MODELLING USER BEHAVIOUR WHILE DRIVING AN

INTELLIGENT WHEELCHAIR

Carsten Fischer, Shi Hui, Cui Jian

SFB/TR8 Spatial Cognition, University of Bremen, Bremen, Germany

Frank Schafmeister, Nils Menrad, Nicole v. Steinb

uchel

Medizinische Psychologie und Medizinische Soziologie, Georg-August-University, G

ottingen, Germany

Kerstin Schill, Bernd Krieg-Br

uckner

SFB/TR8 Spatial Cognition, University of Bremen; DFKI, Bremen,Germany

Keywords:

User modelling, Service robot, Empirical study, Safety assistant, Intelligent wheelchair.

Abstract:

This paper reports on our user modelling work based on an empirical study of users’ behaviour while driving

a power wheelchair with a safety assistant. The focus was on persons with visual perceptional deﬁcits, who

show deﬁcits to control a power wheelchair safely if no additional assistance system is available. Our major

concern is to identify users’ behavioural patterns which lead to the intervention of the safety assistant, although

the empirical study itself covers many different research aspects. The goal of the current work is twofold: to

adapt the safety assistance for users with hemianopsia who lost one half of their visual ﬁeld and to improve

their navigation skill in narrow space via suitable interaction.

1 MOTIVATION

The major goal of intelligent living assistance systems

is to extend an independent and self-consistent life

of persons with physical or cognitive deﬁcits, or of

elderly people, thus enhancing their life quality and

reducing their dependency on personal health care

(Nehmer et al., 2006). An intelligent living assistant

should meet several requirements, for example, to be

personalized to the user’s needs, adaptive to the user’s

actions and environment, anticipating the user’s de-

sires. In addition, new forms of interaction are indis-

pensable for an intelligent living assistant to compen-

sate for users’ deﬁcits and to extend their activities

(Riva, 2005; Morganti and Riva, 2005). Therefore,

the focus of technology research should be on sup-

porting people to cope with their requirements in an

effective and transparent way, along with the develop-

ment of speciﬁc new technologies. This requires the

identiﬁcation of users’ characteristics (Morganti and

Riva, 2005), especially, modelling users’ behaviour

plays an important role in developing intelligent liv-

ing assistance systems like therapy robots (Lee et al.,

2004; Burgar et al., 2000).

Empirical studies of intelligent living assistance

robots with real users have rarely been reported in

the literature. To date several empirical studies on

human-robot interaction have been published (Kanda

et al., 2006). However, to carry out an user based

study with a living assistant, additional conditions

should be taken into consideration. For example, it

is necessary to identify potential user groups based

on medical, cognitive, sensory-motor and epidemio-

logical criteria. The deﬁnition of application scenar-

ios may provide us with exemplary views of their ap-

plications (Hansen et al., 2006), but empirical studies

can help us to identify users’ behaviour while carrying

out pre-deﬁned activities, and can thus be used to im-

prove important features of intelligent assistance sys-

tems such as personalization, adaptivity, acceptance

and feasibility.

Our main concern in the current work is on the

evaluation of a power wheelchair with a safety as-

sistant, which has been developed in our institute

(Lankenau and R

ofer, 2001; Krieg-Br

uckner et al.,

pear), and on the identiﬁcation of users’ behavioural

patterns directly before the interventions of the safety

assistant. The goal is to ﬁnd out whether patients with

hemianopsia, neglect and with motor deﬁcits, are able

to drive a power wheelchair equipped with a safety as-

330

Fischer C., Hui S., Jian C., Schafmeister F., Menrad N., Steinbüchel N., Schill K. and Krieg-Brückner B. (2010).

MODELLING USER BEHAVIOUR WHILE DRIVING AN INTELLIGENT WHEELCHAIR.

In Proceedings of the Third International Conference on Health Informatics, pages 330-336

DOI: 10.5220/0002747803300336

 SciTePress

sistant in their daily living condition, who would oth-

erwise not be able to drive a wheelchair at all. The

aim is also to adapt the safety assistant for supporting

such users more intelligently.

This paper is structured as follows: We begin in

Section 2 with an introduction to the wheelchair and

the empirical study. The data analysis consists of two

major steps: pre-processing and qualiﬁcation, and is

discussed in Section 3. Based on the analysis, we de-

velop several user behaviour patterns focused on ob-

stacle avoidance in Section 4. Before concluding, we

discuss some exceptional situations and possible ap-

plications of these behaviour patterns in Section 5.

2 DO WHEELCHAIR USERS

NEED SAFETY ASSISTANCE:

AN EMPIRICAL STUDY

Rolland (Lankenau and R

ofer, 2001) is an intel-

ligent wheelchair that is equipped with two laser

range sensors on the front and back side, wheel en-

coders recording the speed and direction, and an

onboard computer as extensions of a commercial

power wheelchair. Several assistance functions have

been developed for Rolland. Several similar anti-

collision and guidance systems implemented on pow-

ered wheelchairs and walkers have been reported in

the literature (Dutta and Fernie, 2005; Montesano

et al., 2006; Cortes et al., 2008; R

ofer et al., 2009).

Another approach is proposed by Kulyukin et al who

set up a guidance assistance system which requires

an intelligent environment embedded with inexpen-

sive sensors (Kulyukin et al., 2008).

This work focuses on the safety assistant, which

monitors the surrounding environment using sensor

data gathered by the equipped laser scanners and

brakes in time if an obstacle is dangerously close

to the wheelchair. The user commands are passed

via joystick to the safety assistant and, if no ob-

stacle is inside a predeﬁned safety zone around the

wheelchair, the commands are passed unaltered to

the actuators as target translational and rotational

speeds. If an obstacle is detected inside the safety

zone and the wheelchair is exceeding a safety speed

limit with respect to the remaining distance to the ob-

stacle, the current driving speed and the intended user

commands, the safety assistant reduces the speed of

the wheelchair and brings it to a standstill if neces-

sary. Thus the safety assistant prevents the wheelchair

driver from collisions causing severe injuries or dam-

ages. On the other hand the safety assistant does not

affect the user commands in obstacle free areas attain-

ing a smooth driving behaviour.

Presently, a large group of users with speciﬁc

physical and cognitive disabilities, such as hemianop-

sia patients suffering from the loss of one side of

their visual ﬁeld, are often not allowed to drive power

wheelchairs for safety reasons. In this section we re-

port an empirical study aimed at answering the ques-

tion: “How does the safety assistant support per-

sons suffering from hemianopsia to drive a power

wheelchair?”.

The empirical study took about 4 weeks, in close

cooperation with the Department of Medical Psychol-

ogy at the University of G

ottingen and the St. Mau-

ritius Therapy Clinic Meerbusch. Twelve participants

with different physical and cognitive deﬁcits took part

in the study. One participant in this group suffered

from dementia, three from cerebral palsy; another

three participants suffered from a visual neglect. Five

participants suffered from hemianopsia, but it turned

out that one of them had severe comorbidities. There-

fore we only included test data from four participants

with hemianopsia. Prior to the test runs every partici-

pant had up to ﬁve training sessions of 30 to 45 min-

utes each to get accustomed to a power wheelchair

with or without additional safety assistant.

Figure 1: Plan of the test course.

The test course (see Fig. 1) was 8m wide, 14m

long, and had a stretch of about 35m. The whole

course was delimited by ﬂexible partitions of 80cm

height simulating a long corridor. In the test course

different types of obstacles were arranged. There

were left-right combinations with alternating obsta-

cles of different height, separate low obstacles with

a height of 20cm, and higher obstacles to block the

drivers view onto the following obstacles. In addition,

a wheelchair ramp was placed in the middle of the

course. Thus the obstacle conﬁgurations contained

several critical situations for a wheelchair user with

a restricted visual ﬁeld. Furthermore, the course was

constructed symmetrically, such that it provided the

MODELLING USER BEHAVIOUR WHILE DRIVING AN INTELLIGENT WHEELCHAIR

331

same conditions for the subjects with hemianopsia on

the left or on the right side. The participants were

asked to drive through the test course up to a dead

end representing a narrow elevator. After they had

reached this dead end they had to turn the wheelchair

and return to the starting point.

3 DATA ANALYSIS

3.1 Pre-processing

During the test runs we recorded several wheelchair

data such as the wheelchair translational speeds (de-

noted as OdoSpeedTranslX) and rotational speeds

(i.e. the speed of directional changes, denoted as

OdoSpeedRotation), as well as the joystick com-

mands, including the intended speed and direc-

tion denoted as JoystickSpeed and JoystickDirec-

tion. Furthermore, we recorded the safety speed limit

(MaxSafeSpeed) described in Section 2. Every test

run was also documented with six video cameras

recording the test process from different points of

view, to support the analysis of the driving behaviour

of the participants.

Figure 2: A sample of safety relevant situations.

Persons with hemianopsia are supposed to have

problems identifying obstacles on their blind or not

perceiving side while driving a power wheelchair.

The analysis of the recorded wheelchair data conﬁrms

that participants with hemianopsia had a higher occur-

rence of safety assistant interventions when passing

narrow passages or obstacle conﬁgurations with alter-

nately left and right positioned obstacles. Moreover,

the interventions are concentrated at locations where

participants passed obstacles positioned on the deﬁcit

side. For this reason we decide to use such situations

for our analysis purpose.

Figure 2 shows a sample situation, in which the

odometric data recorded the behaviour of a participant

with hemianopsia on the left side continuously. The

visualized odometric data exhibits that the wheelchair

speed changes from 125 to 30, and from 30 to 120 di-

rectly before the intervention. In the whole time there

is only a slight rotation speed toward the left. The joy-

stick direction changes slightly to the right, and again

to the left. Directly before the intervention the joy-

stick direction is changed to the right again. A global

view of the situation is presented in Figure 3, where

the wheelchair is stopped by the safety assistant.

Figure 3: A participant drives the wheelchair through a

right-left-right obstacle combination while having problems

to pass the second obstacle on the hemianoptic side.

3.2 An Approach to Data Abstraction

Usually, humans explain, perceive and process spa-

tial situations in a qualitative manner, which does

not directly mirror the actual measures in real life.

In Artiﬁcial Intelligence, such mental conceptualiza-

tions of space are formalized and modeled in the sub-

ﬁeld of qualitative spatial representation and reason-

ing. This kind of abstraction can make valuable pre-

dictions about human spatial behavior (Cohn et al.,

1997; Freksa, 1992; Shi and Krieg-Br

uckner, 2008).

The abstraction of the empirical data in the present

work has two additional reasons: to enable a qualita-

tive analysis of the safety relevant situations, and to

identify and specify users’ behaviour patterns.

The recorded wheelchair’s rotation speed has val-

ues between −30 and 30 in millimeters per sec-

ond, and the direction given via the joystick be-

tween −180

◦

and 179

◦

, where the positive values

represent the left side and negative values the right

side. We distinguish them by four qualitative values:

front/back for forward/backward movements or com-

mands, deﬁcit or normal for any rotation or command

toward the subject’s deﬁcit or normal side, respec-

tively. If the subject’s deﬁcit side is on the left, then

the rotation speed −20, for example, is interpreted as

normal, or the direction 50 as deﬁcit. Rotation speed

ranging from −5 to 5 or direction from −10 and 10 is

interpreted as front, and the direction from 170 to 179

or from −180 to −170 as back.

HEALTHINF 2010 - International Conference on Health Informatics

332

Furthermore, four qualitative values have been

introduced to interpret speeds, i.e., standStill,

slowSpeed, midSpeed and fastSpeed. The transla-

tional speed ranges from −400 to 400 millimeters per

second, the speed given via the joystick from −64 to

63, and the maximal safety speed between −3000 and

3000 millimeters per second.

Table 1 contains the mapping between the ranges

of metric speed data and their qualitative correspond-

ings.

Table 1: Abstraction of the metric speeds.

Abstract. standStill slowSpeed midSpeed fastSpeed

Wheel-

chair

speed

(0, 20) (20, 100) (100,

200)

(200,

400)

Joystick

speed

(0, 4) (4, 16) (16, 32) (32, 63)

MaxSafe

Speed

(0, 50) (50, 500) (500,

1000)

(1000,

3000)

We decided on these qualitative representations,

because an abstraction with ﬁner granularity, for ex-

ample, with more qualitative directions or speeds, did

not deliver more information about safety relevant be-

haviour, but increased the complexity of the mod-

elling process. On the other hand, it was impossible

to distinguish any meaningful patterns with a coarser

abstraction.

4 MODELLING USERS’

BEHAVIOUR

4.1 Safety Relevant Situations

A set of situations, which lead to the intervention

of the safety assistant, called safety relevant situa-

tion, has been selected by analyzing the odometric

data. As stated in the last subsection, we ﬁrst in-

terpret these situations qualitatively to enable pattern

identiﬁcation. Each situation contains two safety rel-

evant points: a critical one and an unsafe one. At

the critical point, the user still has enough time to

pass an obstacle located at his/her deﬁcit side with-

out the intervention of the safety assistant. However,

if an unsafe point is reached, the intervention is indis-

pensable to avoid the collision. At a critical or unsafe

point the safety speed limit has a value of midSpeed

or slowSpeed respectively.

Table 2 shows two samples of critical points de-

tected in test runs of a subject with hemianopsia on

the left side, in which the safety speed limit is reduced

in the middle range and the wheelchair is approach-

ing an obstacle on the left hand side. In the ﬁrst case

the wheelchair moved slowly and rotated slightly to

the left, and the command given by the subject was

to drive straight with a moderate speed. The second

case describes a situation, in which the wheelchair

moved slowly forward, and the subject commanded

the wheelchair to move slowly toward the right front.

Table 2: Sample critical points: the subject has the deﬁcit

side on the left.

OdoSpeed

TranslX

OdoSpeed

Rotation

Joystick

Speed

Joystick

Direction

1 slowSpeed leftSlow midSpeed front

2 slowSpeed front slowSpeed rightFront

A critical point may be changed to an unsafe one

by joystick commands given by the user, see the two

examples in Table 3, which followed the two cases in

Table 2. They were reached by giving the left front as

the next direction in the subject’s last command.

Table 3: Sample unsafe points resulting from leftFront as

next direction.

OdoSpeed

TranslX

OdoSpeed

Rotation

Joystick

Speed

Joystick

Direction

1 slowSpeed leftSlow fastSpeed leftFront

2 fastSpeed rightSlow midSpeed leftFront

4.2 Identiﬁcation of Behaviour Patterns

After analyzing a total number of 21 safety relevant

situations found in the test runs of the four partici-

pants with hemianopsia, we identiﬁed the following

three behaviour patterns, each of which is divided into

a critical and an unsafe part.

4.2.1 Pattern 1

In 12 out of 21 safety relevant situations the follow-

ing behaviour has been identiﬁed: the subject drives

toward his or her hemianoptic side – the side with

the visual deﬁcit – and away from an obstacle that

is well perceived on the non-hemianoptic side. The

critical point of this pattern is speciﬁed as follows: the

wheelchair moves forward or rotates toward the user’s

hemianoptic side and the user keeps commanding the

wheelchair to move in this way. At the same time

the qualitative value for the safety speed limit is re-

duced to a value in the middle speed range because of

an approaching obstacle. If the described movement

MODELLING USER BEHAVIOUR WHILE DRIVING AN INTELLIGENT WHEELCHAIR

333

continues and the value for the safety speed limit is re-

duced to slowSpeed at the same time, the unsafe point

of this pattern is reached (see Figure 2 for an exam-

ple). Table 4 contains the qualitative deﬁnition of the

pattern.

4.2.2 Pattern 2

In four of the 21 situations we have identiﬁed a be-

haviour pattern, in which the subject kept driving to-

ward his/her normal side if possible, in order to avoid

obstacles that may occur at his/her deﬁcit side. If

the user tries to avoid an obstacle located on the nor-

mal side by steering the wheelchair to the other side,

but does not notice an obstacle located on the deﬁcit

side, an intervention of the safety assistant is neces-

sary to avoid a collision with the obstacle located on

the deﬁcit side; see Table 4 for its speciﬁcation. In

this pattern, the change of the critical point to the un-

safe one is caused predominantly by the change of the

joystick direction from the normal side to the deﬁcit

side.

4.2.3 Pattern 3

The third pattern describes situations in which the

subject commanded the wheelchair to carry out an

obstacle avoidance maneuver by driving backwards,

however the safety module intervened to avoid a col-

lision with an obstacle located in the back of the

wheelchair; see Table 4 for its deﬁnition. In this pat-

tern the rotational direction of the wheelchair is in-

signiﬁcant. This pattern is found in 3 out of 21 situa-

tions.

5 DISCUSSION

The three behaviour patterns cover 19 of 21 situations,

but the remaining two situations cannot be identiﬁed

by any reasonable pattern.

The related video material exhibited that the sub-

jects tried to apply several different steering com-

mands to navigate the wheelchair through the narrow

obstacle conﬁguration. These changes of driving di-

rections in a short temporal sequence led to unpre-

dictable orientations of the caster front-wheels. Al-

though the intended steering commands are appro-

priate at that moment, the wheelchair could not fol-

low the subject’s commands in time. The subjects

were confused by the wheelchair’s behaviour, since

they were not aware of the position of the caster

front-wheels. As a consequence, they behaved unpre-

dictably. This phenomenon implies the existence of a

mode confusion, in which the wheelchair is situated in

a state different from the one expected by the subject.

Mode confusions occur typically in shared-control

systems (Thimbleby, 1990), such as the wheelchair.

During the experiment we observed another mode

confusion phenomenon: almost all subjects acted un-

common after the intervention of the safety assistant:

they attempted to regain control over the wheelchair,

but the wheelchair did not drive in the way they ex-

pected; as a result they tried to give the wheelchair

some arbitrary commands via the joystick or pressed

the joystick powerfully, which caused an even less ex-

pected movement of the wheelchair. Mode confusion

situations can be avoided through the improvement of

the system’s behaviour or providing the user with suit-

able clariﬁcation (Thimbleby, 1990).

An interesting application of the critical patterns

developed in Section 4 could be to support the re-

habilitation process of neglect and hemianopsia pa-

tients with cognitive and motor deﬁcits. These pat-

terns enable the wheelchair to inform the user about a

possible collision with an approaching obstacle in ad-

vance, which would otherwise be noticed too late or,

even worse, ignored due to their spatial perceptional

problems. Instead of the current feedback from the

wheelchair, i.e., the disregard of the user’s commands

given via the joystick and the braking process in un-

safe situations, some acoustic signal, for example,

could be used to inform the user of an approaching

obstacle on his/her impaired side. As a consequence

the user could be trained to drive the wheelchair more

smoothly or even without the safety module in a nar-

row environment.

On the other hand, the behaviour patterns provide

us with helpful information to improve the existing

safety assistant. In our data we counted up to 5600

interventions in a single test run by some participants,

which is inconvenient in real life in spite of the guar-

anteed safeness. As a possible solution, the safety as-

sistant can be extended with a module that modiﬁes

the joystick commands if a situation matching an un-

safe pattern has been detected, to compensate users’s

visual deﬁcits, and reduce the number of the interven-

tions of the safety assistant.

Finally, the current work makes it clear that user

studies with an intelligent living assistant and real

users present us with hard challenges. They require

close cooperation of engineers, neuropsychologists,

and other medical professionals. In addition to the ex-

periment design, test environment construction, real-

ization, data analysis and user modelling, the accred-

itation of a homogeneous user group turned out to be

the most difﬁcult one, at least if the group consists of

persons with speciﬁc cognitive or physical deﬁcits. In

HEALTHINF 2010 - International Conference on Health Informatics

334

Table 4: Pattern deﬁnitions.

Pattern 1 Pattern 2 Pattern 3

Critical Unsafe Critical Unsafe Critical Unsafe

MaxSafeSpeed midSpeed slowSpeed midSpeed slowSpeed midSpeed slowSpeed

OdoSpeedTranslX ≥ slowSpeed ≥ slowSpeed ≥ slowSpeed ≥ slowSpeed ≥ slowSpeed ≥ slowSpeed

OdoSpeedRotation deﬁcit, or deﬁcit normal, or deﬁcit - -

front front

JoystickDirection deﬁcit deﬁcit normal deﬁcit back back

fact, we originally planned to carry out the experiment

with 12 patients with hemianopsia, however, over the

duration of the experiment only four patients in the

therapy clinic, who met all the requirements, were

available for our experiment.

6 CONCLUSIONS

The research work presented in this paper is an em-

pirical study with a real user group and an intelli-

gent assistance system. Speciﬁcally, we analyzed the

data collected in the experiment carried out with a

group of persons who are affected by hemianopsia,

and studied their behaviour while driving an intelli-

gent wheelchair passing obstacles located on their im-

paired side. The analysis process consists of the se-

lection of relevant data segments, interpreting the rel-

evant odometric data, and abstracting the data qualita-

tively. As a result three behaviour patterns have been

identiﬁed and discussed.

The application of these behaviour patterns is

twofold. First, it enables the development of an adap-

tive user interface, such that users can learn how to

control an intelligent mobile assistant smoothly, even

though he/she has visual/spatial perceptional prob-

lems. On the other hand, the improvement of the

safety assistant by adding an extra functionality to

compensate the user’s deﬁcits is work to do. In future

experiments, we shall evaluate whether the driving

assistant, which has been developed to avoid obsta-

cles automatically (Krieg-Br

uckner et al., pear), will

overcome such situations.

As discussed in Section 5, we are going to handle

the mode confusion problems by comparing the user’s

behaviour with that of the wheelchair after the inter-

ventions of the safety assistant. This requires the con-

struction of a user model using the relevant data col-

lected in the experiment introduced in Section 2. On

the other hand, the abstraction of the wheelchair’s be-

haviour is also necessary, such that they can be com-

pared directly for detecting mode confusions. Fur-

thermore, the qualitative models can even be speciﬁed

formally, thus an automatic mode confusion detection

process is possible by using formal techniques (e.g.

model-checking, cf. (Rushby et al., 1999; Heymann

and Degani, 2002; Bredereke and Lankenau, 2002)).

ACKNOWLEDGEMENTS

We gratefully acknowledge the support of the

Deutsche Forschungsgemeinschaft (DFG) through

the Collaborative Research Center SFB/TR8 - Project

I3-[SharC] ”Shared Control via Interaction”.

We would like to thank Prof. Dr. Volker H

omberg

and his team at St. Mauritius Therapy Clinic Meer-

busch for their great support.

REFERENCES

Bredereke, J. and Lankenau, A. (2002). A rigorous view

of mode confusion. In Proc. of Safecomp 2002, 21st

Int’l Conf. on Computer Safety, Reliability and Secu-

rity, number 2434 in Lecture Notes in Computer Sci-

ence, pages 19–31. Springer-Verlag.

Burgar, C. G., Lum, P. S., Shor, P. C., and Van der Loos, H.

F. M. (2000). Development of robots for rehabilitation

therapy: The palo alto va/stanford experience. Journal

of Rehabilitation Res. Development, 37(6):663–674.

Cohn, A. G., Bennett, B., Gooday, J., and Gotts, N. M.

(1997). Qualitative spatial representation and reason-

ing with the Region Connection Calculus. Geoinfor-

matics, 1.

Cortes, U., Martinez-Velasco, A., Barrue, C., Benedico,

T., Campana, F., Fernandez, J., and Annicchiarico, R.

(2008). A SHARE-it service to elders mobility using

the i-walker. Gerontechnology, 7.

Dutta, T. and Fernie, G. R. (2005). Utilization of ultra-

sound sensors for anti-collision systems of powered

wheelchairs. IEEE Transactions on Neural Systems

and Rehabilitation Engineering, 13:24–32.

Freksa, C. (1992). Using orientation information for qual-

itative spatial reasoning. In Theories and Methods

of Spatio-Temporal Reasoning in Geographic Space,

volume 639 of Lecture Notes in Computer Science,

pages 162–178. Springer-Verlag.

MODELLING USER BEHAVIOUR WHILE DRIVING AN INTELLIGENT WHEELCHAIR

335

Hansen, M., Fabian, B., M

oller, J., and Spiekermann, S.

(2006). Szenarien. In Technikfolgenabsch

atzung

Ubiquit

ares Computing und Informationelle Selb-

stbestimmung. Bundesministerium f

ur Bildung und

Forschung, http://www.datenschutzzentrum.de/ tau-

cis/ita taucis.pdf.

Heymann, M. and Degani, A. (2002). Constructing human-

computer interaction in aeronautics. In Proc. of HCI-

Aero 2002.

Kanda, T., Kamasima, M., Imai, M., Ono, T., Sakamoto,

D., Ishiguro, K., and Anzai, Y. (2006). A humanoid

robot that pretends to listen to route guidance from a

human. Autonomous Robots, 22(1):87–100.

Krieg-Br

uckner, B., T. R

ofer, T., Shi, H., and Gersdorf, B.

(to appear). Mobility Assistance in the Bremen Ambi-

ent Assisted Living Lab. Journal of Gerontology. Spe-

cial Section: Technology and Aging: Integrating Psy-

chological, Medical, and Engineering Perspectives.

Kulyukin, V., Kutiyanawala, A., LoPresti, E., Matthews, J.,

and Simpson, R. (2008). iWalker: Toward a rollator-

mounted wayﬁnding system for the elderly. 2008

IEEE International Conference on RFID, pages 303–

311.

Lankenau, A. and R

ofer, T. (2001). A safe and versatile mo-

bility assistant. IEEE Robotics and Automation Mag-

azine, 1:29–37.

Lee, M., Abdullach, H. A., and Basir, O. A. (2004). Model-

driven interactive system design for therapy robots.

Journal of Intelligent and Robotic Systems, 39:345–

363.

Montesano, L., Minguez, J., Alcubierre, J., and Mon-

tano, L. (2006). Towards the adaptation of a robotic

wheelchair for cognitive disabled children. In Pro-

ceedings of the 2006 IEEE/RSJ International Confer-

ence on Intelligent Robots and Systems, pages 710–

716.

Morganti, F. and Riva, G. (2005). Chapter 15: Am-

bient intelligence for rehabilitation. In Riva,

G., Vatalaro, F., Davide, F., and Alcaniz,

M., editors, Ambient Intelligence. IOS Press,

http://www.ambientintelligence.org.

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

(2006). Living Assistance Systems - an Ambient In-

telligence Approach. In Proceedings of the 28th Inter-

national Conference on Software Engineering.

Riva, G. (2005). The psychology of ambient intel-

ligence: Activity, situation and presence. In

Riva, G., Vatalaro, F., Davide, F., and Al-

caniz, M., editors, Ambient Intelligence. IOS Press,

http://www.ambientintelligence.org.

ofer, T., Laue, T., and Gersdorf, B. (2009). iwalker an

intelligent walker providing services for the elderly.

Technically Assisted Rehabilitation 2009.

Rushby, J., Crow, J., and Palmer, E. (1999). An automated

method to detect potential mode confusions. In Proc.

18th IEEE Digital Avionics Systems Conference.

Shi, H. and Krieg-Br

uckner, B. (2008). Modelling Route

Instructions for Robust Human-Robot Interaction on

Navigation Tasks. International Journal of Software

and Informatics, 2(1):33–60.

Thimbleby, H. (1990). User Interface Design. ACM Press,

New York, USA.

HEALTHINF 2010 - International Conference on Health Informatics

336