A METHODOLOGY FOR CREATING INTELLIGENT

WHEELCHAIR USERS’ PROFILES

Brígida Mónica Faria

1,2,5

, Sérgio Vasconcelos

3,5

, Luís Paulo Reis

4,5

and Nuno Lau

Escola Superior de Tecnologia da Saúde do Porto, Instituto Politécnico do Porto, Vila Nova de Gaia, Portugal

Dep. Elect., Telecomunicações e Informática (DETI/UA),

Inst. Eng. Electrónica e Telemática de Aveiro, Universidade de Aveiro, Aveiro, Portugal

Dep. Eng. Informática, Faculdade de Engenharia, Universidade do Porto, Porto, Portugal

Dep. Sistemas de Informação, Escola de Engenharia, Universidade do Minho, Guimarães, Portugal

Laboratório de Inteligência Artificial e Ciência de Computadores, Universidade do Porto, Porto, Portugal

Keywords: Intelligent wheelchair, Users´ profile, Adaptive interface.

Abstract: Intelligent Wheelchair (IW) is a new concept aiming to allow higher autonomy to people with lower

mobility such as disabled or elderly individuals. Some of the more recent IWs have a multimodal interface,

enabling multiple command modes such as joystick, voice commands, head movements, or even facial

expressions. In these IW it may be very useful to provide the user with the best way of driving it through an

adaptive interface. This paper describes the foundations for creating a simple methodology for extracting

user profiles, which can be used to adequately select the best IW command mode for each user. The

methodology is based on an interactive wizard composed by a flexible set of simple tasks presented to the

user, and a method for extracting and analyzing the user’s execution of those tasks. The results achieved

showed that it is possible to extract simple user profiles, using the proposed method. Thus, the approach

may be further used to extract more complete user profiles, just by extending the set of tasks used, enabling

the adaptation of the IW interface to each user’s characteristics.

1 INTRODUCTION

The fraction of population with physical disabilities

has earned more relevance and has attracted the

attention of international health care organizations,

universities and companies interested in developing

and adapting new products. The actual tendency

reflects the demand for an increase on health and

rehabilitation services, in a way that senior and

handicapped individuals might become more and

more independent performing quotidian tasks.

Regardless the age, mobility is a fundamental

characteristic for every human being. Children with

disabilities are very often deprived of important

opportunities and face serious disadvantages

compared to other children. Adults who lose their

independent means of locomotion become less self

sufficient, raising a negative attitude towards them.

The loss of mobility originates obstacles that reduce

the personal and vocational objectives (Simpson,

2005). Therefore it is necessary to develop

technologies that can aid this population group, in a

way to assure the comfort and independence of the

elderly and handicapped people. Wheelchairs are

important locomotion devices for those individuals.

There is a growing demand for safer and more

comfortable wheelchairs, and therefore, a new

Intelligent Wheelchair (IW) concept was introduced.

However, most of the Intelligent Wheelchairs

developed by distinct research laboratories

(Simpson, 2005), have hardware and software

architectures very specific for the used wheelchair

model/developed project and are typically very

difficult to configure in order for the user to start

using them.

The rest of the paper is organized as follows.

Section 2 presents the state of art on intelligent

wheelchairs. Section 3 contains a description of the

users’ interfaces already developed and how the

interface is integrated in our work. Section 4

presents the implementation and methodology for

creating an intelligent wheelchair user interface. The

experiments and the results achieved are presented

in section 5. Finally some conclusions and future

work is described in the last section.

171

Mónica Faria B., Vasconcelos S., Paulo Reis L. and Lau N..

A METHODOLOGY FOR CREATING INTELLIGENT WHEELCHAIR USERS’ PROFILES.

DOI: 10.5220/0003749701710179

In Proceedings of the 4th International Conference on Agents and Artiﬁcial Intelligence (ICAART-2012), pages 171-179

ISBN: 978-989-8425-95-9

 2012 SCITEPRESS (Science and Technology Publications, Lda.)

2 INTELLIGENT

WHEELCHAIRS

In the last years several prototypes of Intelligent

Wheelchairs (Figure 1) have been developed and

many scientific work has been published (Braga et

al., 2009) (Reis et al., 2010) in this area. Simpson

(Simpson, 2005) provides a comprehensive review

of IW projects with several descriptions of

intelligent wheelchairs. The main characteristics of

an IW are (Braga et al., 2009) (Jia, 2007):

interaction with the user using distinct types of

devices such as joysticks, voice interaction, vision

and other sensor based controls like pressure

sensors; autonomous navigation with safety,

flexibility and obstacle avoidance capabilities and

communication with others devices such automatic

doors and other wheelchairs.

The first project of an autonomous wheelchair

for physical handicapped was proposed by Madarasz

in 1986 (Madarasz, 1986). It was planned as a

wheelchair with a micro computer, a digital camera

and an ultra-sound scanner with the objective of

developing a vehicle that could move around in

populated environments without human intervention.

Hoyer and Holper (Hoyer and Holper, 1993)

presented a modular control architecture for an

Omni-directional wheelchair. The characteristics of

NavChair (1996), such as the capacity of following

walls and avoid obstacles by deviation are described

in (Simpson, 1998) (Bell et al., 1994) (Levine,

1999). Miller and Slak (Miller and Slak, 1995)

(Miller, 1998) proposed the system Tin Man I with

three operation modes: one individual driving a

wheelchair with automatic obstacles deviation;

moving through-out a track and moving to a point

(x,y). This kind of chair evolved to Tin Man II

which included advanced characteristics such as

storing travel information, return to the starting

point, follow walls, pass through doors and recharge

battery. Wellman (Wellman et al., 1994) proposed a

hybrid wheelchair equipped with two extra legs in

addition to its four wheels, to allow stair climbing

and movement on rough terrain. FRIEND is a robot

for rehabilitation which consists of a motorized

wheelchair and a MANUS manipulator (Borgerding

et al., 1999) (Volosyak et al., 2005). In this case,

both the vehicle and the manipulator are controlled

by voice commands. Some projects present solutions

for quadriplegic individuals, where facial

expressions recognition is used to control the

wheelchair (Jia et al., 2007) (Ng and Silva, 2001)

(Adachi et al., 1998). In 2002, Pruski presented

VAHM, a user adapted intelligent wheelchair

(Pruski et al., 2002).

Figure 1: Several prototypes of Intelligent Wheelchairs.

Satoh and Sakaue (Satoh and Sakaue, 2007)

presented an omni-directional stereo vision-based

IW which detects both the potential hazards in a

moving environment and the postures and gestures

of a user, using a stereo omni-directional system,

which is capable of acquiring omni-directional color

image sequences and range data simultaneously in

real time. In 2008 John Spletzer studied the

performance of LIDAR based localization for

docking an IW system (Spletzer et al., 2008) and in

2009 Horn and Kreutner (Horn and Kreutner, 2009)

showed how the odometric, ultrasound, and vision

sensors are used in a complementary way in order to

locate the wheelchair in a known environment. In

fact, the research on IW has suffered a lot of

developments in the last few years. Some IW

prototypes are controlled with "thought". This type

of technology uses sensors that pick up

electromagnetic waves of the brain (Hamagami and

Hirata, 2004) (Lakany, 2005) (Rebsamen, 2006).

ICAART 2012 - International Conference on Agents and Artificial Intelligence

172

2.1 IntellWheels Project

This section presents a brief overview of the

Intelligent Wheelchair project that is being

developed at the Faculty of Engineering of the

University of Porto (FEUP) in collaboration with

INESC-P and the University of Aveiro. Also, the

primary results of this project that have already been

published are also presented. The main objective of

the IntellWheels Project is to develop an intelligent

wheelchair platform that may be easily adapted to

any commercial wheelchair and aid any person with

special mobility needs. Initially, an evaluation of

distinct motorized commercial wheelchair platforms

was carried out and a first prototype was developed

in order to test the concept. The first prototype was

focused on the development of the modules that

provide the interface with the motorized wheelchair

electronics using a portable computer and other

sensors.

Figure 2: Generic gamepad, headset with microphone and

Nintendo Wii Remote.

Several different modules have been developed

in order to allow different ways of conveying

commands to the IW. These include, for example,

joystick control with USB, voice commands, control

with head movements and gestures, and facial

expressions recognition (Faria et al., 2007). Figure 2

shows the three commands already available in the

IntellWheels IW. The project research team

considered the difficulty that some patients have

while controlling a wheelchair using traditional

commands such as the traditional joystick.

Therefore, new ways of interaction between the

wheelchair and the user have been integrated,

creating a system of multiple entries based on a

multimodal interface. The system allows users to

choose which type of command best fits their needs,

increasing the level of comfort and safety.

Another possibility enabled by a system of

multiple entries is the use of software for intelligent

control of inputs. This application has the task of

determining the confidence level of each of the

entries, or even cancels them if it detects the

presence of conflicts or noise in the surrounding

environment. For example, in a very dark or very

bright room, where the patient's face is not fully

recognized, the intelligent control of inputs would

decrease the degree of confidence of this type of

software commands sent by the recognition of facial

expressions, and would provide greater importance

to the joystick, voice and/or head movements based

commands.

3 USER PROFILES

In addition to the wide variety of disabilities that

cause different types of limited mobility, each

person has specific characteristics, which may be

related to physical and cognitive factors. Thus,

individuals with similar symptoms may have

significant differences. It is fair to say that

characteristics such as moving the head, ability to

pronounce words, ability to move the hand and

fingers, can vary substantially from individual to

individual. Similarly, the time of learning and

proficiency in using assistive devices may also vary

greatly.

3.1 User Interfaces

The interface between a human and a computer is

called a user interface and it is a very important part

of any computerized system. Moreover, an adaptive

user interface (Langley, 1999) is a software entity

that improves its ability to interact with a user by

constructing a user model based on past experience

with that user. The emerging area of adaptive and

intelligent user interfaces has been exploring

applications in which these paradigms are useful and

facilitate the human machine communication (Ross,

2000). In fact, if an intelligent user interface has a

model of the user, this user model can be used to

automatically adapt the interface. Additionally,

adaptive user interfaces may use machine learning

techniques to improve the interaction with

individuals in order to have the users reach their goal

more easily, faster and with a higher level of

satisfaction. It is also essential for an adaptive

interface to obtain knowledge included in four

distinct domains: knowledge of the user; knowledge

of the interaction (modalities of interaction and

dialogue management); knowledge of the

task/domain; and knowledge of the system

characteristics (Norcio and Stanley, 1989).

A METHODOLOGY FOR CREATING INTELLIGENT WHEELCHAIR USERS' PROFILES

173

3.2 Adaptive Interfaces

Ross (Ross, 2000) presented a comprehensive

classification of adaptive/intelligent interfaces. His

classification contains three main classes. The first

class involves the addition of adaptation to an

existing direct manipulation interface. Examples of

this class are adding extra interface objects in order

to hold the predicted future commands or designing

an interface with multiple commands. The second

class is composed by interfaces acting as an

intermediary between the user and the direct

manipulation interface by filtering information or

generating suggested data values. The third class is

composed by the agent interfaces, in which

autonomous agents (Maes, 1996) (Wooldridge,

2002) can provide pro-active support to the user,

typically can make suggestions and give advice.

It is also mentioned that many intelligent interfaces

can be viewed as adaptive user interfaces, because

they change their behaviour to adapt to an individual

or assignment (Ross, 2000). Another taxonomy

defended by Langley (Langley, 1997) for adaptive

user interfaces (AUI) is based on separating them

into two groups: Informative Interfaces and

Generative Interfaces. The first class selects

information for the user and presents the items he

will find interesting or practical. The second process

tries to generate a useful knowledge structure like

spread-sheets, document preparation or drawing

packages.

Also, in the literature, another class of adaptive

interfaces is presented and studied. This class is

designated as Programming by Demonstration

(Cypher and Halbert, 1994) (Ross, 2000). This class

is distinct from the previous since generative

interfaces produce data values but programming by

demonstration systems produce commands with

arguments.

3.3 Intellwheels User Profile

Tracing a user diagnostic can be very useful to

adjust certain settings allowing for an optimized

configuration and improved interaction between the

user and the multimodal interface.

Accordingly, the Intellwheels Multimodal

Interface should contain a module capable of

performing series of training sessions, composed of

small tests for each input modality. These tests may

consist, for example, of asking the user to press a

certain sequence of buttons on the gamepad, or to

move one of the gamepads' joysticks to a certain

position. Another test may consist in asking the user

to pronounce a set of voice commands, or to perform

a specific head movement. Figure 3 shows where the

user needs to click to start the User Profiler module.

Figure 3: Starting user´s profile module.

The tests should be performed sequentially and

should have an increasing difficulty. Additionally,

the tests should be reconfigurable and extensible.

Finally, the tests sets and theirs results should be

saved on a database, accessible by the Intellwheels

Multimodal Interface. Therefore, the following user

characteristics should be extracted. These

characteristics are separated in two different types:

quantitative and qualitative. The quantitative

measures consist of: the time taken to perform a full

button sequence; the average time between pressing

two buttons; the average time to place a gamepad

analogical switch on a certain position; the average

time to position the head on a certain position; the

trust level of speech recognition; maximum

amplitude achieved with the gamepad analogical

switches in different directions; maximum amplitude

achieved with the head in different directions and

number of errors made using the gamepad. Using the

quantitative measures, the following qualitative

measures should be estimated: user ability to use the

gamepad buttons; user ability to perform head

movements and user ability to pronounce voice

commands.

At the end of the training session, the tracked user

information should be saved to an external database,

containing the users' profile. The user profile can be

used to improve security, by defining, for each user,

a global trust level for each input modality. The trust

level can be used to advice the user of each modality

to use, at the creation of a new association. Also, it

could be useful to activate confirmation events

whenever a user requests a certain output action

using an input level with a low trust level.

4 IMPLEMENTATION

This section presents the implementation for the

proposed User Profile feature. Firstly, it explains the

approach followed to specify which the test sets are

going to be loaded by the module responsible for

tracking the users’ profile. Secondly, we show the

simple profiling methods that were implemented to

ICAART 2012 - International Conference on Agents and Artificial Intelligence

174

create in future a user classification. Following, it is

explained how the extracted information was used to

adjust certain settings of the interface. Finally, a

demonstration of how the profile is stored to enable

future use is also made.

4.1 Definition of the Sets

To perform the measures described in the previous

section, a simple XML grammar was defined. It

implements four configurable distinct test types:

sequences of gamepad buttons; voice commands;

positions for both joysticks and positions for head.

Example of XML containing user profile test set:

<INTELLWHEELS_PROFILER>

<BINARY_JOYSTICK>

<item>

<sequence>joystick.1

joystick.2 </sequence>

</item>

</BINARY_JOYSTICK>

<ANALOG_JOYSTICK>

(…)

<ANALOG_WIIMOTE>

<item>

</item>

</ANALOG_WIIMOTE>

<item>go forward</item>

<item>turn right</item>

<item>create new sequence</item>

</SPEECH>

</INTELLWHEELS_PROFILER>

The proposed XML grammar makes it possible

for an external operator to configure a test set that

they may find appropriate for a specific context or

user. When a user starts the training session, the four

different types of tests are iterated. In order to attain

a consistent classification of the user, the defined

grammar should be sufficiently extensive. The test

set present on the XML file is iteratively shown to

the user. It starts by asking the user to perform the

gamepad button sequence as can be observed in

Figure 4.

When the user ends the ﬁrst component of the

user proﬁler module, the navigation assistant asks

the user to pronounce the voice commands stored in

the XML. Also, the quantitative results for the

gamepad buttons test are presented.

The last part of the user proﬁler test is shown in

Figure 4: User proﬁler gamepad and voice tests.

Figure 5. The user is invited to place the gamepad’s

joystick into certain positions. A similar approach is

used for the head movements test.

Figure 5: User proﬁler joystick test.

To define the user proficiency in using the

gamepad buttons, a very simple method was

implemented. Each sequence defined on the

grammar should have an associated difficulty level

(easy, medium or hard). The difficulty type of a

sequence may be related to its size, and to the

physical distance between the buttons on the

gamepad. Since the layout of a generic gamepad

may change depending on the model, defining

whether or not a sequence is of easy, medium or

difficulty level is left to the operator.

When the user completes the gamepad sequences

training part, an error rate is calculated for each of

the difficulty levels. If these rates are higher than a

minimum acceptable configurable value, the user

classification in this item is immediately defined.

A METHODOLOGY FOR CREATING INTELLIGENT WHEELCHAIR USERS' PROFILES

175

This classification is then used to turn on the

security feature, which is characterized by a

confirmation event performed by the navigation

assistant. For a grammar with 5 sequences of

difficulty type easy, the maximum number of

accepted errors would be 1. If the user fails more

than one sequence, the confirmation event is

triggered for any input sequence, of any difficulty

type, and the gamepad training session is terminated.

If the error rate for the easy type is less than 20%

(=1/5) the training with the sub-set composed by the

sequences of medium difficulty is initiated. At the

end, a similar method is applied. If the error rate for

the medium level is higher than 30%, the

confirmation is triggered for the medium and hard

levels of difficulty, and the training session is

terminated. Finally, if the user makes it to the last

level of difficulty, the training for the hard

sequences sub-set is started. If the error rate is

higher than 50%, the confirmation event is triggered

only for sequences with a hard difficulty level. The

best scenario takes place when the user is able to

surpass the maximum accepted error rates for all the

difficulty levels. In this situation, the confirmation

event is turned off, and an output request is

immediately triggered for any kind of input

sequence composed only by gamepad buttons.

Defining the ideal maximum acceptable error

rates is not easy. With this in mind, we made it

possible to also configure these values in the XML

grammar.

The joystick phase of the training session can be

used to calculate the maximum amplitude achieved

by the user. This value can then be used to

parameterize the maximum speed value. Imagining a

user who can only push the joystick to 50% of its

maximum amplitude, the speed can be calculated by

multiplying the axis value by two. This feature was

not implemented. However, all the background

preparation to implement it was set for future work.

The speech component of the training session

was used to define the recognition trust level for

each of the voice commands. The trust level is a

percentage value retrieved by the speech recognition

engine. This value is used to set the minimum

recognition level for the recognition module.

Finally, the head movement phase of the training

session has a similar purpose to the joystick's phase.

Additionally, the maximum amplitude for each

direction can be used to determine the range that will

trigger each one of the leaning inputs of the head

gestures recognition.

5 EXPERIMENTS AND RESULTS

The main objective of the experimental work was to

make a preliminary study of the tasks that can be

implemented and the responses of the individuals in

order of get information for the user profiling. The

experiments involved 33 voluntaries, with a mean

age of 24, a standard deviation of 4.2 and without

any movements’ restrictions.

The first experiment consisted in performing the

sequence tasks with several levels of difficulty. In

the first sequence the users needed to push the

gamepad buttons GP1 - GP2 (easy level of

difficulty); the second sequence was GP3 - GP8

(easy level of difficulty); the third sequence was

GP5 - GP8 - GP9 (medium level of difficulty) and

the last sequence was GP6 - GP1 - GP7 - GP4 - GP2

(hard level of difficulty). For the experiments with

voice commands the individuals had to pronounce

the sentences: “Go forward”; “Go back”; “Turn

right”; “Turn left”; “Right spin”; Left Spin” and

“Stop” to get the information about the recognition

trust level for each voice command.

Figure 6: User proﬁler joystick tests.

The last two experiments involved the precision

of the gamepad’s joystick and the head movements.

The voluntaries should move the small dot into the

bigger one with the gamepad’s joystick and with the

wiimote controller. Figure 6 shows some of the tasks

that were asked. The positions were moving right;

up; down; northeast; north-west; southeast and a

sequence northeast - northwest - southeast without

going back to the initial position in the center of the

target.

In general, the achieved results show the good

performance of the individuals using gamepad and

voice commands. The behaviour with head

movements reflects more asymmetry and

heterogeneous results, since several moderate and

severe outliers exist in the time results. The time

ICAART 2012 - International Conference on Agents and Artificial Intelligence

176

consumed to perform the sequences confirmed the

complexity of the tasks as can be seen in Figure 7. In

terms of average time between buttons (Figure 8) it

is interesting to notice the results for the last

sequence. Although it is more complex and longer it

has a positive asymmetry distribution. This probably

reveals that training may improve the user’s

performance.

Figure 7: Time to perform the sequences.

Figure 8: Average time between gamepad buttons.

In terms of errors, the third sequence presents a

higher result with at least one fail. The last sequence

presented a case where 12 errors were committed.

Table 1: Contingency table with the errors of sequences.

Number of Errors

Seq 0 1 2 3 4 5 6 12

1 30 1 2 0 0 0 0 0

2 31 2 0 0 0 0 0 0

3 20 7 3 1 1 0 1 0

4 27 1 1 1 0 2 0 1

Table 2 presents several descriptive statistics,

such as central tendency (mean, median) and

dispersion (standard deviation, minimum and

maximum), for the trust level of speech recognition.

Table 2: Descriptive Statistics for the trust level of speech

recognition.

Sentence Mean Median S. Dev Min Max

“Go Forward”

95.36 95.50

0.51 93.9 95.9

“Go Back” 94.37 95.00 2.44 82.2 95.9

“Turn Right” 95.31 95.40

0.42

94.4 95.9

“Turn Left” 94.76 95.20 1.42 88.4 95.8

“Left Spin” 93.69 94.90 2.88 83.1 95.8

“Right Spin” 94.82 95.00 1.25 89.7 97.2

“Stop”

92.67

94.30

3.85

82.2 95.8

Total Sentences 94.43 94.99 1.08 92.24 95.93

The speech recognition has very good results. In

fact, the minimum of minimums was 82.2 for the

sentences “Go Back” and “Stop”. The expression

“Go Forward” has the highest mean and median.

The sentence “Stop” is more heterogeneous since it

has the higher standard deviation (3.85).

The paired samples t test was applied with a

significance level of 0.05 to compare the means of

time using joystick and head movements. It was

established the null hypothesis: the means of time to

perform the target tasks with joystick and head

movements were equal. The alternative hypothesis

is: the means of time to perform the target tasks with

joystick and head movements were not different.

The achieved power was of 0.80 with an effect size

of 0.5. Table 3 contains the p values of the paired

sample t tests and the 95% confidence interval of the

difference. Observing the results for the positions

Down and Northwest, it is valid to claim there are

statistical evidences to affirm that the mean of time

with joystick and head movements is different. This

reveals the different performance by using in the

same experience the joystick and the head

movements.

Clustering analysis is a technique that can be

used to obtain the information about similar groups.

In the future, this can be used to extract

characteristics for classification and users’ profiling.

The results obtained by hierarchical clustering,

using the nearest neighbour method and squared

Euclidean distance, show the similar performance of

subjects except one individual. In this case, using the

Table 3: Confidence intervals of the difference and p

values.

95% Confidence Interval of

the difference

Move the red dot to:

Lower Upper P

value

Right -2.29 0.67 0.273

Up -1.38 0.08 0.080

Down -9.67 -1.87 0.005

Northeast -2.89 0.66 0.211

Northwest -2.74 -0.17 0.028

Southeast -6.26 1.00 0.150

Northeast - Northwest -

Southeast

-5.32 0.37 0.085

A METHODOLOGY FOR CREATING INTELLIGENT WHEELCHAIR USERS' PROFILES

177

R-square criteria, the number of necessary clusters

to achieve 80% of the total variability retain by the

clusters is 12. Since the sample of volunteers was

from the same population, this kind of conclusions

are very natural. So the next step will consist in

obtain information about handicapped people. In

fact, if the clusters of subjects could be defined then

it should be interesting to work with supervised

classification in which the best command mode

would be the class.

6 CONCLUSIONS AND FUTURE

WORK

Although many Intelligent Wheelchair prototypes

are being developed in several research projects

around the world, the adaptation of user interfaces to

each specific patient is an often neglected research

topic. Typically, the interfaces are very rigid and

adapted to a single user or user group. Intellwheels

project is aiming at developing a new concept of

Intelligent Wheelchair controlled using high-level

commands processed by a multimodal interface.

However, in order to fully control the wheelchair,

users must have a wheelchair interface adapted to

their characteristics. In order to collect the

characteristics of individuals it is important to have

variables that can produce a user profile. The first

stage must be a statistical analysis to extract

knowledge of user and the surrounding. The second

stage must be a supervised classification to use

Machine Learning algorithms in order to construct a

model for automatic classification of new cases.

This paper mainly refers to the proposal of a set of

tasks for extracting the required information for

generating user profiles. A preliminary study has

been done with several voluntaries, enabling to test

the proposed methodology before going to the field

and acquiring information with disabled individuals.

In fact, this will be the next step for future work. The

test set presented in this paper will be tested by a

group of disabled individuals, and the results of both

experiments will be compared to check if the

performances of both populations are similar. Also,

in order to collect feedback regarding the system

usability, disabled users will be invited to drive the

wheelchair in a number of real and simulated

scenarios.

ACKNOWLEDGEMENTS

The authors would like to acknowledge to

FCT – Portuguese Science and Technology

Foundation for the INTELLWHEELS project

funding (RIPD/ADA/109636/2009), for the PhD

Scholarship FCT/SFRH/BD/44541/2008, LIACC –

Laboratório de Inteligência Artificial e de

Computadores, DETI/UA – Dep. Electrónica,

Telecomunicações e Informática and ESTSP/IPP –

Escola Superior de Tecnologia da Saúde Porto –

IPP.

REFERENCES

Adachi, Y., Kuno, Y., Shimada, N., Shirai, N., 1998.

Intelligent wheelchair using visual information on

human faces. In International Conference in Intelligent

Robots and Systems, vol. 1, pp. 354-359.

Bell, D. A., Borenstein, J., Levine, S. P., Koren, Y.; Jaros,

J., 1994. An assistive navigation system for

wheelchairs based upon mobile robot obstacle

avoidance. In IEEE Conf. on Robotics and

Automation, pp. 2018-2022.

Borgerding, B., Ivlev, O., Martens, C., Ruchel, N., Gräser,

A., 1999. FRIEND: Functional robot arm with user

friendly interface for disabled people. In 5th European

Conf. for the Ad-vancement of Assistive Technology.

Braga, R., Petry, M., Moreira, A.P., Reis, L.P., 2009.

Concept and Design of the Intellwheels Platform for

Developing Intelligent Wheelchairs. In LNEE/

Informatics in Control, Automation and Robotics, vol.

37, pp. 191-203.

Cypher, A., Halbert, D. C., 1994. Watch what I do:

programming by demonstration. A. Cypher and D. C.

Halbert, Eds. Massachusetts, USA: Library of

Congress.

Faria, P.M., Braga, R., Valgôde, E., Reis, L.P., 2007.

Interface framework to drive an intelligent wheelchair

using facial expressions. In IEEE International

Symposium on Industrial Electronics, Vigo, pp. 1791-

1796.

Gao, C., Hoffman, I., Miller, T., Panzarella, T., Spletzer,

J., 2008. Performance Characterization of LIDAR

Based Localization for Docking a Smart Wheelchair

System. In Internationsl Conference on Intelligent

Robots and Systems, San Diego.

Hamagami, T., Hirata, H., 2004. Development of

Intelligent Wheelchair acquiring autonomous,

cooperative and collaborative behaviour. In IEEE

International Conference on Systems Man and

Cybernetics, vol. 4, pp. 3525-3530.

Horn, O., Kreutner, M., 2009. Smart wheelchair

perception using odometry, ultrasound sensors and

camera. Robotica, vol. 27, no. 2, pp. 303-310, March.

Hoyer, H., Hölper, R., 1993. Open control architecture for

an intelligent omnidirectional wheelchair. In Proc.1st

TIDE Congress, Brussels, pp. 93-97.

Jia, P., Hu, H., Lu, T., Yuan, K., 2007. Head Gesture

Recognition for Hands-free Control of an Intelligent

ICAART 2012 - International Conference on Agents and Artificial Intelligence

178

Wheelchair. Journal of Industrial Robot, vol. 34, no. 1,

pp. 60-68.

Jia, P., Hu, H., Lu, T., Yuan, K., 2007. Head Gesture

Recognition for Hands-free Control of an Intelligent

Wheelchair. Journal of Industrial Robot, vol. 34, no. 1.

Lakany, H., 2005. Steering a wheelchair by thought. IEE

Digest, vol. 2005, no. 11059, pp. 199-202, The IEE

International Workshop on Intelligent Environments.

Langley, P., 1997. Machine learning for adaptive user

interfaces. In Proceedings of the 21st Ger-man Annual

Conference on Artificial Intelligence, Freiburg, p. 53–

62.

Langley, P., 1999. User modeling in adaptive interface. In

Proceedings of the seventh international conference on

User modeling, Banff, pp. 357-370.

Levine, S. P., Bell, D. A., Jaros, L. A., Simpson, R. C.,

Koren, Y., 1999. The NavChair assistive wheelchair

navigation system. In IEEE Transactions on

Rehabilitation Engineering, vol. 7, pp. 443-451.

Madarasz, R. L., Heiny, L. C., Cromp, R. F., Mazur, N.

M., 1986. The design of an autonomous vehicle for the

disabled. IEEE Journal of Robotics and Automation,

vol. 2, no. 3, pp. 117-126, September.

Maes, P., 1996. Intelligent Software: Programs That Can

Act Independently Will Ease the Burdens that

Computers Put on People. IEEE Expert Systems, vol.

11, no. 6, pp. 62-63, February.

Miller D., Slack, M., 1995. Design and testing of a low-

cost robotic wheelchair. In Autonomous Robots, vol.

2, pp. 77-88.

Miller, D.P., 1998. Assistive Robotics: An Overview. In

Assistive Technology and AI, pp. 126-136.

Ng, P. C., De Silva, L. C., 2001. Head gestures

recognition. In Proceedings International Conference

on Image Processing, pp. 266-269.

Norcio, A. F., Stanley, J., 1989. Adaptive Human-

Computer Interfaces: A Literature Survey and

Perspective. IEEE Transactions on Systems, Man and

Cybernetics, vol. 19, no. 2, pp. 399-408, March.

Pruski, A., Ennaji, M., Morere, Y., 2002. VAHM: A user

adapted intelligent wheelchair. In Proceedings of the

2002 IEEE International Conference on Control

Applications, Glasgow, p. 784789.

Rebsamen B., Rebsamen, B., Burdet, E., Guan, C.,

Haihong Zhang, Chee Leong Teo, Qiang Zeng, Ang,

M., Laugier, C., 2006. A Brain-Controlled Wheelchair

Based on P300 and Path Guidance. In IEEE/RAS-

EMBS International Conference , vol. 20, pp. 1101-

1106.

Reis, L. P., Braga, R., Sousa, M., Moreira, A. P., 2010.

Intellwheels MMI: A Flexible Interface for an

Intelligent Wheelchair. RoboCup 2009: Robot Soccer

World Cup XIII, Springer Ber-lin/Heidelberg, LNCS,

vol. 5949, Graz, pp. 296-307.

Ross, E., 2000. Intelligent User Interfaces: Survey and

Research Directions. University of Bristol, Bristol,

Technical Report: CSTR-00-004.

Satoh Y., Sakaue, K., 2007. An Omnidirectional Stereo

Vision-Based Smart Wheelchair. EURASIP Journal on

Image and Video, p. 11.

Simpson, R., 1998. NavChair: An Assistive Wheelchair

Navigation System with Automatic Adaptation. In

Assistive Technology and Artificial Intelligence.

Berlin: Springer-Verlag Berlin Heidelberg, p. 235.

Simpson, R. C., 2005. Smart wheelchairs: A literature

review. Journal of Rehabilitation Research and

Development, vol. 42 (4), pp. 423–436.

Volosyak, I., Ivlev, O., Graser, A., 2005. Rehabilitation

robot FRIEND II - the general concept and current

implementation. In ICORR 2005 - 9th International

Conference on Rehabilitation Robotics, Chicago, pp.

540-544.

Wellman, P., Krovi, V., Kumar, V., 1994. An adaptive

mobility system for the disabled. In Proc. IEEE Int.

Conf. on Robotics and Automation.

Wooldridge, M., 2002. An Introduction to Multi-Agent

Systems. John Wiley & Sons.

A METHODOLOGY FOR CREATING INTELLIGENT WHEELCHAIR USERS' PROFILES

179