in the rule base. After
the initial setup phase is complete, the FTA starts controlling
the entities within the ecology in order to fulfil its current
task (e.g. “Heat the room to a comfortable level”). As the
characteristics of the entities are subject to a high level of
uncertainty (e.g. sensors give inaccurate readings, actuator
performance changes over time, etc.) and as the user
preferences (e.g. what temperature is “comfortable”?) are
subject to change - which can be seen as another source of
uncertainty, we are applying a lifelong learning/adaptation
approach, based on the concepts detailed by Doctor et al. in
[10].
The sphere adaptation is with the help of the FTA,
realised on two levels. First, the actual rules which can be
created, deleted or modified to reflect changes within the
ecology or the user preferences. Second, the actual fuzzy
sets which represent the linguistic variables are modified to
reflect changes in the perception of the user of these
variables (e.g. if a user changes his concept of “cold”, the
fuzzy set should change to reflect this). We are aiming to use
the strong uncertainty modelling capabilities of type-2 fuzzy
logic sets [11] in order to model and adapt the meaning of
the linguistic variables as accurately as possible while
maintaining a small number of rules, which in turn should
improve performance and facilitate human interpretation and
as such system accountability.
During its operation, the FTA continually monitors the
entities which are associated with its inputs and controls the
entities associated with its outputs according to its rulebase
which is constantly adapted. A schematic view of this
process can be seen in Fig. 2. As the FTA adapts, it exposes
the modified fuzzy set definitions and the rulebase through
its ontology to the sphere ontology. This allows for other
components of the system to react to new information
encoded within the FTA’s rules as well as maintaining an
up-to-date view of the definition of specific linguistic
variables.
C. Interaction Agent
The Interaction Agent (IA) is responsible for the direct
interaction between the user and the system within the
current ecology. It is in charge of the interpretation of the
messages produced by the user and the presentation of
messages produced by the system (through the PA).
Different communication modalities may be used, either by
the user or the IA, to express messages (speech, touch
buttons, text, graphics, avatars, etc.). The IA has to
determine the best means of interaction (modalities and
devices) [12] according to the current ecology. Similar to the
FTA, an individual IA is instantiated by the sphere manager
for the realization of each concrete task. During
instantiation, the IA can access the following elements to
achieve its mission:
Available input and output interaction devices such
as speakers, microphones, touch screens, etc.
Available input and output interaction modalities
(e.g., text, graphic, touch, speech): the list of
available interaction devices is not enough for the
IA to achieve its mission because an interaction
device may support different interaction modalities.
For instance, a speaker can support different
modalities, such as earcon (sound patternused to
represent a specific item), speech synthesis, music,
etc. The presence of a speaker in the list of
available interaction devices doesn't necessarily
mean that the speech synthesis modality is also
available. As such, the IA needs to be aware of
what interaction modalities are available within the
current ecology.
Available input and output modes: modes refer to
the human sensory-motor systems such as vision
and audition. The IA needs to know which modes
are available and/or preferred at a given moment to
determine the best modalities to use. This
availability/unavailability and preference may be
temporary (for instance the user is currently closing
his eyes) or permanent (the user is blind). Such
information may be retrieved from the user profile,
and it is important to not select an interaction
modality which is not preferred by the user because
it may not be perceived by him.
Physical environment characteristics: the
characteristics of the physical environment (such as
ambient light level, ambient sound level, etc.) may
influence the interaction process (for instance, a
sound modality will be inadvisable if the ambient
sound level is loud). The IA needs this information
to select the best modalities to use.
Message characteristics: the characteristics of the
interaction message (such as its semantic content,
its importance, its emergency state, etc.) may also
influence the way the message will be expressed.
For instance, redundant modalities (speech
synthesis and text display) may be used to express
an important message.
Based on all these elements, the IA will use a behavioural
model, which identifies the best interaction components
(modalities and devices) adapted to the current ecology. Its
formalization can be made in different ways: rules [13],
automats [14], Petri networks [15], etc.
V. R
EALIZATION OF AMBIENT ECOLOGIES
The main concept of our approach is that we use
ontologies not only for storing and classifying data but also
for the communicative exchange of data and knowledge
within a sphere. The sphere ontology is used as an
intermediate layer between active and passive entities of a
sphere. As will be described in this section, passive entities
contribute their state to the ontology, while active entities
may also change it.
A. Role of the ontology
An ontology is usually defined as “a formal, explicit
specification of a shared conceptualization” [16]. A
“conceptualization” refers to an abstract model of some
phenomenon in the world, which identifies the relevant
concepts of that phenomenon. “Explicit” means that the type
of concepts used and the constraints on their use are
explicitly defined. “Formal” refers to the fact that the
ontology should be machine readable. “Shared” reflects the
notion that an ontology captures consensual knowledge, that
is, it is not private of some individual, but accepted by a
group. Thus, an ontology is a structure of knowledge, used
as a means of knowledge sharing within a community of
heterogeneous entities.
In our approach, we assume that each AE entity (user,
agent, device or service) maintains locally an ontology
(could only be a protocol header or a simple description),
where its properties, capabilities and current state are
described. This local ontology is managed only by the
owning entity. In other words, the local ontology of an entity
represents the complete set of knowledge associated with
this entity.
Based on the concrete task plan, which details the entities
that must be used in order to realize an abstract task, the
ontology manager forms the sphere ontology, which
contains information about:
- the states of the devices and services that participate
in the task;
- the knowledge bases of the AS agents;
- the user profile;
- the constraints and policies that apply to the
realization of the goal and its tasks.
The sphere ontology is formed by accessing the local
ontologies of these entities and attempting to interface them
with the help of ontology matching algorithms. Ontology
matching is the process of finding relationships or
correspondences between entities of two different
ontologies. Its output is a set of correspondences between
two ontologies, that is, relations holding, or supposed to
hold, between entities of different ontologies, according to a
particular algorithm, or individual. This set of
correspondences is called an alignment. According to [17],
the ontology alignment process is described as: given two
ontologies, each describing a set of discrete entities (which
can be classes, properties, rules, predicates, or even
formulas), find the correspondences, e.g. equivalences or
subsumptions, holding between these entities.
Based on these alignments, one can apply ontology
merging in order to produce the top level, sphere ontology,
which realizes an activity sphere [18]. When performing
ontology merging, a new ontology is created, which is the
union of the source ontologies. The new ontology, in
general, replaces the original source ontologies. The
challenge in ontology merging is to ensure that all
correspondences and differences between the ontologies are
reflected in the merged ontology. However, ontology
merging cannot be directly applied in our system, although it
would speed up the knowledge management within a sphere,
because the constituent (local) ontologies will continue to
exist and change, thus the sphere ontology will require
continuous re-engineering.
B. Functional description
The main function provided by the sphere ontology is
information sharing amongst entities populating the sphere.
Apart from the passive entities which simply provide a local
ontology and keep it synchronized with their actual state,
agents play a more sophisticated role within the AS: agents
can read and write data encoded in all the ontologies they are
aligned to. To realize, for example, the task “Heat the room
to a comfortable level” the PA initially may identify the
following agents and corresponding passive entities which
are necessary for this task:
User (passive);
Radiator (passive);
Window (passive);
FTA (active);
IA (active).
The OM is now able to align these local ontologies and
build up a common knowledge base. The FTA might find
information about the preferred temperature in the user’s
ontology, which can be used as a starting point for learning.
It can also directly edit the values in the local ontologies of
the radiator and the window to change their state. Due to the
alignment of the FTA and the IA, it is also possible for the
IA to react to a direct user input about temperature by
triggering the FTA to change the radiator’s or the window’s
state. Whenever a structural change happens (i.e., a new
device is discovered by a device registration service), the
OM will do a realignment to update the sphere ontology and
triggers the agents to include the new elements they are
aligned to in deducing their behaviour. If a device changes
its state the OM does not have to realign; it can either throw
events to trigger the agents to read a new value or the agents
themselves can constantly read values of entities they are
aligned to in order to react adequately.
We can use the same paradigm to demonstrate how a
complete goal and not only one single task is fulfilled. Since
a sphere is directly linked to a specific goal, it is possible to
align, map or merge different ontologies in respect to several
tasks. Thus, the communication within the sphere depends
on the current activity and for this reason only relevant data
will be transferred. We divide the members of the ecology
into active and passive for the following reason: all passive
entities can use their local ontology only to provide
information, i.e., they behave passive.
By contrast agents can also express the requirement for
additional information not contained in their individual local
ontology, i.e., they behave active. If the ecology provides
knowledge related to the agent’s requirement, the ontology
manager aligns this knowledge with the local ontology of the
agent and by doing so enables it to act in a more intelligent
way. The managers within the ecology are by definition
active, as they throw events, form structures (i.e., ontologies)
and align, map or merge data provided in local ontologies.
C. Example
The following short scenario should help to understand
how we want to use our concept for adaptation:
Suki has been living in this new apartment for the past 10
months. His is no ordinary house; it is not a commonplace
first generation smart house: it is a brand new adaptive
house! When he comes home after work he wants to feel
comfortable and for this reason the house should adapt the
temperature, the level of lighting and sometimes present his
favourite TV program. Suki prefers a warm living room and
a cold bedroom.
The goal “Feel comfortable after work” can be described,
for example, with abstract tasks such as:
1) Set a comfortable temperature (TEMP)
a. Sense the indoor and outdoor temperatures
b. Adjust room heating/cooling according to
the user preferences and context.
2) Set a comfortable level of lighting (LIGHT)
a. Sense the indoor light levels
b. Adjust indoor light levels according to the
user preferences and context.
3) Select favourite TV program (FAVTV)
a. Check Media options
b. Set Media according to the user preference
and context.
At home the abstract task TEMP could be made concrete
by the PA with the use of the entities mentioned in the last
section. In this scenario the concrete task may look like:
1) Concrete TEMP
a. FTA senses indoor temperature using
Sensor S1 in the living room.
b. FTA senses indoor temperature using
Sensor S2 in the bedroom.
c. FTA senses outdoor temperature using
Sensor S3.
d. FTA checks Suki’s temperature
preferences stored in his local ontology.
e. FTA deduces Suki’s favourite temperature
for the several rooms.
f. FTA adjusts temperature by using the
windows
g. FTA adjusts temperature by using the
radiator
h. IA provides control dialogues (using
different modalities) for allowing Suki to
directly adjust the temperature.
The OM aligns the entities referring to this plan and Suki
is glad about the setting but on the next day the scenario
continues as follows:
Suki has to stay the night at a hotel in the city close to his
office due to a public transportation strike. When Suki enters
the hotel room the AS detects that the goal “Feel
comfortable after work” should be set up in a different
environment.
The PA checks the hotel room’s device registry and
adapts to the new environment by evolving the abstract task
TEMP into a different concrete one:
1) Concrete TEMP
a. FTA senses indoor temperature using
Sensor S1 in the hotel room.
b. FTA checks a weather forecast service to
get the outside temperature.
c. FTA checks Suki’s temperature
preferences stored in his local ontology.
d. FTA deduces Suki’s favourite temperature
for the hotel room.
e. FTA adjusts temperature by using the
HVAC
f. IA provides a more sophisticated dialogue
(using speech modality) for allowing Suki
to directly adjust the temperature in the
new environment.
The FTA recognizes the hotel room as a bedroom and sets
up the HVAC to cold. Suki does not want to go to bed but
wants to finish some work and for this reason he wants the
AS to adjust the temperature to a normal level. He starts a
set-up dialogue and tells the system always to set the
temperature to a normal level if there is only one room
available.
The outcome of this set-up dialogue will be stored by the
IA in Suki’s local ontology and makes it possible for the
FTA to involve such information in future decision to make
automatic adaptation more exact.
VI. C
ONCLUSIONS
We have presented an approach for supporting the
realization of adaptive pervasive applications. It is based on
(a) the existence of heterogeneous smart components
(devices, services) within an Ambient Intelligence
environment, (b) the fact that these components maintain
and make available local representations of their self and
state and (c) their ability to communicate and collaborate.
We have termed this set of components an Ambient Ecology
and we use it to realize user tasks. We conceptualize as an
Activity Sphere a set of coherent tasks aiming to the same
goal.
Our system supports the instantiation of an Activity
Sphere within a specific Ambient Ecology. To achieve this,
we propose the engineering of a sphere ontology by
matching the local ontologies of the ecology components.
The sphere ontology will serve as a common repository of
knowledge and information among passive and active sphere
entities. The latter are users and agents of three kinds: fuzzy
task agent, planning agent, and interaction agent. All these
maintain their local knowledge bases, which are also
matched in the sphere ontology.
After drawing the conceptual picture of ATRACO our
future work aims toward implementing a basic working
system that can be used as a starting point for a evaluation
sessions with real users.
A
CKNOWLEDGMENT
The described work has been supported by the European
Commission funded research project ATRACO
(www.atraco.org) (Grant agreement No. 216837) within the
7
th
Framework Programme (FP7) as part of the FET
Pervasive Adaptation (PERADA) initiative.
R
EFERENCES
[1] C. Goumopoulos and A. Kameas, “Ambient ecologies in smart homes”,
The Computer Journal, August 2008.
[2] F. De Paoli and M. Loregian, “Context-aware applications with
distributed ontologies”, in Proc. 18th International Conference on
Advanced Information Systems Engineering, 2006.
[3] H. Chen, T. Finin, and A. Joshi, “An ontology for context-aware
pervasive computing environments”, Knowledge Engineering, Special Issue
on Ontologies for Distributed Systems, 2003.
[4] M. Roman, C. K. Hess, R. Cerqueira, A. Ranganathan, R. H. Campbell,
and K. Nahrstedt, “Gaia: A middleware infrastructure to enable active
spaces”, IEEE Pervasive Computing, vol. 1, no. 4, pp. 74–83, 2002.
[5] ICT 7th Framework Programme and Ageing – Grant Agreement No:
215754, “The oasis project”, http://www.oasis-project.eu/.
[6] Bundesministerium für Bildung und Forschung (Förderkennzeichen
01ISC27 x), “The dynamite project”, http://www.dynamite-project.org/.
[7] B. Schattenberg and S. Biundo, “A unifying framework for hybrid
planning and scheduling”, in Proc. 29th German Conference on Artificial
Intelligence (KI 2006), LNAI 4314, U. Furbach, Ed., 2007, pp. 361–373.
[8] J. Bidot, B. Schattenberg, and S. Biundo, “Plan repair in hybrid
planning”, in 31st German Conference on Artificial Intelligence (KI 2008),
LNAI 5243, 2008, pp. 169–176.
[9] F. Doctor, H. Hagras, and V. Callaghan, “An intelligent fuzzy agent
approach for realising ambient intelligence in intelligent inhabited
environments”, IEEE Transactions on System, Man & Cybernetics, vol.
35(1), pp. 55–65, January 2005.
[10] H. Hagras, F. Doctor, V. Callaghan, and A. Lopez, “An incremental
adaptive life long learning approach for type-2 fuzzy embedded agents in
ambient intelligent environments”, IEEE Transactions on Fuzzy Systems,
vol. 15(1), pp. 41–55, February 2007.
[11] J. Mendel, Uncertain Rule-Based Fuzzy Logic Systems, 2001.
[12] C. Rousseau, Y. Bellik, F. Vernier, and D. Bazalgette, “A framework
for the intelligent multimodal presentation of information”, Signal
Processing, vol. 86(12), pp. 3696–3713, December 2006.
[13] C. Stephanidis, C. Karagiannidis, and A. Koumpis, “Decision making
in intelligent user interfaces”, pp. 195–202, January 1997.
[14] Y. Bellik and D. Teil, “A multimodal dialogue controller for
multimodal user interface management system, application: a multimodal
window manager”, pp. 93–94, April 1993.
[15] R. Bastide, P. Palanque, H. Le Duc, and J. Munoz, “Integrating
rendering specifications into a formalism for the design of interactive
systems”, Proc. 5th Eurographics Workshop on Design: Specification and
Verification of Interactive systems (DSV-IS'98), June 1998.
[16] T. Gruber, “A translation approach to portable ontologies”,
Knowledge Acquisition, vol. 5(2), pp. 199–220, 1993.
[17] J. Euzenat and P. Schvaiko, Ontology Matching, 2007.
[18] L. Seremeti and A. Kameas, “Ontology-based high-level task
composition in ubiquitous computing applications”, in Proc. IET 4th
International Conference on Intelligent Environments (IE08), 2008.
... Adaptation and evolution are supported with genetic algorithms, machine and statistical learning. Heterogeneity is dealt with the application of ontology matching and merging techniques [2,7,10,11,15,17,18]. ...
Article
Full-text available
This paper proposes an environmental sensor bridge system named COSPI for autonomous communication robots. In the near future, various sensors will get installed in day-to-day environments and robots will use information from these sensors to recognize ...
Article
Pervasive Awareness Systems (PASs) collect awareness information from sensors and devices in smart environments and can present it to remote users by using various awareness objects. In this paper, we present a pervasive awareness management framework for supporting the conception, design and realisation of applications that foster social connectedness between communities. The framework consists of: 1) an awareness model that considers user disturbance and privacy; 2) a service-oriented platform that provides core awareness services; 3) assorted tools that support communities to create, adapt and appropriate such applications. A PAS prototype is presented along with a usability evaluation of the tools.
Conference Paper
Prioritisation is an important technique for resolving planning conflicts between agents with shared resources, such as robots moving through a shared space. This paper explores the use of genetic-based machine learning to assign priority dynamically, to improve performance of a team of agents without unduly impacting individual agents' performance. A decoupled heuristic approach is used for flexibility, whereby individual XCS agents learn to optimise their behaviour first, and then a high-level planner agent is introduced and trained to resolve conflicts by assigning priority. The approach is designed for Partially Observable Markov Decision Process (POMDP) environments and demonstrated on a problem in 3D aircraft path planning.
Conference Paper
In the last decade, a technological tendency is observed related to the integration of the digital objects in the daily routine of people. Latest technology achievements as wireless networks, tangible interfaces, embedded interaction and in general the diffusion of computing power in small units through space, reveal that the boundaries between physical and digital space are blurring. From contemporary technology applications, turns up that digital spaces can be easily adapted to our everyday life in the urban environment. That way a new perspective for architects appears: to manage simultaneously digital and physical urban space. This paper aims at indicating cases where digital tools help sustainable development in the urban environment and highlighting a potential research field for architecture that combines AmI, built environment and sustainability, by analyzing them. Through ambient or pervasive computing with specific applications and examples that will be extensively presented, it is shown how digital technology can be used for sustainable urban environments especially in the field of architecture, with respect to the local conditions of place.
Conference Paper
Full-text available
Management of trust for federation activities requires intensive monitoring and reconfiguration services to accommodate dynamic trust behaviors across trust domains. To detect and control potential conflicts between these behaviors, trust services in federation systems need to be able to perform dynamic conflict analysis. Policy analysis provides a way to detect potential conflicts in system behaviors. We introduce event calculus (EC) into policy analysis to provide dynamic policy conflict analysis for detection and control of dynamic conflicts in trust services for federations. An application of this dynamic analysis to a Web-service-based healthcare system is also described.
Conference Paper
Next Generation Ambient Intelligent Environments (NGAIEs) will offer intelligent and trustworthy services (although probably not for free), will be populated with Ambient Ecologies of smart objects, and will support the Activity Spheres of their multiple occupants. Ubiquitous computing applications will be composed from heterogeneous services offered by objects connected via the web and NGAIEs. These applications will inherently exhibit increasingly intelligent behavior, perceive the operational environment and adapt to it, provide optimized resource usage and support consistent functionality and human-centric operation. This paper tracks the disappearance of computers and the advent of Ambient Intelligence.
Conference Paper
One of the principal objectives of today's ambient intelligence environments is to provide users with services according to their activity, i.e. preferences in order to accomplisha specific user goal. For this purpose, miscellaneous user information must be collected and structured into user profiles. These profiles offer the advantage of being easily enriched and exploitable by the environment, in order to deliver to the user, at any moment and at any place, the best fitted service, with regard to his activity. In this paper we present an ontology-based user profile model which provides the static data about a user but also the context of the user in a given situation in real time.
Article
As the computer disappears in the environments surrounding our activities, the objects therein become augmented with sensors, actuators, processors, memories, wireless communication modules and they can receive, store, process and transmit information. In addition to objects, spaces also undergo a change towards becoming smart and eventually Ambient Intelligence (AmI) spaces. In order to model the way everyday activities are carried out within an AmI environment, we introduce the notion of “activity sphere”. In this paper, we are interested in the ontology-based representation of activity spheres from two different perspectives (as creators and as observers), as well as the modeling and control of the dynamic nature of activity spheres.
Article
Full-text available
We envision a future where people's living spaces are interactive and programmable. Users interact with their offices, homes, cars, malls and airports to request information, benefit from the resources available, and configure the habitat's behavior. Data and tasks are always accessible and are mapped dynamically to convenient resources present in the current location. Users may extend the habitat with personal devices that seamlessly integrate with the environment. Such user-oriented interactive environments may require a novel software infrastructure to operate their resources, sense context properties, and assist in the development and execution of applications. In this article, we present an experimental middleware infrastructure called Gaia that we have used to prototype the resource management of and provide the user- oriented interfaces for such physical spaces populated with network-enabled computing resources. To limit the scope of our research, we focus on physical spaces used for teaching; classrooms, offices, and lecture rooms. The system described in this paper is derived from a series of experiments starting in 1996. We show how, by applying the concepts of a conventional operating system to middleware, we can manage the resources, devices and distributed objects in a room, building, or physical space, how a distributed extension of the model-view-controller that is use in personal computers simplifies and structures practical applications for these environments, and how, by driving context-sensitivity into its data storage mechanisms, the system can help satisfy the requirements for user-centricity and mobility.
Article
Full-text available
In this paper we describe a novel life long learning approach for intelligent agents that are embedded in intelligent environments. The agents aim to realise the vision of Ambient Intelligence in Intelligent Inhabited Environments (IIE) by providing 'ubiquitous computing intelligence in the environment supporting the activities of the user. An unsupervised, data-driven, fuzzy, technique is proposed for extracting fuzzy membership functions and rules that represent the user's particularised behaviours in the environment. The user's learnt behaviours can then be adapted online in a life long mode to satisfy the different user and system objectives. We have performed unique experiments in which the intelligent agent has learnt and adapted to the user's behaviour, during a stay of five consecutive days in the intelligent Dormitory (iDorm) which is a real ubiquitous computing environment test bed. Both offline and online experimental results are presented comparing the performance of our technique with other approaches. The results show that our proposed system has outperformed the other systems while operating online in a life long mode to realise the ambient intelligence vision.
Conference Paper
Full-text available
In this paper, we will present a novel system for learning and incrementally adapting type-2 Fuzzy Logic Controllers (FLCs) for agents embedded in Ambient Intelligent Environments (AIEs). The system learns the rules and the type-2 Membership Functions (MFs) for the type-2 FLC that models the user behavior. Over long term operations, the agent incrementally adapts the type-2 FLC rules and MFs in a life long learning mode to accommodate for the short term and long term uncertainties encountered in AIEs. We will present unique experiments carried out by different users over the course of the year in the Essex intelligent Dormitory (iDorm) which is a real AIE test bed. We will show how the type-2 agent learnt and adapted to the occupant's behavior, whilst handling the encountered short term and long term uncertainties to give a very good performance that outperformed the type-1 fuzzy agents while using smaller rule bases.
Conference Paper
Full-text available
In interactive systems, the term rendering applies to any form of communication directed from the application towards the users. The present paper deals with the specification of rendering, and its relationship with the formal specification of the dialogue between application and user. We first present a taxonomy of rendering according to its function in the application. We briefly recall the basics of the ICO formalism, which is used for the formal specification of the application. We then present a case study illustrating how various categories of rendering are taken into account in the ICO formalism. Lastly, we show how mathematical analysis can be performed on the ICO models to verify predictability properties of the interactive system. 1. Introduction Although a lot of work has recently been devoted to the formal specification of interactive systems, the presentation component has received much less consideration than the dialogue component. However, as stated in [10], "arguments...
Conference Paper
Full-text available
This paper presents a multimodal dialogue controller which can be integrated in a MUIMS (Multimodal User Interface Management System). The well-known A.T.N. (Augmented Transition Networks) model [3] is used to represent the multimodal grammar of a user interface. This type of model has been used before to speci~ monomodal user interfaces [4] [5]. The work presented here shows it is possible to use the A.T.N. model for multimodal user interfaces by adding specific extensions.
Article
To support the sharing and reuse of formally represented knowledge among AI systems, it is useful to define the common vocabulary in which shared knowledge is represented. A specification of a representational vocabulary for a shared domain of discourse—definitions of classes, relations, functions, and other objects—is called an ontology. This paper describes a mechanism for defining ontologies that are portable over representation systems. Definitions written in a standard format for predicate calculus are translated by a system called Ontolingua into specialized representations, including frame-based systems as well as relational languages. This allows researchers to share and reuse ontologies, while retaining the computational benefits of specialized implementations.We discuss how the translation approach to portability addresses several technical problems. One problem is how to accommodate the stylistic and organizational differences among representations while preserving declarative content. Another is how to translate from a very expressive language into restricted languages, remaining system-independent while preserving the computational efficiency of implemented systems. We describe how these problems are addressed by basing Ontolingua itself on an ontology of domain-independent, representational idioms.
Article
Intelligent multimodal presentation of information aims at using several communication modalities to produce the most relevant user outputs. This problem involves different concepts related to information structure, interaction components (modes, modalities, devices) and context. In this paper we study the influence of interaction context on system outputs. More precisely, we propose a conceptual model for intelligent multimodal presentation of information. This model, called WWHT, is based on four concepts: “What”, “Which”, “How” and “Then”. These concepts describe the life cycle of a multimodal presentation from its “birth” to its “death” including its evolution. On the basis of this model, we present the ELOQUENCE software platform for the specification, the simulation and the execution of output multimodal systems. Finally, we describe two applications of this framework: the first one concerns the simulation of an incoming call in an intelligent mobile phone and the second one is related to a task of marking out a target on the ground in a fighter plane cockpit.
Conference Paper
Many real-world application domains that demand planning and scheduling support do not allow for a clear separation of these capabilities. Typically, an adequate mixture of both methodologies is required, since some aspects of the underlying planning problem imply consequences on the scheduling part and vice versa. Several integration efforts have been undertaken to couple planning and scheduling methods, most of them using separate planning and scheduling components which iteratively exchange partial solutions until both agree on a result. This paper presents a framework that provides a uniform integration of hybrid planning –the combination of operator based partial order planning and abstraction based hierarchical task network planning– and a hierarchical scheduling approach. It is based on a proper formal account of refinement planning, which allows for the formal definition of hybrid planning, scheduling, and search strategies. In a first step, the scheduling functionality is used to produce plans that comply with time restrictions and resource bounds. We show how the resulting framework is thereby able to perform novel kinds of search strategies that opportunistically interleave what used to be separate planning and scheduling processes.
Conference Paper
We present a domain-independent approach to plan repair in a formal framework for hybrid planning. It exploits the generation process of the failed plan by retracting decisions that led to the failed plan fragments. They are selectively replaced by suitable alternatives, and the repaired plan is completed by following the previous generation process as close as possible. This way, a stable solution is obtained, i.e. a repair of the failed plan that causes minimal perturbation.