IAAN: Intelligent Animated Agent with Natural Behaviour for Online

Tutoring Platforms

Helen V. Diez, Sara Garc

ıa , Jairo R. S

anchez, Maria del Puy Carretero and David Oyarzun

Vicomtech-IK4 Research Center, San Sebasti

an, Spain

Keywords:

Agents, Artiﬁcial Intelligence, Natural behaviour, Behaviour Markup Language.

Abstract:

The goal of the work presented in this paper is to develop an Intelligent Animated Agent with Natural Be-

haviour (IAAN). This agent is integrated into e-learning platforms in order to perform the role of an online

tutor. The system stores into a database personalized information of each student regarding their level of edu-

cation, their learning progress and their interaction with the platform. This information is then used by the 3D

modeled virtual agent to give personalized feedback to each student; the purpose of the agent is to guide the

students throughout the lectures taking into account their personal needs and interacting with them by means

of verbal and non-verbal communication. To achieve this work a thorough study of natural behaviour has been

held and a complex state machine is being developed in order to provide IAAN with the sufﬁcient artiﬁcial

intelligence as to enhance the students motivation and engagement with the learning process.

1 INTRODUCTION

Online tutoring has become very popular in the past

years and according to several studies such as the lat-

est survey by Ambient Insight Research (Adkins,

2013); the aggregate growth rate for self-paced e-

learning products and services expected for the next

ﬁve year period (2011-2016) is 7.6%. Distance learn-

ing offers a series of beneﬁts that traditional learning

cannot compete with, for example in terms of mobil-

ity, affordability or ﬂexibility e-learning happens to

be much more suitable for nowadays lifestyle. How-

ever, online learning also has its drawbacks, the lack

of supervision from a tutor in the courses may lead

to demotivation, boredom and the ﬁnal drop of the

courses. This is why the integration of virtual agents

represented as human characters into these platforms

can be an effective solution to make the students feel

supported and accompanied throughout the course as

a real teacher would.

Michael Graham Moore (Moore, 1989) classiﬁed

the possible interactions in distance education into

three types:

• learner-content

• learner-instructor

• learner-learner

He acknowledged the “learner-instructor” inter-

action as the most important. Moreover, studies

like (Bloom, 1984) demonstrated the effectiveness

of a one-on-one human tutoring system against other

methods of teaching, and (Lepper et al., 1993), de-

fended the idea that education could be globally im-

proved if every student was provided with a personal

tutor. This is something almost impossible to achieve

in traditional education but it is not so in online learn-

ing.

This work introduces IAAN, a virtual agent rep-

resented as a 3D modeled character endowed with in-

telligence and natural behaviour, designed to perform

the role of a real tutor in e-learning platforms. IAAN

reacts in real time to the students’ interaction with the

platform by means of verbal and non-verbal commu-

nication. Furthermore, in order to make IAAN as re-

alistic as possible an Intelligent Animated Agent Ed-

itor is included into the system allowing real tutors to

deﬁne the appearance and behaviour of IAAN in re-

sponse to different situations. This editor is based on

the Behaviour Markup Language standard (BML

Additionally, this work is entirely web based so

it solves the interoperability issues presented by most

commonly used e-learning systems with other plat-

forms such as Learning Management Systems, web-

based virtual world platforms, Virtual Reality learn-

ing systems or simulators.

IAAN has been partially integrated into Moodle

http://www.mindmakers.org/projects/bml-1-0/wiki

123

V. Diez H., García S., R. Sánchez J., del Puy Carretero M. and Oyarzun D..

IAAN: Intelligent Animated Agent with Natural Behaviour for Online Tutoring Platforms.

DOI: 10.5220/0004756401230130

In Proceedings of the 6th International Conference on Agents and Artiﬁcial Intelligence (ICAART-2014), pages 123-130

ISBN: 978-989-758-016-1

 2014 SCITEPRESS (Science and Technology Publications, Lda.)

(Moodle, 2013) Learning Management System.

This paper is organized as follows; Section 2 an-

alyzes the related work carried out in the last years

concerning agents and artiﬁcial intelligence, Section

3 describes the architecture followed to accomplish

the goals of this work, Section 4 shows integration

results from the work. The ﬁnal section is about con-

clusions and future work.

2 RELATED WORK

The purpose of this work is to integrate an intelligent

3D virtual agent into e-learning platforms. Concern-

ing this process, Buraga (Buraga, 2003) proposes an

agent-oriented extensible framework based on XML

family for building a hypermedia e-learning system

available on the world-wide-web. This intelligent tu-

toring system is composed of four major components,

the information processed by each component can be

stored by XML documents. Some of the components

are implemented as intelligent agents.

Angehrn et al. (Angehrn et al., 2001) sug-

gests the use of K-InCA to provide a personalized e-

learning system to help people learn and adopt new

behaviours. The agent continuously analyses the ac-

tions of the user in order to build and maintain a “be-

havioural proﬁle” reﬂecting the level of adoption of

the “desired” behaviours. Using this proﬁle, the agent

provides customized guidance, mentoring, motivation

and stimuli, supporting the gradual transformation of

the users behaviours.

Deﬁning natural behaviour is not an easy task ei-

ther. The literature and theory of affective comput-

ing imply several conditions for synthesized motion

to appear natural (Abrilian et al., 2005). Speed of

interaction and emotion/speech-correlated believable

body motion are among the most important function-

alities (Mlakar and Rojc, 2011). Rieger (Rieger

et al., 2003) developed a series of rules in order to

increase the acceptance of virtual agents in Human-

Computer Communication and established a correla-

tion table between the message to rely and the emo-

tion to show.

Steve (Johnson and Rickel, 1997) was one of the

ﬁrst pedagogical agents capable of expressing emo-

tions; it was designed as a stereoscopic 3D charac-

ter that cohabited with learners, it has been applied

to naval training tasks. However, Steve was origi-

nally designed to operate in immersive virtual envi-

ronments and not over the Web.

Project GRETA (Poggi et al., 2005) presents

a multimodal Embodied Conversational Agent (ECA)

capable of interpreting APML

mark-up lan-

guage to generate synchronized speech, face, gaze

and gesture animations.

More recently, (Benin et al., 2012) presented a

three-dimensional animated talking head which re-

peats any input text in six different emotional ways.

3 INTELLIGENT ANIMATED

AGENT ROLES

The ﬁnal goal of the work presented is to integrate a

3D animated agent into e-learning platforms in order

to asses and guide students as a real teacher would.

Many studies have been held to identify the qual-

ities of a good teacher (Azer, 2005) (Korthagen,

2004). Besides, students and teachers do not always

agree in the importance of these qualities. From the

perspective of students, Brown and McIntyre (Brown

and McIntyre, 1993) and Batten (Batten et al., 1993)

found the two qualities with highest frequency of

mention were the teachers ability to “explain clearly”,

and “help us with our work”. On the other hand, two

qualities seen by teachers as crucial, but not men-

tioned by students, were “planning, structuring and

organising the classroom, and fostering student in-

volvement and participation”.

Considering previous research, IAAN has been

designed to perform several roles throughout the

course depending on the needs of the lesson, these

roles have been divided into three states:

• Explanation State. IAAN will be able to explain

new concepts to the user. Humans tend to gesticu-

late when introducing an idea to others; perform-

ing arm movements or pointing out objects, IAAN

will act alike.

• Evaluation State. One of the most important

roles that IAAN must perform is the one of a

real tutor. IAAN will guide the students through

the lectures and will interact with them giving

them personal feedback, responding in real-time

to their interaction with the platform.

• Waiting State. When IAAN is not in any

of the previous states, for example, when self-

explanatory audiovisual content is being dis-

played to the student, IAAN will enter a waiting

state mode. IAAN will not interact directly with

the student but he will be animated. Humans do

not stand hieratical when waiting for something

to happen; we balance our body, gaze, cross our

arms, etc. As IAAN is endowed with natural be-

http://apml.areyoupayingattention.com

ICAART2014-InternationalConferenceonAgentsandArtificialIntelligence

124

haviour, he will also perform these kind of move-

ments in this state.

The following section introduces the Intelligent

Animated Agent Editor, this tool is being developed

to ease the job of deﬁning the virtual agent’s appear-

ance and behaviour. With the help of this tool the

course editor will be able to deﬁne the Explanation

and Waiting state. This tool is the core of the Be-

haviour Module. The Evaluation state will be dis-

cussed in more detail in Section 5.

3.1 Intelligent Animated Agent Editor

This tool is being implemented to allow real tutors

to design IAAN’s appearance and behaviour as they

deem most appropriate in each case. To ensure quality

education it is important to leverage the knowledge

and expertise provided by real teachers.

Regarding IAAN’s appearance, McCloud (Mc-

Cloud, 1994) stated that individuals see themselves as

iconic images but see others in a more detailed form,

that is, as realistic images. Gulz and Haake (Gulz

and Haake, 2006) extended this idea to the role of an-

imated pedagogical agents and stated that if the agent

is acting as a teacher, the student will see it as “the

other person” and therefore it is better to represent it

in a human form. However, the risk of falling into

the “uncanny valley” (Mori, 1970) also exists and

furthermore depending on the target of the course the

editor may consider to represent IAAN in a more car-

toonish shape, for example, if the course is targeted

for kids. So the decision of IAAN’s appearance is en-

tirely left in the hands and teaching experience of the

real tutor. The agent’s appearance is selected from

a list of predeﬁned 3D models and imported into the

IAA Editor.

With respect to facial expression, the IAA editor

is based on Ekman’s six universal emotions (Ekman

and Friesen, 1981), so IAAN will be able to show

the following facial expressions: happiness, surprise,

anger, sadness, disgust and fear. An example of IAAN

performing these facial expressions can be found in

Figure 1.

IAAN performs several hand gestures, the scien-

tiﬁc community has established four type of hand ges-

tures (Cassell et al., 1994):

• Iconic: or illustrators, they are descriptive ges-

tures often used to illustrate speech.

• Metaphoric: or representational gestures, rep-

resent an abstract feature concurrently spoken

about.

• Deictic: indicate a point in the space.

Figure 1: IAAN represented as a 3D virtual agent express-

ing facial emotions (happiness, surprise, anger, sadness).

• Beats: small formless waves of the hand that oc-

cur to emphasize words.

IAAN communicates in a verbal way with the stu-

dents, for this purpose the IAA Editor includes a text

editor. This text is then transformed into speech by a

Text-to-Speech synthesizer.

As shown in Figure 2 the IAA Editor is composed

of three main areas; the animations regarding natu-

ral behaviour can be found in the Options Area, the

real tutor selects the desired animation from one of the

available menus and drags it onto the Timeline Area,

this action is repeated as many times as necessary un-

til the desired behaviour is achieved. Then the ﬁnal

result is visualized in the Viewer Area.

The composition created in the timeline is trans-

lated into a BML ﬁle. In the following BML example

created with the IAA Editor, a welcoming message

has been designed. This BML ﬁle is then used as an

input to the Animation Engine.

<bml xmlns=

"http://www.bml-initiative.org/bml/bml-1.0"

character="Iaan"

id="bml1">

<gesture id="behavior1" lexeme="hello-waving"

start="2" end="3"/>

<faceLexeme id="behavior2" lexeme="happy"

amount="0.8" start="2" end="3"/>

<text>Wellcome to the first lesson!</text>

</speech>

</bml>

4 IMPLEMENTATION

The architecture chosen to accomplish the goals pre-

sented in the previous section is shown in Figure 3.

IAAN:IntelligentAnimatedAgentwithNaturalBehaviourforOnlineTutoringPlatforms

125

Figure 2: Intelligent Animated Agent Editor.

Figure 3: Platform Modules.

The main modules involved in the design of IAAN are

the Behaviour Module, the Evaluation Module and

the Animated Engine.

The Behaviour Module has been described in Sec-

tion 3. This module is in charge of deﬁning the be-

haviour of IAAN in each situation, to make IAAN as

realistic as possible a real tutor is in charge of describ-

ing IAAN’s interaction with the students. To ease this

job the IAA Editor has been developed.

Next the Evaluation Module and the Animated

Engine will be explained in more detail.

4.1 Evaluation Module

IAAN must accompany the students throughout the

course and interact with them whenever is necessary.

To accomplish this IAAN will keep track of each stu-

dent along the course in order to give personalized

feedback to each individual.

This is achieved by storing multiple information

into a database and by deﬁning a complex state ma-

chine that will inform IAAN when and how to interact

with each student.

4.1.1 User Proﬁle Database

It is crucial to know as much as possible about each

student in order to assess them according to their per-

sonal needs. Table 1 shows the information stored

into the database for further analyses, regarding per-

sonal information as well as the student’s interaction

with the e-learning platform.

Table 1: User Proﬁle Information.

Personal Information Platform Usage

Name Logins

Age Session Duration

Address Interaction Speed

Knowledge Level Mistake Frequency

Taking this information into account the course

editor is able to design speciﬁc evaluation rules for

each student.

The ﬁrst thing the course editor must decide is

the EVALUATION HARSHNESS, initially this parameter

ICAART2014-InternationalConferenceonAgentsandArtificialIntelligence

126

is set depending on the student’s age and knowledge

level, but as the course progresses and the student ac-

quires more knowledge this scale can be modiﬁed.

The Platform Usage information is designed to es-

tablish a BEHAVIOUR PATTERN for each student, for

example; the amount of logins per week, the ses-

sion duration, the frequency in answering questions,

the mistaken answers. Bearing this pattern in mind,

IAAN will interact with the student whenever a disor-

der in the pattern takes place.

4.1.2 Evaluation Manager

The Evaluation Manager is designed as a ﬁ-

nite state machine (FSM). As seen in Figure 4

the Evaluation Manager takes as inputs the

EVALUATION HARSHNESS, the BEHAVIOUR PATTERN

and the BML FILES that deﬁne IAAN’s natural

behaviour which have been previously designed with

the IAA editor .

Taking these inputs into account the Evaluation

Manager waits for behaviour pattern alerts and if they

occur the manager orders IAAN to interact with the

student by means of the previously designed BML

ﬁles. Regarding the User Proﬁle Information the pos-

sible alerts have been deﬁned as follows:

• LOGIN ALERT. The course editor is in charge of

establishing an amount of logins per week (or

month) for each student in order to fulﬁl his as-

signments. If this recommendation is altered in

any way the platform is notiﬁed with an alert mes-

sage and IAAN is launched to interact with the

student.

• INTERACTION ALERT: Once the student has

logged in he interacts with the platform in a deter-

mined frequency. Modifying this frequency may

mean several things, for example, if the interac-

tion speed increases it may indicate the student

ﬁnds the lesson too easy, on the contrary if the

speed decreases the lesson might be too difﬁcult

or it may simply mean the student is taking a

break. IAAN is able to interact with the student

to ﬁnd out what is happening.

• MISTAKE ALERT. If the student commits more

mistakes than usual the course level might not be

appropriate or the student might not be paying at-

tention. IAAN might enter the Explanation state

in order to clarify concepts or draw the student’s

attention by introducing a multimedia effect.

• SESSION ALERT. A recommended minimum and

maximum time is set to perform each session, if

this time is altered IAAN shows up to check if the

student has ﬁnished his assignment.

The messages in round boxes from Figure 4 rep-

resent examples of IAAN communicating with the

student in each alert situation. For example, if the

Evaluation Manager detects an INTERACTION ALERT

IAAN launches the BML ﬁle describing the be-

haviour to adopt under this circumstance, for in-

stance, IAAN will ask the student: ”Are you taking

a break?”. Another example would be in the event

of a MISTAKE ALERT, in this case IAAN will express

concern by asking: ”Is the lesson too difﬁcult?”.

With these queries IAAN seeks to encourage the

student to continue working, letting him know he is

not alone and that he is being supervised.

Figure 4: Evaluation FSM Example.

4.2 Animation Engine

The Animation Engine is composed by several mod-

ules developed using JavaScript programming lan-

guage and following all the HTML5 and Web3D stan-

dards. These modules have been developed as an ab-

straction layer over O3D

engine which has been se-

lected amongst other engines (GLGE, x3dom, etc.)

for its beneﬁts, as it is not a very high level API it al-

lows great ﬂexibility when developing new features.

WebGL (Leung and Salga, 2010) technology

has been used to render IAAN via the web. We-

bGL is based on OpenGL, which is a widely used

open source 3D graphics standard. Nowadays, most

common browsers support this technology; Google

Chrome, Mozilla Firefox, Apple Safari or Opera.

The modules that compose the Animation Engine

are in charge, amongst other features, of rendering 3D

modeled characters into web browsers, parsing BML

ﬁles which deﬁne the character’s behaviour and ani-

http://code.google.com/p/o3d/

IAAN:IntelligentAnimatedAgentwithNaturalBehaviourforOnlineTutoringPlatforms

127

Figure 5: Integration of IAAN into Moodle.

mations and communicating with the e-learning plat-

form.

5 INTEGRATION

The work presented in this paper is entirely web

based, this fact makes the platform compatible

with most widely-used learning management systems

(LMS)

; Edmodo (Edmodo, 2013), Moodle (Moo-

dle, 2013), Blackboard (Blackboard, 2013), SumTo-

tal Systems (SumTotal, 2013), etc.

In the work presented IAAN has been partially in-

tegrated into Moodle. Moodle is based on a Model-

View-Controller coding design pattern. To integrate

IAAN into Moodle the modules from this work must

be added to the conﬁguration ﬁle of the platform.

This platform offers a very interesting feature for

our work; the Conﬁgurable Reports. This block is a

Moodle custom reports builder designed in a modular

way to allow developers to create new plugins. The

types of reports available are:

• Courses reports, with information regarding

courses.

• Categories reports, with information regarding

categories.

http://edudemic.com/wp-content/uploads/2012/10/top-

20-lms-software-solutions.png

• User reports, with information regarding users

and their activity in a course.

• Timeline reports, this is a special type of report

that displays a timeline showing data depending

on the start and end time of the current row.

• Custom SQL Reports, custom SQL queries.

Taking advantage of this feature a Synchroniza-

tion Module is being developed as a communication

bridge between Moodle and the Animation Engine

(Figure 3).

For this work a very simple English Course has

been created in Moodle with multiple choice quizzes

for the student to answer. Figure 5 shows the integra-

tion of IAAN into the created course.

Validation results regarding the entire platform in-

tegration have not yet been performed as it is still

work in progress. However, some of our ﬁnal users in

different applications have been able to interact with

IAAN and they have pointed out its natural behaviour

and communication as an engaging and realistic way

of interaction.

6 CONCLUSIONS AND FUTURE

WORK

An intelligent virtual agent represented as a 3D

modeled character has been presented in this paper.

Thanks to the IAA editor the agent is gifted with nat-

ICAART2014-InternationalConferenceonAgentsandArtificialIntelligence

128

ural behaviour, allowing real tutors to use their expe-

rience to deﬁne the agent’s reactions in different cir-

cumstances.

The Evaluation Module described in this work is

still being developed. The main goal of this module is

to turn the animated agent into an autonomous agent

that needs no exterior intervention to respond to stu-

dents’ interaction in real-time.

The modules of this work have been developed

following a Model-View-Controller coding pattern to

ease the integration with the selected e-learning plat-

form. IAAN has been successfully integrated into

Moodle, though only partially functional as the Eval-

uation Module is still under development.

The Animation Engine introduced in this work is

constantly improved to suit new needs. The ﬁnal goal

is to develop an animation engine capable of repro-

ducing human behaviour as realistic as possible.

A synchronization module is being developed in

order to optimize the communication between IAAN

and Moodle. Furthermore, we are studying other

LMS in order to develop a general synchronization

module turning IAAN into a multiplatform assistant.

Finally, we are deﬁning the validation phase to test

the work presented in this paper with real students in

order to conﬁrm IAAN’s positive effect. The results

of these evaluations will verify whether IAAN’s natu-

ral behaviour and real-time emotional response has a

beneﬁcial effect on the students learning engagement

and ﬁnal cognitive results.

REFERENCES

Abrilian, S., Devillers, L., Buisine, S., and Martin, J.-C.

(2005). Emotv1: Annotation of real-life emotions for

the speciﬁcation of multimodal affective interfaces. In

HCI International.

Adkins (2013). The worldwide market for self-paced

elearning products and services: 2011-2016 forecast

and analysis.

Angehrn, A. A., Nabeth, T., Razmerita, L., and Roda, C.

(2001). K-inca: Using artiﬁcial agents to help peo-

ple learn and adopt new behaviours. In ICALT, pages

225–226.

Azer, S. A. (2005). The qualities of a good teacher: how

can they be acquired and sustained? Journal of the

Royal Society of Medicine, 98(2):67–69.

Batten, M., Marland, P., Khamis, M., and for Educa-

tional Research., A. C. (1993). Knowing how to teach

well : teachers reﬂect on their classroom practice /

Margaret Batten, Perc Marland, Mon Khamis. ACER

Hawthorn, Vic.

Benin, A., Leone, G. R., and Cosi, P. (2012). A 3d talking

head for mobile devices based on unofﬁcial ios we-

bgl support. In Proceedings of the 17th International

Conference on 3D Web Technology, pages 117–120.

ACM.

Blackboard (2013). Wherever your learning takes place.

http://uki.blackboard.com. Accessed November,

2013.

Bloom, B. S. (1984). The 2 sigma problem: The search

for methods of group instruction as effective as one-

to-one tutoring. Educational Researcher, 13(6):4–16.

Brown, S. and McIntyre, D. (1993). Making Sense of Teach-

ing. Open University Press.

Buraga, S.-C. (2003). Developing agent-oriented e-learning

systems. In in Proceedings of The 14th International

Conference on Control Systems And Computer Sci-

ence vol. II, I. Dumitrache and C. Buiu, Eds, Po-

litehnica. Press.

Cassell, J., Pelachaud, C., Badler, N., Steedman, M.,

Achorn, B., Becket, T., Douville, B., Prevost, S., and

Stone, M. (1994). Animated conversation: rule-based

generation of facial expression, gesture & spoken in-

tonation for multiple conversational agents. In Pro-

ceedings of the 21st annual conference on Computer

graphics and interactive techniques, pages 413–420.

ACM.

Edmodo (2013). Where learning happens.

https://www.edmodo.com. Accessed November,

2013.

Ekman, P. and Friesen, W. V. (1981). The repertoire of non-

verbal behavior: Categories, origins, usage, and cod-

ing. Nonverbal communication, interaction, and ges-

ture, pages 57–106.

Gulz, A. and Haake, M. (2006). Design of animated peda-

gogical agentsa look at their look. International Jour-

nal of Human-Computer Studies, 64(4):322–339.

Johnson, W. L. and Rickel, J. (1997). Steve: An animated

pedagogical agent for procedural training in virtual

environments. ACM SIGART Bulletin, 8(1-4):16–21.

Korthagen, F. A. (2004). In search of the essence of a good

teacher: towards a more holistic approach in teacher

education. Teaching and Teacher Education, 20(1):77

– 97.

Lepper, M. R., Woolverton, M., Mumme, D. L., and Gurt-

ner, J. L. (1993). Motivational techniques of expert

human tutors: Lessons for the design of computer-

based tutors. In Lajoie, S. P. and Derry, S. J., ed-

itors, Computers as Cognitive Tools, pages 75–105.

Lawrence Erlbaum Associates.

Leung, C. and Salga, A. (2010). Enabling webgl. In

Proceedings of the 19th international conference on

World wide web, pages 1369–1370. ACM.

McCloud, S. (1994). Understanding comics. William Mor-

row Paperbacks.

Mlakar, I. and Rojc, M. (2011). Towards ecas animation

of expressive complex behaviour. Analysis of Verbal

and Nonverbal Communication and Enactment. The

Processing Issues, pages 185–198.

Moodle (2013). Open-source community-based tools for

learning. https://moodle.org. Accessed November,

2013.

Moore, M. G. (1989). Three types of interaction. The Amer-

ican Journal of Distance Education, 3:1–6.

IAAN:IntelligentAnimatedAgentwithNaturalBehaviourforOnlineTutoringPlatforms

129

Mori, M. (1970). Bukimi no tani [the uncanny valley]. En-

ergy, 7:33–35.

Poggi, I., Pelachaud, C., de Rosis, F., Caroﬁglio, V., and

de Carolis, B. (2005). GRETA. A Believable Embod-

ied Conversational Agent. pages 27–45.

Rieger, T. et al. (2003). Avatar gestures. Journal of WSCG

2003, 11:379–386.

SumTotal, S. (2013). Learning management and talent man-

agement solutions. http://www.sumtotalsystems.com.

Accessed November, 2013.

ICAART2014-InternationalConferenceonAgentsandArtificialIntelligence

130