Collaborative Explanation and Response in Assisted Living

Environments Enhanced with Humanoid Robots

Antonis Bikakis

, Patrice Caire

, Keith Clark

, Gary Cornelius

, Jiefei Ma

, Rob Miller

Alessandra Russo

and Holger Voos

2, ∗

Department of Information Studies, University College London, London WC1E 6BT, U.K.

Interdisciplinary Centre for Security, Reliability and Trust, University of Luxembourg, L-1359 Luxembourg, Luxembourg

Department of Computing, Imperial College London, London SW7 2RH, U.K.

Keywords:

Assisted Living, Distributed Reasoning, Evidence Gathering, Explanation Generation, Service Robots.

Abstract:

An ageing population with increased social care needs has provided recent impetus for research into assisted

living technologies, as the need for different approaches to providing supportive environments for senior citi-

zens becomes paramount. Ambient intelligence (AmI) systems are already contributing to this endeavour. A

key feature of future AmI systems will be the ability to identify causes and explanations for changes to the

environment, in order to react appropriately. We identify some of the challenges that arise in this respect,

and argue that an iterative and distributed approach to explanation generation is required, interleaved with di-

rected data gathering. We further argue that this can be realised by developing and combining state-of-the art

techniques in automated distributed reasoning, activity recognition, robotics, and knowledge-based control.

1 INTRODUCTION

Electronic health services give opportunities for pro-

viding better care, particularly for the elderly. But

their development gives rise to a number of research

challenges, for example in terms of privacy, user-

friendliness, conviviality and security. A prominent

area of research and development has been in Ambi-

ent Intelligence (AmI) systems, in particular for as-

sisted living environments. The long term vision of

the AmI research community is to provide systems

that intelligently and unobtrusively assist human in-

habitants with tasks in their everyday environment.

This environment is dynamic and complex, and in or-

der to operate effectively, an AmI system must have

some ability to identify and explain the events occur-

ring within it, sometimes in the face of incomplete,

uncertain or seemingly inconsistent information.

Humans have a number of key abilities for coping

with events in their environment. One is the ability

to quickly ﬁlter out events of interest or that need a

response from the normal environmental background.

Another is the ability to identify possible causes or ex-

planations for such events, and then act appropriately

∗

Authors are listed alphabetically.

to eliminate or conﬁrm them. This often involves tak-

ing action to obtain further information. In this pa-

per we propose an approach to AmI systems that mir-

rors these abilities, using state-of-the art techniques in

automated distributed reasoning, activity recognition,

robotics, and knowledge-based control. Our particu-

lar focus is on directed data gathering, triggered by

the generation of tentative explanations. Robots are

a particularly useful tool in this respect, because of

their multi-functional and mobile capabilities.

The research questions we wish to address are: (1)

how to identify events of interest, e.g. ones that could

lead to emergency situations, (2) how to determine

the causes of these events, and (3) how to then decide

what to do. Although traditionally these problems

are handled separately and sequentially, we propose

an iterative process which interleaves these activities,

using a knowledge-based approach underpinned by a

distributed, abductive inference engine.

The remainder of the paper is structured as fol-

lows. Section 2 illustrates the challenges in address-

ing the above questions with an example scenario.

Section 3 summarises the current state of the art, Sec-

tion 4 presents our proposed methodology, and Sec-

tion 5 concludes and outlines our next steps.

506

Bikakis, A., Caire, P., Clark, K., Cornelius, G., Ma, J., Miller, R., Russo, A. and Voos, H.

Collaborative Explanation and Response in Assisted Living Environments Enhanced with Humanoid Robots.

DOI: 10.5220/0005823405060511

In Proceedings of the 8th International Conference on Agents and Artiﬁcial Intelligence (ICAART 2016) - Volume 2, pages 506-511

ISBN: 978-989-758-172-4

2 OUR VISION

2.1 Running Example

Eighty-year-old Wally lives alone, with health de-

clining. An intelligent, agent-based home care sys-

tem has been installed to help him, including an

alarm connected to a local call centre that Wally can

trigger himself while at home, and two humanoid

Robotic Care Assistants (RoCAs). These small,

gentle-looking mobile humanoid robots rest at charg-

ing stations unless tasked by the system. One day

at 11am the automated Home Care Agent (HCA) re-

ceives an input indicating that Wally has pressed his

alarm button. The HCA queries the help centre to ask

if the operators there have managed to communicate

with Wally via his home or mobile phones. They have

not, and so are waiting the pre-agreed 10 minutes for

the system to generate explanations on which to act.

The HCA therefore tries to generate explanations

for the alarm. It asks the two RoCAs (one upstairs,

one downstairs) to try to locate both Wally and the

alarm button. The latter is painted with a distinc-

tive pattern that the robots can easily recognise. At

the same time the HCA tries to locate Wally’s mo-

bile phone, which is integrated into the system as an

agent, using GPS. It locates this 0.5km away, moving

up Fairtree Road at 30km/hr. The mobile’s log does

not show any recent outgoing calls, and conﬁrms that

an unanswered call was made from the call center. It

also indicates that the mobile has been set to “silent”.

The HCA’s proﬁle of Wally contains the following

information: (i) Wally habitually switches his mobile

to silent and does not answer calls while on public

transport, (ii) a list of places that Wally often vis-

its, including the house of his friend Ernie. Wally’s

proﬁle states that he habitually takes the no. 27 bus

along Fairtree Road when visiting Ernie, so the sys-

tem generates and ranks the following explanation as

highly likely: The alarm was triggered erroneously or

is faulty, because Wally is on the bus to visit a friend.

Meanwhile, neither robots nor house cameras

have been able to locate Wally. A house camera re-

ports a location for the alarm button, but a nearby

robot checks then refutes this. So the HCA sends a

message to the help centre with Ernie’s contact details

and a suggestion that they call him to ask if Wally is

visiting. Ernie conﬁrms that a visit is planned, and

shortly after phones back to conﬁrm Wally’s arrival.

An expensive ambulance call-out is thus avoided.

Finally the HCA double-checks via a text ex-

change with Wally that he did not press his alarm,

then sends a text to the alarm system maintenance ser-

vice asking for an appointment to check the system.

2.2 Research Challenges

In cases like the running example, answering the three

questions we posed in Section 1 brings about several

research challenges. Although in this example the in-

put indicating that Wally has pressed the alarm but-

ton can easily be identiﬁed as an event that requires

further investigation, identifying events of interest is

not always straightforward. Furthermore, detecting

the causes of such events, and deciding how to react

to them in such complex environments requires han-

dling challenges such as:

• How to decide which from the possibly large

amount of available information is relevant to the

event. In our example: how can the system de-

termine that the fact that Wally’s mobile phone is

silent is relevant to explaining the triggered alarm.

• How to collect the relevant information. In the

running example locating the alarm button inside

the house requires the two robots to move and

search in places that may not be reached by static

sensors, and to recognise the distinctive pattern

painted on the alarm button.

• How to combine the available information to gen-

erate possible explanations, especially when this

information comes from diverse sources and in

different formats. In our example, the conclusion

that Wally took bus 27 to visit Ernie is based on

a combination of different types of information

such as Wally’s location, recent calls and habits,

coming from different sources, i.e. Wally’s mo-

bile phone and the HCA.

• How to deal with imperfections in the available

information, e.g. inaccuracy, incompleteness or

conﬂicts. In the running example, the house

camera and a robot report conﬂicting information

about the location of the alarm button.

• How to make the system robust to failures. In our

example, the system should be able to generate

sensible explanations for the triggered alarm even

if one of the RoCAs or Wally’s mobile phone fails.

• How to protect the privacy of all involved people.

In our scenario, Ernie must have given his consent

to Wally’s HCA to share his contact details with

the help centre in case of emergency.

• How to react to events taking into account the in-

dividual requirements of all involved parties. In

our example, apart from the overall goal of tak-

ing care of Wally, Wally may wish to keep his

personal information private, while the system

managers may be required to run the system ef-

ﬁciently.

Collaborative Explanation and Response in Assisted Living Environments Enhanced with Humanoid Robots

507

3 STATE OF THE ART

In the literature, one can ﬁnd several proposals for

reasoning with imperfect information in AmI envi-

ronments (Bettini et al., 2010). Many of them use

Machine Learning (ML) techniques such as Bayesian

Networks (Petzold et al., 2005), Time Series predic-

tions (Das et al., 2002), Markov Models (Gellert and

Vintan, 2006) and Neural Networks (Pansiot et al.,

2007) for speciﬁc reasoning tasks such as identifying

a particular user activity. Most recently, the use of Re-

inforcement Learning has been proposed to provide

an “implicit feedback loop” between context predic-

tion and actuation decisions (Boytsov and Zaslavsky,

2010). Although ML solutions are acceptably accu-

rate, due to the lack of an explicit knowledge rep-

resentation, they cannot provide high-level explana-

tions for automatically chosen actions, nor can they

be easily extended and modiﬁed in a modular fashion

by the users or managers of an AmI system. More-

over, they cannot cope with run-time dynamicity, as

any enhancement to the system or changes to the en-

vironment necessitate re-training of the system.

Rule-based approaches overcome some of the

above limitations. A variety of decidable and

tractable formalisms can be used to create knowledge-

based domain models, thus facilitating efﬁcient, for-

mal reasoning about context (Broda et al., 2009).

Their formality, expressiveness, modularity and ex-

tensibility allow them to better satisfy the needs for

interoperability among heterogeneous components,

adaptability to changes in the environment, and main-

tainability of large knowledge bases (Bikakis and An-

toniou, 2010b).

Most current rule-based solutions adopt a cen-

tralised reasoning approach, but the need for dis-

tributed reasoning has also been acknowledged

mainly for better scalability and robustness. Most

distributed approaches, though, are either not fully

decentralised, e.g. knowledge is distributed, but rea-

soning is local and agents do not exchange context

information (Rom

an et al., 2002); or are limited in

their reasoning capabilities. For example, reasoning

in (Viterbo and Endler, 2012) is distributed in that dif-

ferent computational nodes cooperate to infer a global

context state; the reasoning process is however lim-

ited to ontology-based inference, and is not resilient

to inaccurate or conﬂicting information. Finally, all

existing knowledge-based reasoning approaches as-

sume a single direction information ﬂow: from sensor

input, through to data analysis and decision-making,

and then to reaction and control, irrespective of the

extent to which each of these phases are decentralised

(Snchez-Garzn et al., 2012; Valero et al., 2013).

distributed inference, including

(partial) explanation generation

and decision-making

distributed knowledge corpus

distributed control processes

including directed data gathering

distributed

data interpretation

control decisions

control queries

high-level events

Figure 1: Top Level Architecture.

4 PROPOSED APPROACH

The proposed approach would integrate (1) com-

putational logic techniques for distributed reasoning

and explanation generation in the presence of partial

knowledge, inconsistency and noisy data, (2) control

procedures capable of performing proactive and ro-

bust responses of the system in an interleaved man-

ner with the event-driven knowledge-based inference

and (3) enhanced cognitive-driven robotic capabilities

such as visual scene understanding and privacy and

security awareness. Figure 1 illustrates our top level

architecture, where ‘S’s and ‘A’s signify various sen-

sors and actuators in the environment.

4.1 Distributed Defeasible Abduction

The ﬁrst component of our proposed system would

extend the DAREC engine, proposed in (Ma et al.,

2010; Ma et al., 2011), with contextual defeasible in-

ference (Bikakis and Antoniou, 2010a; Bikakis et al.,

2011). DAREC is a general-purpose system that

permits collaborative reasoning between agents over

decentralised incomplete knowledge. It is particu-

larly suited to compute collaboratively explanations

of given observations. For instance, in cognitive

robotics it can be used to collectively abduce expla-

nations, in terms of descriptions of the world, from

sensor data. Agents can recruit other agents on-the-

ﬂy, based on their local knowledge and reasoning

capabilities, and recover from other agents’ failures

during a distributed computation task. The DAREC

distributed algorithm is ﬂexible and efﬁcient, able

to perform constraint satisfaction and customisable

ICAART 2016 - 8th International Conference on Agents and Artiﬁcial Intelligence

508

with application-dependent coordination strategies to

better balance collaborative inference and inter-agent

communication according to the particular domain

and computational infrastructure. However, one of

the main underlying assumptions is global consis-

tency. An explanation has to be guaranteed to be

consistent with respect to observed information, con-

straints and local knowledge of the agents involved

in the collaborative inference. In practice, conﬂicting

information may occur when integrating information

from different sources.

Contextual Defeasible Logic (CDL), on the other

hand has been used, also in a decentralised manner,

to tackle the issue of distribute deductive inference in

the presence of inconsistent information using non-

monotonic inference and priorities among the con-

ﬂicting chains of reasoning (arguments) (Bikakis and

Antoniou, 2010a; Bikakis et al., 2011). Such priori-

ties may represent different levels of conﬁdence in the

content of the arguments or different levels of trust in

the information sources. These approaches, however,

cannot compute collaborative explanations from dis-

tributed observations. We propose overcoming these

two limitations by reformulating the DAREC dis-

tributed abductive inference mechanism in CDL. The

main idea is to be able to collectively compute expla-

nations of distributed observations but also determin-

ing the contextual information for which that explana-

tion is plausible. This would for instance enable the

generation of an explanation from the sensor data, that

Wally is not in the house despite the camera reporting

an indoor location for the alarm button.

4.2 Knowledge Representation

To communicate effectively, the agents of our pro-

posed system would need to share a common vocab-

ulary and ontology for referring to objects in their en-

vironment and the relationships between them. De-

veloping such an ontology, and expressing sufﬁcient

“commonsense” knowledge about the environment

with it, is itself a major research challenge. The on-

tological framework for expressing such knowledge

should facilitate basic spatial, temporal and causal

reasoning. For example, the system might at some

point need to utilise the knowledge that spilling boil-

ing water onto an electrical device (e.g. Wally’s alarm

button) typically damages the device, and causes it

to be hot and wet, and to remain wet for some time.

The Event Calculus (EC) (Kowalski and Sergot, 1986;

Miller and Shanahan, 2002) is a prime candidate for

this type of knowledge representation and reason-

ing. The EC is a logical mechanism for inferring the

cumulative effects of a sequence of events recorded

along a time line, or (used abductively (Shanahan,

1989)) inferring possible causes of a temporal se-

quence of observations (e.g. sensor readings). The

EC has been extended with several features that are

particulaly relevant in the present context. One is

the ability to infer compound or “high level” events

(e.g. ‘Wally has gone upstairs’) from a sequence of

“smaller” events (e.g. the triggering of movement sen-

sors in sequence from the bottom to top of the stair-

case) (Artikis and Paliouras, 2009; Alrajeh et al.,

2013). Another is the ability to reason epistemically

about the agent’s own future knowledge should it per-

form sensing or data-gathering actions (e.g. phoning

Ernie will result in knowing whether Wally is visiting

him) (Ma et al., 2013). For these reasons, our pro-

posed system would embed EC-style ontology inside

the agent’s knowledge.

4.3 Knowledge based Control

To address the ﬁrst three challenges in Section 2.2 our

proposal would allow for proactive multi-agent ex-

planation generation and evidence gathering. As de-

scribed in Section 4.1, distributed defeasible abduc-

tive reasoning would enable multi-agent computation

of context-dependent explanations to distributed ob-

servations in the presence of uncertain and conﬂict-

ing information. This process would be most effec-

tive when done proactively and in an interleaved man-

ner with multi-agent evidence gathering. Our under-

lying infrastructure of sensors/actuators network and

mobile robots would be capable of performing proac-

tive data gathering: responding to requests for speciﬁc

data gathering, deemed by the knowledge-based in-

ference to be relevant for computation of more likely

explanations. It could also trigger requests for infer-

ence tasks for knowledge-driven control.

Knowledge-driven control can be achieved using a

multi-thread agent architecture that embeds Nilsson’s

Teleo-Reactive (TR) procedures (Nilsson, 1994) for

robot control. A TR procedure is an ordered sequence

of condition-action rules, in which the conditions can

access the current agent’s store of observed and in-

ferred (i.e. deduced or abduced) dynamic beliefs: sen-

sor percepts, told and remembered beliefs about the

environment and its inhabitants. The rule actions are

either device control actions, calls to TR procedures

including recursive calls, or forking of information

gathering or distrubuted reasoning threads. In each

called procedure the ﬁrst rule with an inferable guard

is ﬁred eventually resulting in device actions or thread

forking. Device actions typically continue until new

device actions are determined. The belief store of an

agent is continuously and asynchronously updated as

Collaborative Explanation and Response in Assisted Living Environments Enhanced with Humanoid Robots

509

the (reactive) control procedures execute. On each up-

date the last rule ﬁrings are reconsidered starting with

the ﬁring of the initially called procedure of each task

thread. This unique operational semantics means that

TR control is robust and opportunistic. If helped a

TR controlled robot will automatically skip actions,

if hindered it will redo actions.

Reasoning about the expected effects of control

gives the system a means for intelligently monitor-

ing its own performance. For instance, integration

of sensing/actuation and knowledge-base inference

may enhance the visual perceptions of mobile robots.

Computer vision tasks (e.g. 3D reconstruction, object

recognition and tracking), can be complemented with

knowledge inference to provide the service robot with

the ability to draw conclusions, generate plausible ex-

planations from what it sees and gain real-time vi-

sual scene understanding. The robot would be able

to reason about the objects in the environment and its

inhabitants, explain current observations, or even re-

construct a narrative of the scene by deriving and us-

ing information that it cannot see, thus making more

intelligent decisions about its actions.

Communication between multi-agent components

could make use of an agent acquaintance model ac-

quired using publish/subscribe (Robinson and Clark,

2010) to recruit agents likely to be able to contribute

to an inference task. As new agents are added to the

system, they subscribe for event notiﬁcations of in-

terest to them, updating subscriptions to reﬂect focus

of interest. These agents then exert control over the

monitored system by posting action request notiﬁca-

tions to be routed to other agents and devices. No

component needs to know the identities of other com-

ponents, or even what other components there are. All

that has to be decided is the ontology for notiﬁcations

and subscriptions.

4.4 Robotics Considerations

In applications such as ours, hardware considerations

have to be taken into account. To satisfy the system

requirements our service robots would include differ-

ent kinds of capabilities from efﬁciency and cost to

conviviality (Caire et al., 2011). So we need to exam-

ine a range of potential service robots which could

fulﬁl the requested tasks. For example, to provide

the basic functionality of looking for Wally in case

of emergency, the simplest robot could consist of a

Kinect camera attached to a mobile robot platform.

For a more sophisticated robot, a tablet attached to a

pole could be added as a feature. This would enhance

its interactions with Wally, and provide a basis to be

used as a healthcare assistant. A higher end robot

could consist of a humanoid robot with more subtle

and varied interaction capabilities, possibly equipped

with 3D camera. Such a service robot could be poten-

tially be used as a personal companion at home.

4.5 Privacy and Security

Of course, and particularly in AAL and scenarios

such as ours, privacy and security issues are always

a concern. Our approach in this domain is to ﬁrst,

ensure that communication on the health network

is encrypted. Any secure method, such as Trans-

port Layer Security (TLS) and Secure Sockets Layer

(SSL) would be appropriate. The TLS protocol al-

lows distributed applications to communicate across

a network in such a way as to prevent eavesdropping

and tampering, while the SSL ensures conﬁdentiality,

integrity, and authenticity of individual packets.

Second, strict access policy can be put in place

to ensure that, for example, only doctors have direct

access to patients. Provisions could then be set to,

for example, relax the strict policy and allow doc-

tors to, under certain conditions, share access, or part

of it, with others, e.g. nurses. Approaches such as

Role-Based Access Control or Attribute-Based Ac-

cess Control could be used, e.g. (Kateb et al., 2014).

The latter method offers the interesting perspective

whereby a subject’s requests to perform operations on

objects are granted or denied based on the attributes

of the subject, the object and the environment.

In our scenario, the robots only navigate within

the house. Privacy with respect to video and sound

is therefore not an issue. Furthermore, they trans-

mit video and sound solely when there is an emer-

gency. Areas such as the bathroom, could by default,

i.e. when no emergency has been triggered, be de-

clared no-go areas. In all other cases the video and

sound data would only be used locally by the robot to

interact with its environment.

As for the recording devices themselves, blurring

ﬁlters may be used on cameras to allow only a general

view of the scene, i.e. no details. This could be used

in some cases, such as alarms, to check whether Wally

is in the bathroom. As for microphones, switches

have to be implemented with an off default setting.

Indeed, it is preferable to keep microphones switched

off at all times except for emergencies, e.g. where the

robots are looking for Wally.

5 CONCLUSION

In this paper we argue that the ability of an AmI sys-

tem to explain the causes of perceived events in its en-

ICAART 2016 - 8th International Conference on Agents and Artiﬁcial Intelligence

510

vironment is key to its success, and is best achieved by

an iterative process of tentative explanation genera-

tion interleaved with directed evidence gathering. We

further argued that a number of recently developed

knowledge-based technologies and methods could be

extended and combined into a next generation of AmI

systems in the area of assisted living, in particular

utilising the latest generation of mobile robots act-

ing as agents in a distributed computational setting.

We outlined some of the associated research chal-

lenges and described a composite approach to their

solution using the latest methods in distributed ab-

ductive and defeasible reasoning, teleoreactive con-

trol, and event-based knowledge representation. Such

a system would capitalise on the advantages of a dis-

tributed, knowledge-based approach, such as trans-

parency of computation, easy adaptability and extend-

ability, no single point-of-failure, and closeness of ﬁt

with human-level reasoning.

REFERENCES

Alrajeh, D., Miller, R., Russo, A., and Uchitel, S. (2013).

Reasoning about triggered scenarios in logic program-

ming. In Proc. ICLP’13.

Artikis, A. and Paliouras, G. (2009). Behaviour Recogni-

tion Using the Event Calculus. Artiﬁcial Intelligence

Applications and Innovations III, pages 469–478.

Bettini, C. et al. (2010). A survey of context modelling and

reasoning techniques. Pervasive and Mobile Comput-

ing, 6(2):161 – 180.

Bikakis, A. and Antoniou, G. (2010a). Defeasible Con-

textual Reasoning with Arguments in Ambient Intel-

ligence. IEEE Transactions on Knowledge and Data

Engineering, 22(11):1492–1506.

Bikakis, A. and Antoniou, G. (2010b). Rule-based con-

textual reasoning in ambient intelligence. In Proc.

RuleML 2010, volume 6403 of LNCS, pages 74–88.

Springer.

Bikakis, A., Antoniou, G., and Hassapis, P. (2011). Strate-

gies for contextual reasoning with conﬂicts in Ambi-

ent Intelligence. Knowledge and Information Systems,

27(1):45–84.

Boytsov, A. and Zaslavsky, A. B. (2010). Extending con-

text spaces theory by proactive adaptation. In Proc.

NEW2AN 2010, volume 6294 of Lecture Notes in

Computer Science, pages 1–12. Springer.

Broda, K., Clark, K., Miller, R., and Russo, A. (2009).

Sage: A logical agent-based environment monitoring

and control system. In Proc. AmI 2009, volume 5859

of LNCS, pages 112–117. Springer.

Caire, P., Alcade, B., van der Torre, L., and Sombattheera,

C. (2011). Conviviality measures. In Proc. AAMAS

2011.

Das, S. K., Cook, D. J., Bhattacharya, A., III, E. O. H., and

Lin, T.-Y. (2002). The role of prediction algorithms in

the mavhome smart home architecture. IEEE Wireless

Communications, 9(6):77–84.

Gellert, A. and Vintan, L. (2006). Person movement pre-

diction using hidden markov models. Studies in Infor-

matics and Control, 15(1):17–30.

Kateb, D. E., Zannone, N., Caire, P., Moawad, A., Nain, G.,

Mouelhi, T., and Traon, Y. L. (2014). Conviviality-

driven access control policy. Requirements Engineer-

ing Journal, pages 1–20.

Kowalski, R. A. and Sergot, M. J. (1986). A Logic-Based

Calculus of Events. New Generation Computing,

4:67–95.

Ma, J., Miller, R., Morgenstern, L., and Patkos, T. (2013).

An epistemic event calculus for asp-based reasoning

about knowledge of the past, present and future. In

Proc. LPAR-19.

Ma, J., Russo, A., Broda, K., and Lupu, E. (2010). Dis-

tributed abductive reasoning with constraints. In Proc.

DALT 2010.

Ma, J., Russo, A., Broda, K., and Lupu, E. (2011). Multi-

agent conﬁdential abductive reasoning. In Proc. ICLP

2011.

Miller, R. and Shanahan, M. (2002). Some Alternative For-

mulations of the Event Calculus. In Computational

Logic: Logic Programming and Beyond, Essays in

Honour of R. A. Kowalski, Part II, pages 452–490.

Nilsson, N. (1994). Teleo-reactive programs for agent

control. Journal of Artiﬁcial Intelligence Research,

1:139158.

Pansiot, J., Stoyanov, D., McIlwraith, D. G., Lo, B., and

Yang, G. (2007). Ambient and wearable sensor fusion

for activity recognition in healthcare monitoring sys-

tems. In Proc. BSN 2007, pages 208–212. Springer.

Petzold, J., Pietzowski, A., Bagci, F., Trumler, W., and Un-

gerer, T. (2005). Prediction of indoor movements us-

ing bayesian networks. In Proc. LoCA 2005, volume

3479 of LNCS, pages 211–222. Springer.

Robinson, P. and Clark, K. (2010). Pedro: a Pub-

lish/Subscribe Server using Prolog Technology. Soft-

ware: Practice and Experience, 40(4):313329.

Rom

an, M., Hess, C., Cerqueira, R., Ranganathan, A.,

Campbell, R. H., and Nahrstedt, K. (2002). A middle-

ware infrastructure for active spaces. IEEE Pervasive

Computing, 1(4):74–83.

Shanahan, M. (1989). Prediction is deduction but explana-

tion is abduction. In Proc. IJCAI’89.

Snchez-Garzn, I., Milla-Milln, G., and Fernndez-Olivares,

J. (2012). Context-aware generation and adaptive

execution of daily living care pathways. In Proc.

IWAAL 2012, volume 7657 of LNCS, pages 362–370.

Springer.

Valero, M. n., Bravo, J., Garca, J. M., de Ipia, D. L., and

Gmez, A. (2013). A knowledge based framework to

support active aging at home based environments. In

Proc. IWAAL 2013, volume 8277 of LNCS, pages 1–8.

Springer.

Viterbo, J. and Endler, M. (2012). Decentralized Reasoning

in Ambient Intelligence. Springer Briefs in Computer

Science. Springer.

Collaborative Explanation and Response in Assisted Living Environments Enhanced with Humanoid Robots

511