Personalization of Virtual Coaching Applications using Procedural

Modeling

Ren

e Zmugg

1,5

, Andreas Braun

, Peter Roelofsma

, Wolfgang Thaller

, Lisette Moeskops

, Sven

Havemann

, Gabrijela Reljic

and Dieter W. Fellner

1,2,5

Institute of Computer Graphics and Knowledge Visualization, Graz University of Technology, Graz, Austria

Fraunhofer IGD, TU Darmstadt, Darmstadt, Germany

Vrije Universiteit, Amsterdam, Netherlands

University of Luxembourg, Luxembourg, Luxembourg

Fraunhofer Austria, Graz University of Technology, Graz, Austria

Keywords:

Virtual Coaching, Procedural Modeling, Intelligent Environments.

Abstract:

Virtual coaching is an application area that allows individuals to improve existing skills or learn new ones; it

ranges from simple textual tutoring tools to fully immersive 3D learning situations. The latter aim at improv-

ing the learning experience with realistic 3D environments. In highly individual training scenarios it can be

beneﬁcial to provide some level of personalization of the environment. This can be supported using procedural

modeling that allows to easily modify shape, look and contents of an environment. We present the applica-

tion of personalization using procedural modeling in learning applications in the project V2me. This project

combines virtual and social networks to help senior citizens maintain and create meaningful relationships. We

present a system that uses a procedurally generated ambient virtual coaching environment that can be adjusted

by training subjects themselves or in collaboration. A small user experience study has been executed that gives

ﬁrst insight to the acceptance of such an approach.

1 INTRODUCTION

The usage of virtual coaching has greatly increased

over the last few years, for various reasons. In

many cases it augments or even replaces traditional

coaching, in other instances it allows non-typical user

groups to access coaching resources; but most impor-

tantly, the demand of individuals for improving their

skill set has increased (Kozak et al., 1993; Rose et al.,

2000). Virtual coaching applications have a consider-

able diversity, ranging from simple text-based online

tools to immersive 3D environments that have been

conﬁrmed to increase various metrics relevant to the

learning process and thus should be the preferred so-

lution (Ijsselsteijn et al., 2006). However, the cre-

ation of immersive learning environments typically

demands collaboration between designers of the 3D

world that may lack knowledge of the learning pro-

cess and creators of the coaching tools that might not

be able to design 3D environments. Such collabo-

rations increase costs and thus prevent wide-spread

usage of immersive virtual coaching. Another im-

portant factor is acceptance of the system, which can

be increased by providing personalization within the

training environments, e.g. creating the appearance

of their own premises. In this work we provide initial

results on the acceptance of a personalization tool for

virtual coaching system for elderly users.

Procedural modeling uses algorithms to automati-

cally generate 3D models from a limited set of high-

level instructions. While the initial investments in the

tools may be higher, it allows generating further mod-

els easily. In coaching applications this means that a

single procedural modeling tool allows the designers

to easily and rapidly design new immersive learning

environments, without requiring advanced knowledge

in 3D modeling. It is easy to modify existing proce-

dural environments, thus allowing a higher versatility

and adaptation of the model according to the desires

of the user as well as to the current learning situation.

V2me - Virtual Coach reaches out to me (V2me,

2013), is a collaborative European research project

that aims at providing a coaching platform for older

adults that allows them to create and maintain mean-

Zmugg R., Braun A., Roelofsma P., Thaller W., Moeskops L., Havemann S., Reljic G. and Fellner D..

Personalization of Virtual Coaching Applications using Procedural Modeling.

DOI: 10.5220/0005435600370044

In Proceedings of the 1st International Conference on Information and Communication Technologies for Ageing Well and e-Health (ICT4AgeingWell-

2015), pages 37-44

ISBN: 978-989-758-102-1

 2015 SCITEPRESS (Science and Technology Publications, Lda.)

ingful social relations. This improved social connect-

edness helps senior citizens to stay active and health-

ier for longer, as loneliness and lack of social ac-

tivities is a major contributing factor to depression.

Group therapies for teaching friendship skills have

proven efﬁcient in the past (Stevens et al., 2006).

However, it is difﬁcult and costly to provide such

measures for a larger population. An alternative ap-

proach is virtual coaching that is supported by a lim-

ited set of technical systems and can be provided to

a larger user group at decreased cost. This approach

has been evaluated in V2me. It was selected as a case

study to evaluate the advantages of the use of proce-

durally generated environments, which allow person-

alization of the content. These 3D environments are

generated and personalized through a separate proce-

dural scene editor, whose features are illustrated in

this paper. We envision that procedural modeling sys-

tems can help designers and care takers to quickly cre-

ate environments that closely resemble the actual en-

vironment of a system user. Therefore, a combined

system is presented that allows swift creation and

adaption of rooms through procedural modeling. In a

second step those rooms can personalized by adding

further items to the environment, adjusting their po-

sition and changing color and appearance of speciﬁc

aspects.

In this paper we present the prototype tools that

are created for this purpose and how they are applied

in V2me. The system for procedural environment

generation has been evaluated by social science ex-

perts in a mutual workshop with potential users. We

present the initial ﬁndings of this study with a partic-

ular focus on the personalization aspects.

2 RELATED WORK

2.1 Virtual Coaching Applications

Virtual coaches have been in focus of research for

several decades. Heylen et al. gave an overview of

early examples (Heylen et al., 2009). The charac-

ter “Karin” is a virtual receptionist residing behind

the information desk at the lobby of the Virtual Mu-

sic Center (VMC) - a virtual replica of the theatre in

Enschede (Nijholt and Hulstijn, 2000). The dialogue

skills of Karin are based on an earlier non-embodied

dialogue system that interfaces to a database contain-

ing the information on the performances in the actual

theatre. A dialogue system allowed people to query

the information about performances and also to or-

der tickets. Another example, INES, is an intelligent

tutoring system primarily designed to help students

practicing nursing tasks with a haptic device in a vir-

tual environment (Heylen et al., 2005). INES takes

into account elements of the student’s character, his

or her conﬁdence level, and the difﬁculty of the task.

It, moreover, gives appraisal of the student’s actions.

Carcasso proposed a natural human computer in-

teraction paradigm for persons with cognitive impair-

ments such as Alzheimer’s disease (Carrasco et al.,

2008). The paradigm consists of a realistic virtual

character, rendered on a television set, playing the

role of a virtual personal assistant that shows re-

minders, notiﬁcations, and performs short dialogues

with the user. In this paradigm, the television remote

control is used as a return channel to capture the user’s

responses. Roelofsma and Sevim (Roelofsma and Se-

vim, 2012) suggested a framework for virtual coach-

ing in changing lifestyles, including the domain of in-

trapersonal dilemmas that occur when people make

choices that are in the best interest of themselves at

the moment of choice, but not in the best interest

of themselves in the long run. This framework is

based on multiple selves’ theory that proposes a self-

perception of various personal roles, such as a planner

and a doer, which might have conﬂicting intentions

over time. Ortiz et al. performed an empirical study

with older adults and avatars, showing that subjects

follow instructions better, that they can understand the

facial impressions of an avatar even when having cog-

nitive impairments, and that interaction with an emo-

tionally responsive avatar was regarded as pleasant

(Ortiz et al., 2007). However, the presence had no per-

ceptible inﬂuence on recall. They suggest that appli-

cation of an avatar should depend closely on the spe-

ciﬁc task. In A

, 2013), virtual coaches are

connected to several sensors, which include an activ-

ity and a movement sensor, as well as blood pressure

and weight sensors. The Virtual Coach interacts with

the elderly persons in real time in their home environ-

ment and motivates the elderly to a healthy lifestyle.

2.2 Procedural Modeling

A conventional 3D model consists of geometry data

that describes the shape of objects and their pose (po-

sition + orientation) in a 3D scene. By contrast, a

procedural model is a computer program that gener-

ates this geometry data on the ﬂy when it is run. Thus,

it describes the operations required to create an object.

A procedural model can, therefore, describe a whole

family of objects or scenes.

Procedural modeling is used in various kinds of

applications. First and foremost, it has become es-

sential for content creation for computer games and

motion pictures. Here, huge landscapes and urban

ICT4AgeingWell2015-InternationalConferenceonInformationandCommunicationTechnologiesforAgeingWelland

e-Health

structures are often needed, and with the ever-growing

scale of such projects manual construction is becom-

ing increasingly infeasible. Most notably here is

Esri’s City Engine (Esri, 2015), which is used to gen-

erate large cities and buildings through split gram-

mars (Wonka et al., 2003; M

uller et al., 2006; Thaller

et al., 2013b; Zmugg et al., 2014).

As for coaching applications, it is important that

the users remain interested in the system, leading to

adaptation and personalization as important factors.

Modiﬁcations can happen on two levels. First, the

care givers may want to re-use existing environments

and adapt them to provide the users with new content.

And second, users may want to change parts of the

displayed environment to their personal preferences.

A procedural modeling system allows both: Changes

can be made at any point of the generated scene with-

out much effort. Changes of the environment can also

happen fully automatically based on different aspects.

For example, if there is a procedural model of a bed-

room for two people, it can be automatically enlarged

to ﬁt for four people.

Procedural models are usually described in some

sort of scripting language; but this makes the system

not very accessible for non-expert users. Therefore,

an interface without scripting is necessary. The mod-

eling packages Houdini (Side Effects Software, 2015)

and Grasshopper (McNeel and Associates, 2015)

show a graph-based representation of the procedural

model. This graph helps to understand how the model

is built and what dependencies exist without showing

the underlying code. In our system, whose theoretical

foundations were initially presented by Thaller et al.

(Thaller et al., 2013a), neither the code nor the under-

lying dataﬂow graph are shown to the user, but depen-

dencies are illustrated by highlighting related parts in

3D. Furthermore, hierarchically arranged scenes can

be traversed in an intuitive manner. Our system uses a

dataﬂow graph only in the background, which is used

to generate the necessary script code automatically.

This happens completely transparent to the user.

3 V2me AND PERSONALIZED

ENVIRONMENTS

V2me is a research project that provides a coaching

platform for older adults that gives lessons to maintain

or create meaningful relations. It is based on a specif-

ically designed social network that reduces cognitive

load during use. Simpliﬁed communication systems

allow the users to interact with their peers and poten-

tial new friends through a virtual friendship enrich-

ment course (Stevens, 2001). In order to increase

Figure 1: The Virtual Coach (source: (A

, 2013)) in a vir-

tual environment created with the editor, which is displayed

on the Home Platform.

the efﬁciency of these aspects, an immersive Virtual

Coach – a 3D character that provides lessons, expert

information and links to different aspects of the sys-

tem functionality – resides in the apartments of the

users. The virtual environment of this Virtual Coach

should be adaptive to the current situations, both in

regards of user input and expert presets that are stored

in the system or updated from a central web service

regularly. A major requirement for the created tools

was that they have to be used by non-technical per-

sonnel to allow independent creation of lessons and

environments. The system is modular and has three

main components. The Mobile Platform is a portable

tablet computer that is used independently by the par-

ticipant. It provides the Virtual Coach functionality

as text-and-audio representation and enables interac-

tion with other users. The Home Platform is a large-

screen PC. It communicates with the server to syn-

chronize content and host the 3D representation of the

Virtual Coach and environments. The Web Platform

is a web-based engine that allows family and friends

to communicate with network participants.

3.1 The Home Platform

The Home Platform visualizes the content created

from the procedural descriptions. The generated 3D

environments are the “home” of the Virtual Coach.

The Virtual Coach interacts with the user in the pro-

vided environments that can be, for example, a ﬂat

(see Figure 1), or a park. The procedural environ-

ments are generated with a separate piece of software,

the Content Editor, which will be presented in the fol-

lowing Section 3.2.

The Home Platform is responsible for augmenting

all the actions done on the mobile device. The Mo-

bile Platform is able to trigger all animations deﬁned

for the environment in the editor. Facial and body

animations and the text-to-speech engine of the Vir-

tual Coach can also be triggered independently. Upon

PersonalizationofVirtualCoachingApplicationsusingProceduralModeling

completion of a task by the user, the Virtual Coach

reacts with appropriate animations, such as clapping.

The Home Platform provides an immersive inter-

action environment for the users of the system. In-

stead of the small screen on the mobile device, the

larger screen provides a more sophisticated form of

interaction. This enables an engaging learning experi-

ence for the user that should result in increased partic-

ipation, improved learning speed, uptake and recall.

3.2 The Content Editor

We utilize the procedural modeling approach in the

Content Editor, which is used to generate environ-

ments for the Home Platform. This editor is based

on the Generative Modeling Language GML (Have-

mann, 2005) and uses procedural techniques in var-

ious ways to create non-static content that can be

personalized. These include procedural modeling of

backdrops, procedural scene graphs, procedural ob-

ject and camera animations, and personalization of

parameters and objects.

In combination, these techniques can be used to

generate 3D animated stories that can be personal-

ized for individual persons and used to guide, help,

and entertain the target group. This system is adapted

from one Zmugg et al. presented in the Cultural Her-

itage domain (Zmugg et al., 2012a; Zmugg et al.,

2012b). The different steps for the generation of pro-

cedural environments are discussed in the next sec-

tions. For V2me, we have created only a pilot pro-

totype of a procedural scene editor; as in many re-

search projects, much remains to be done to be pro-

duction ready, but the feedback we got conﬁrmed that

the overall approach is understood and appreciated by

our pilot scene production team (see Section 4).

3.2.1 Modeling of Backdrops

Backdrops are the environment where the Virtual

Coach and other objects are positioned in (see the

ﬂat in Figure 1). Backdrops can be generated interac-

tively without any need of scripting. The parametric

system is based on a set of basic modeling operations

and procedural assets. The modeling operations are

inspired by split grammars and feature several opera-

tions to partition existing shapes into smaller pieces.

Modeling usually proceeds in a coarse to ﬁne manner,

meaning that an initial shape is sequentially reﬁned

to achieve the desired result. To remain concise, the

modeling operations are not explained individually in

this paper. For a detailed explanation of the modeling

operations we refer to (Zmugg et al., 2012a).

Procedural assets are ready-made parametric

models. Our system features several libraries which

Figure 2: Selected steps from creating a circular building

using the procedural modeling engine. First and second

row: The the wall is subdivided into parts to deﬁne the space

for the parametric arch. Bottom row: Decisions made at

the beginning (number of sides, radius) can still be changed

later on, which allows change and personalization of proce-

dural models (Zmugg et al., 2012b).

include, for example, windows (like in (Thaller

et al., 2013c)), arches, or other architectural elements.

These models are, however, not static, instead they

provide parameters to approximate a wide range of

different elements. They, furthermore, adapt their size

dependent on the shape they are inserted into. An ex-

ample demonstrating procedural modeling operations

and procedural assets is shown in Figure 2.

3.2.2 Modeling the Scene Hierarchy

After generating and texturing the background, ele-

ments can be positioned in it. We use the concept of

procedural scene graphs, which was proposed in pre-

vious work (Zmugg et al., 2012b). A scene graph is

a hierarchical structure to describe dependencies be-

tween objects in environments. Every object in the

scene is represented by a node in the scene graph.

Each node (except for the root node) has one parent

and an arbitrary number of children. The parent-child

relation propagates all changes made to the parent to

all of its descendants. Technically, every node fea-

tures a 3D pose transformation (4×3 matrix) that af-

fects all its children. As a simple example, imagine a

vase that is placed on top of a table. Without a scene

graph, moving the table does not move the vase, and

vice versa. By using a scene graph, the node repre-

senting the vase can be made a child of the node rep-

resenting the table, and, therefore, moving the table

moves the vase too, while the vase, as the child of the

ICT4AgeingWell2015-InternationalConferenceonInformationandCommunicationTechnologiesforAgeingWelland

e-Health

Figure 3: A camera ﬂight on the Home Platform visual-

ized as a sequence of images. From the view on the Virtual

Coach the camera moves to a close-up view on the tablet

placed on the table.

table, can still be freely moved on the table.

The scene graph can be manipulated by the user

through 3D widgets. This way, e.g. furniture ob-

jects, collected from SketchUp 3D Warehouse (Trim-

ble Navigation Ltd, 2015), can be arranged in the 3D

scene. Technically speaking, the 3D model of the Vir-

tual Coach is part of the scene graph too.

3.2.3 Animating the Scene

Finally, animations can be deﬁned for all the objects

arranged within the scene. To achieve animations,

key frames storing position and orientation can be de-

ﬁned for any object in the scene graph. During ani-

mation, these key frames are interpolated, and can be

assigned to speciﬁc time slots. For these time slots

speciﬁc avatar animations can be set too. As an ex-

ample, the walking animation of the avatar can be ac-

tivated while the scene graph node, which contains

the avatar, moves in the scene. The view on the scene

during these animations is controlled via camera ob-

jects that can also be placed on scene graph nodes. By

animating camera objects, it is easy to produce cam-

era view transitions (see Figure 3). The time slots can

be queued and also arranged in parallel to create si-

multaneous animation time lines. The environments,

together with the deﬁned animation time lines, can be

executed on demand on the Home Platform.

3.2.4 Personalization of the Scene

The curator mode for this system was introduced in

previous work (Zmugg et al., 2012b) in a Cultural

Heritage context. It can be utilized within an Ambient

Assisted Living context in a different, but related way.

The curator mode was designed to provide a museum

curator with a way to make ﬁnal changes to the scene

layout. The scene is deﬁned through a scene template

Figure 4: Different ways for furniture placement in a living

room realized through use of the personalization mode.

beforehand. This template encodes everything the cu-

rator is allowed to change and possibilities are delib-

erately limited so that they are shielded from acciden-

tally doing harm to the scene, and to work in a more

targeted and efﬁcient way.

The situation within the V2me system is similar.

Caregivers can design environments and the animated

stories within them. Afterwards, the elderly people

could use this so-called personalization mode to per-

sonalize the environment. For instance, it is possible

to deﬁne within the template which parameters can

be edited by the elderly users. Take, for example, the

model from Figure 2. It may be allowed to change the

radius of this model freely, but the position and the

number of segments may be ﬁxed or constrained. An-

other feature are the so-called drop targets, which are

positions for additional content. For example, around

a table a set of six drop targets with restricted move-

ment capabilities can be placed. The user can then

decide how many (up to six), and which chairs they

want to place around the table. This way, users can

create entirely different environments based on a sin-

gle template, by re-arranging objects and even placing

new objects in the scene, such as pictures on a wall

(see Figure 4). For the animations, which tell a story

through the procedural environment, to remain the

same, the conﬁguration options of story-related ani-

mations and objects are not available or are restricted.

3.3 Workﬂow of Creating and

Personalizing an Environment

To give a concrete example we will brieﬂy outline the

workﬂow (see Figure 5) of generating a personalized

environment using the tools presented. At ﬁrst, the

basic apartment is created using procedural modeling.

This step can be performed either by a designer or by

a care taker that has been trained for this task. This

model can then be stored as a template and be fur-

ther used to create new apartments even quicker. To

create such models knowledge of the modeling tool

functionality is required.

The next steps are to be performed by a care taker

that has knowledge of the user’s environment. First,

different details are added to the scene. The next step

is adding furniture objects to the scene, such as chairs,

PersonalizationofVirtualCoachingApplicationsusingProceduralModeling

Figure 5: Workﬂow of creating and personalizing an envi-

ronment.

tables, or plants. The 3D model of the Virtual Coach

is also placed in the scene. Afterwards, animations of

the before placed objects are deﬁned in the scene. In

this step, the care taker will deﬁne where the avatar

should reside and perform most of his actions. This is

useful, e.g. to have the avatar appear in favorite places

of the user. This increases the feeling of immersion

for the user. Lastly, the care taker will deﬁne the cam-

era views and transitions to highlight speciﬁc parts of

the scene, or the actions and mimic of the avatar.

The last step is the personalization of the environ-

ment that happens in a co-creation step between user

and care taker in the personalization mode of the sys-

tem. Here, personal items can be placed or wrong

assumptions about preferences can be redeemed. The

environment is visible to both user and care taker and

changes are immediately represented. The ﬁnal result

is an environment that closely approximates the ac-

tual surroundings without requiring the intervention

of seasoned designers in most steps of the process.

4 EVALUATION

The concept for friendship coaching within V2me has

been adapted throughout the runtime of the project

(Muuraiskangas et al., 2012). At the end of the

project, a long term study was conducted with several

users in the Netherlands. In this scope, we wanted

to test various aspects of personalization of the sys-

tem and how procedural modeling can contribute to

this task. Interviews were conducted with all partic-

ipants, covering the aspects on how the quality of a

procedurally modeled environment is regarded, what

elements are required in the environment, and how the

environment can be improved.

4.1 Study Design

Three women and four men, ranging from 64 to 77

years in age, participated in the long term evaluation

of the V2me system. The subjects who participated

were living independently and alone. Subjects were

selected based on their score on the De Jong Gierveld

Loneliness Scale (de Jong Gierveld and van Tilburg,

2006) (scores should be moderate to high) and on

openness to technology (scores should be high). Sub-

jects received an instruction session after the V2me

system was installed in their homes. This system

setup included a predeﬁned apartment environment

and stories featuring the Virtual Coach. The func-

tionalities of the personalization mode itself have not

been directly part of this study. The predeﬁned setup

was used to assess the interest of the users in chang-

ing and personalizing the environment, hence using

the personalization mode in the future.

The loneliness intervention lasted for two months.

Afterwards, they received the loneliness and open-

ness to technology questionnaires again, and a semi-

structured interview was held about their experiences

with the Home Platform. In this interview they were

asked questions like what they thought of the Home

Platform and the interior of the dwelling, and what

they would like to see changed, if they could change

it. Because the participants are usually not used to

terms like procedural modeling, these questions had

to be designed properly to assess their interest in per-

sonalization features.

4.2 Interview Results

Regarding the quality of procedurally modeled envi-

ronments, each participant showed to have their own

preferences for this. One subject mentioned that “Its

very beautiful, very well made. (...) I like the in-

terior a lot.” and “I love how the parquet has been

made”. Answering on the question what elements are

missing from the virtual environment the responses

varied from “the house lacks a lot of stuff” to “it

is too crowded”. Asking how the environment can

be improved, some participants would have preferred

the coach to reside in a more comfortable environ-

ment. They would “add a couch, a little lamp, so he

can sit down and read.” or conclude that “he needs

a stereo set, a TV, and a leather couch.” Or “he

could use a fridge or a TV. Or a stove in his kitchen.”

Most participants would have preferred their coach to

have some company, including a female housemate,

or pets. Finally one participant wanted more interac-

tive elements in the environment, such as “a TV that

shows clips from Youtube or news items.”.

4.3 Discussion

The users were generally happy with the visual qual-

ity of the procedurally generated environment and the

existing objects added to it. Therefore, we can as-

sume that similar environments can be equally well

ICT4AgeingWell2015-InternationalConferenceonInformationandCommunicationTechnologiesforAgeingWelland

e-Health

received. Preference plays a huge part in the ques-

tion how comprehensive the provided environment

was perceived. Some users found it to be limited, and

would have preferred additional personal items for the

coach. Providing more objects was the most com-

mon suggestion for improvement in the future. Here

it is important to create an interface that enables to

access this variety in a simple fashion. Additionally,

some more interaction for the coach should be added,

such as companions in the environment. This is more

difﬁcult to accomplish, however the required anima-

tions and models can also be procedurally gener-

ated. The ﬁnal aspect is interactivity, whereas objects

should link to additional services, such as Youtube or

Wikipedia. Again, this is interesting to pursue in a

future work by creating a library of web services that

can be linked to objects in the environment.

All in all, the results indicate that every user has

their own speciﬁc preferences for the design of the en-

vironments of the Home Platform, so personalization

is a desirable feature for virtual coaching systems.

5 CONCLUSION

We have presented a system which is capable of gen-

erating personalizeable environments for elderly peo-

ple within coaching applications that rely on 3D con-

tent. Procedural modeling techniques are utilized to

generate environments, animations and camera ﬂights

within an editor, which is designed to be accessible

for non-expert users; so programming is not neces-

sary to use it. This editor is used within the V2me

system to generate environments in which the Vir-

tual Coach is presented and interacts with the user.

All deﬁned animations and camera ﬂights can be trig-

gered by actions of the user. This in combination with

lessons that are designed by professional psycholo-

gists is used to tell stories and give tasks to motivate

people to get involved in social activities again. Af-

ter all, the ultimate goal of the project is to stimulate

richer real-life social experiences.

All elements positioned in the environments cre-

ated by our editor can be repositioned and extended

at any time. In a separate personalization mode, the

conﬁguration options can be limited, so elderly users

gain more control and are shielded from doing harm

by not accessing certain parameters. The results from

the user study explained in Section 4 show that there

is a big interest in personalization of environments.

The answers of the participants indicate that a sin-

gle environment is not sufﬁcient because the priori-

ties and preferences of the participants differ vastly.

This shows that procedural modeling needs to be fa-

cilitated so that the wishes of elderly people for virtual

coaching systems can be fulﬁlled with ease.

First results have shown that our approach is a

promising new ICT-based method for bringing care

to the homes of the elderly people in a socially mean-

ingful way. Especially the ﬂexibility and use of the

generic aspects of the content editor proved to be

promising for care givers and professionals.

5.1 Limitations and Future Work

The system has not been adapted for usage by the el-

derly end users of V2me yet. For the current train-

ing applications it was more suitable to have a sys-

tem that provides developers of training systems with

tools that allow them to create immersive environ-

ments. For this a rich set of functions have been de-

veloped in collaboration with social scientists. The

environments for the user study were created by this

team. However, with a limited amount of training

they should be able to create their own environments.

The system is prepared for the environment person-

alization. Modiﬁcations should be accessible for all

users in an intuitive way. The focus of the prototype

up until now was on functionality rather than usabil-

ity. Increasing the usability of the system and make it

accessible for elderly user groups is the most impor-

tant step in our future work. We anticipate stimulating

psychological effects if the user can inﬂuence the sur-

roundings of the Virtual Coach with no assistance.

Additional improvements and future work for the

prototype include a more realistic rendering and a

wider range of possible objects. We currently work

on functions to design also the avatar animations pro-

cedurally, so that the animations of the Virtual Coach

can be dynamically adapted, e.g., controlling the level

of smiling, or specifying animations that adapt to dif-

ferent chair and table heights. This is indispensable

for more interesting plots and stories.

ACKNOWLEDGEMENTS

This work was supported by the AAL Joint Pro-

gramme project V2me (grant no. AAL-2009-2-107).

We want to thank all participants of the user studies.

REFERENCES

(2013). Available: http://www.a2e2.eu (Accessed:

17-Feb-2015).

Carrasco, E., Epelde, G., and Moreno, A. (2008). Nat-

ural interaction between avatars and persons with

PersonalizationofVirtualCoachingApplicationsusingProceduralModeling

alzheimer’s disease. In Computers helping people

with special needs, pages 38–45.

de Jong Gierveld, J. and van Tilburg, T. (2006). A six-item

scale for overall, emotional and social loneliness: con-

ﬁrmatory tests on survey data. Research on aging: A

Bimonthly on Aging and the Life Course, 28:582–598.

Esri (2015). CityEngine. Available:

www.esri.com/software/cityengine (Accessed:

17-Feb-2015).

Havemann, S. (2005). Generative Mesh Modeling. PhD

thesis, Institute of Computer Graphics, Faculty of

Computer Science, Braunschweig Technical Univer-

sity, Germany.

Heylen, D., Nijholt, A., and op den Akker, R. (2005). Affect

in tutoring dialogues. Applied Artiﬁcial Intelligence,

19(3-4):287–311.

Heylen, D., Theune, M., and Nijholt, A. (2009). Social

Agents: the ﬁrst generations. In ACII.

Ijsselsteijn, W. A., de Kort, Y. A. W., Westerink, J., de Jager,

M., and Bonants, R. (2006). Virtual ﬁtness: stim-

ulating exercise behavior through media technology.

Presence: Teleoperators and Virtual Environments,

15(6):688–698.

Kozak, J., Hancock, P., Arthur, E., and Chrysler, S. (1993).

Transfer of training from virtual reality. Ergonomics.

McNeel, R. and Associates (2015). Grasshopper 3D. Avail-

able: http://www.grasshopper3d.com (Accessed: 17-

Feb-2015).

uller, P., Wonka, P., Haegler, S., Ulmer, A., and Van Gool,

L. (2006). Procedural modeling of buildings. ACM

Transactions on Graphics, 25(3):614–623.

Muuraiskangas, S., Leist, A., Braun, A., Klau, K., Roelof-

sma, P. H. M. P., Wichert, R., Klein, P., and Ferring,

D. (2012). V2me: Evaluating the ﬁrst steps in mobile

friendship coaching. JAISE, 4(6):517–534.

Nijholt, A. and Hulstijn, J. (2000). Multimodal interactions

with agents in virtual worlds. Future Directions for

Intelligent Information Systems and Information Sci-

ence, 45:148–173.

Ortiz, A., del Puy Carretero, M., Oyarzun, D., Yanguas,

J. J., Buiza, C., Gonzalez, M. F., and Etxeberria, I.

(2007). Elderly users in ambient intelligence: Does an

avatar improve the interaction? In Universal Access

in AmI Environments, pages 99–114. Springer.

Roelofsma, P. H. M. P. and Sevim, K. (2012). Using multi-

ple self theory of planner and doer as a virtual coach-

ing framework for changing lifestyles: The role of

expert, motivator and mentor coaches. In Sixth Int.

Workshop on HAI.

Rose, F., Attree, E., and Brooks, B. (2000). Training in

virtual environments: transfer to real world tasks and

equivalence to real task training. Ergonomics.

Side Effects Software (2015). Houdini. Available:

http://www.sidefx.com (Accessed: 17-Feb-2015).

Stevens, N. (2001). Combating loneliness: a friendship en-

richment programme for older women. Ageing and

Society, 21:183–202.

Stevens, N. L., Martina, C. M., and Westerhof, G. J. (2006).

Meeting the need to belong: Predicting effects of

a friendship enrichment program for older women.

Gerontologist, 46(4):495–502.

Thaller, W., Krispel, U., Zmugg, R., Havemann, S., and

Fellner, D. W. (2013a). A Graph-Based Language for

Direct Manipulation of Procedural Models. Int. Jour-

nal On Advances in Software, 6(3 & 4):236–255.

Thaller, W., Krispel, U., Zmugg, R., Havemann, S., and

Fellner, D. W. (2013b). Shape grammars on convex

polyhedra. Computers & Graphics, 37(6):707–717.

Thaller, W., Zmugg, R., Krispel, U., Posch, M., Havemann,

S., and Fellner, D. (2013c). Creating Procedural Win-

dow Building Blocks using the Generative Fact La-

beling Method. In Proc. ISPRS Workshop 3D-ARCH

2013.

Trimble Navigation Ltd (2015). 3d warehouse. Avail-

able: https://3dwarehouse.sketchup.com/ (Accessed:

17-Feb-2015).

V2me (2013). Available: http://www.v2me.org (Accessed:

17-Feb-2015).

Wonka, P., Wimmer, M., Sillion, F., and Ribarsky, W.

(2003). Instant architecture. ACM Transactions on

Graphics, 22(3):669–677.

Zmugg, R., Krispel, U., Thaller, W., Havemann, S., Pszeida,

M., and Fellner, D. W. (2012a). A New Approach for

Interactive Procedural Modelling in Cultural Heritage.

In Proc. Comp. Applications & Quantitative Methods

in Archaeology 2012.

Zmugg, R., Thaller, W., Hecher, M., Schiffer, T., Have-

mann, S., and Fellner, D. W. (2012b). Authoring An-

imated Interactive 3D Museum Exhibits using a Dig-

ital Repository. In VAST, pages 73–80. Eurographics

Association.

Zmugg, R., Thaller, W., Krispel, U., Edelsbrunner, J., Have-

mann, S., and Fellner, D. W. (2014). Procedural archi-

tecture using deformation-aware split grammars. The

Visual Computer, 30(9):1009–1019.

ICT4AgeingWell2015-InternationalConferenceonInformationandCommunicationTechnologiesforAgeingWelland

e-Health