Virtual Agent Behavior Modeling in Case of a Risky Situation in a

Virtual Electrical Substation

Dilyana Budakova

, Velyo Vasilev

, Lyudmil Dakovski

and Stanimir Stefanov

Technical University of Sofia, Plovdiv Branch, Plovdiv, Bulgaria

European Polytechnic University, Pernik, Bulgaria

Keywords: Virtual Reality Technology, Electrical Substation Modeling, Agent Modeling, Intelligent Virtual Agents,

Risky Situation, Goal Change, Priority Change, Social Power.

Abstract: In this paper, the behavior of a realistically represented intelligent virtual agent (IVA) that accompanies

students during their visit to a virtual electrical substation is modeled. The choice of technologies for modeling

the agent and the task environment is considered. The properties of the task environment are discussed. The

agent’s behavior when a risky situation occurs is investigated. For this purpose, an IVA behavior model, based

on psychological theories of motivation, emotions, and power is proposed. A change in the IVA priorities and,

as a consequence, a change in its goal is modeled. Results of a survey, studying the trust, which the IVA

receives from the students, are presented. To have a more realistic IVA, the model includes knowledge of the

environment, the shortest evacuation route learning, visitor training locations, priorities, emotions, social

power strategy, set goals, abilities to learn, abilities to change priorities and goals when a risk occurs, and a

role of a specialist – electrician.

1 INTRODUCTION

Prediction and prevention of risky situations,

disasters, and accidents unite the efforts of

researchers and experts. When they do occur, they

require an immediate response, a change in priorities

for all affected, a search for an exit, evacuation, and

damage limitation.

The behavior of intelligent agents, designed to

communicate with users in critical situations includes

some requirements for their modeling so that the

users could trust them. These are, for example:

having knowledge of the environment; having up-to-

date information about the course of events; being

able to show empathy, being able to change priorities,

having a strategy for demonstrating power, having

social communication skills, and being realistic.

A great advantage is the possibility to model the

risky situation in a way that allows presenting the risk

most realistically and, at the same time, organizing all

corresponding actions in a safe way for the users.

Virtual Reality technology (Favian, 2019, Gartner,

2018, Resnick, 2022) enables appropriate means for

achieving extremely realistic and at the same time

https://orcid.org/0000-0001-8933-9999

completely safe experiences of the modeled events by

the customers. Therefore, this technology was chosen

for conducting experiments within the presented

study.

Electrical energy is generated and used in real

time. A large part of the facilities in an electric grid

such as substations, overhead power lines, renewable

energy sources, and conductors, are located outdoors.

They are dependent on the meteorological conditions;

the change in the amount of consumed power; the

health of the facilities and many others. Power

generators and consumers may be located at great

distances. Power consumption is constantly growing

and becoming a key factor in industrial development

and social life. The electrical substation in particular

is an example of a risky working environment (B2B,

EPRI, OSHA, saVRee), which justifies its choice as

an object for conducting the experiments.

In this paper, the behavior of a realistic intelligent

virtual agent (IVA) that accompanies students during

their visit to a virtual electrical substation is modeled.

The properties of the task environment are discussed.

The agent’s behavior when a risky situation occurs is

investigated. For this purpose, an IVA behavior

Budakova, D., Vasilev, V., Dakovski, L. and Stefanov, S.

Virtual Agent Behavior Modeling in Case of a Risky Situation in a Virtual Electrical Substation.

DOI: 10.5220/0011623700003393

In Proceedings of the 15th International Conference on Agents and Artiﬁcial Intelligence (ICAART 2023) - Volume 1, pages 189-198

ISBN: 978-989-758-623-1; ISSN: 2184-433X

 2023 by SCITEPRESS – Science and Technology Publications, Lda. Under CC license (CC BY-NC-ND 4.0)

189

model, based on psychological theories of personality

motivation, emotions, and power, is proposed. A

change in the IVA priorities and, as a consequence, a

change in its goal is modeled.

The remainder of this paper is organized as

follows: Section 2 justifies the choice of the

technology Virtual Reality to recreate a risky

situation and IVA behavior in the course of its

development. Existing models in the field of electric

power engineering, using Virtual Reality technology

are also discussed.

Section 3 presents modeling the IVA behavior in

a risky situation when a change of goals and priorities

is needed. The IVA must have a Self-Transcendence

Need and knowledge of the hierarchy of all needs

according to subsection 3.1. The need for an Expert

Social Power Strategy Model using is explained in

subsection 3.2. Modeling the Agent Type, the

Performance Measure, the Task Environment, the

Actuators, and the Sensors is shown in subsection 3.3.

Section 4 considers modeling the appearance of

the IVA, its role, and the scene. Using Unity Game

Engine reasons are explained in subsection 4.1;

Modeling the appearance of the IVA and ins role are

discussed in subsection 4.2; Modeling the scene in 4.3

respectively.

Experimental settings and experimental results

are presented in Section 5 as follows: in 5.1. and 5.2.

The experimental setting of the first and second

experiments are explained; in subsection 5.3. A

Survey of the students' opinions after the conducted

experiments is presented.

The conclusion, contributions, and future work

are discussed in Section 6.

2 VIRTUAL REALITY

TECHNOLOGY AND

PROJECTS IN THE FIELD OF

ENERGETICS, USING IT

The new technologies of Virtual reality (VR),

Augmented reality (AR) and Mixed Reality (MR)

reveal new possibilities for realistic and more

impactful modeling of processes, phenomena,

behavior, objects, environment, IVAs, NPC-non-

player characters and the interrelations between them

(Flavian, 2019). Evidence of the great interest in these

technologies and the applications, using them, are the

numerous developed applications in various fields;

the ranking of these technologies in the top 10

strategic technologies for 2018 (Gartner, 2018); the

outline of trends, stimulating the development of

metaverse technologies today and over the next three

to five years (Gartner, 2018, Resnick, 2022).

With VR technology, users have the opportunity

to interact only with virtual objects in a virtual

environment. Being completely safe, they can

experience in a truly realistic way the course of life-

threatening situations, such as storms, fires, and

accidents. They can manipulate the objects with no

consequences for them in case of making a mistake.

By using VR, the user achieves the psychological

feeling of being present at the location of the

simulation. It is also called immersion or

embodiment.

The provision of innovative training approaches

through interactive 3D lessons in virtual reality is of

interest to several modern technological companies

from the electric power system (B2B, 2022, EPRI,

2022, OSHA, 2022, saVRee, 2022). The lessons

cover training in maintaining workplace safety by

complying with Occupational Safety and Health

Administration - OSHA standards (OSHA, 2022),

transformer oil sampling, familiarization with high

voltage electrical substation equipment such as power

transformers, oil circuit breakers, re-closers,

switchgear, etc., viewing 3D models of the

constituent components of the equipment, the power

line operators’ training system, etc. (Vanfretti, 2020,

Perez-Ramirez, 2019, Sier 2022). Other examples of

Immersive virtual training for substation electricians

are (Silva, 2021, Memik, 2021, Tanaka, 2017,

Hernandez, 2016).

According to the Electric Power Research

Institute – EPRI (EPRI, 2022), AI technologies

should be actively used in the electric power industry

(Vanfretti, 2020, Hernandez, 2016, Silva, 2021).

Usually, electrical substation training projects,

based on Virtual Reality technology, lack an

intelligent agent model, which actively

communicates with the learners and assists them by

guiding them.

The development of a virtual consultant-

electrician can help achieve automatic diagnosis and

monitoring of the equipment in the electrical

substation and ensure the safety of the workers. Our

motivation for modeling an IVA - consultant to

students-visitors to a virtual electrical substation was

to study its behavior, on the one hand, and to see how

it is perceived by the students.

ICAART 2023 - 15th International Conference on Agents and Artiﬁcial Intelligence

190

3 MODELING THE IVA

BEHAVIOR IN A RISKY

SITUATION WHEN A CHANGE

OF GOALS AND PRIORITIES IS

NEEDED

There are a lot of examples in the literature of solving

conflict situations, evoking mixed emotions (Lee,

et.al, 2006, Campos, et.al, 2012, Campos 2012,

Basheer, et.al, 2013). It is relevant to consider a

complex system that solves problems in a dynamic

environment, in real-time, as a rational agent that has

Beliefs, Desires, and Intentions (BDI) (Rao and

Georgeff, 1995). Rao and Georgeff propose a BDI

agent-based air-traffic management system. Desires

are seen as motivational states of the system. The

occurrence of events may require a change in the goal

chosen to be achieved by the system or the intention

(Kinny and Georgeff, 1991). A system with evaluator

CBR (Case-Based Reasoning) and advisor BDI

agents is used for a web-based risk management

system for small and medium businesses (De Paz,

et.al, 2011). BDI agents are used for social

simulations (Adam and Gaudou, 2016, Puica M. A,

Florea A. M., 2013).

In works (Moffaert, 2016, Natarajan, et.al. 2005)

an IVA is trained to achieve goals, which, to varying

degrees, satisfy the controversial requirements of the

users. Pareto front goals are those, which propose

different ways of balancing conflicting demands. The

users have to choose one that balances their

requirements in the best way. (Moffaert, 2016,

Natarajan, et.al. 2005).

Other authors focus on the way of achieving a

given goal (Budakova, et. al. 2020, 2020a). They

introduce a hierarchy of requirements concerning the

way of achieving a goal. When it is not possible to

achieve a goal by complying with all the

requirements, compromises are suggested. For this

purpose, the characteristics-requirements are

classified as acceptable and unacceptable for the

users.

The IVA is trained to make only acceptable

compromises to manage the way to achieve the goal

and not to make unnecessary compromises. An

algorithm of learning by reinforcement learning is

used, in which a matrix of the characteristics is

introduced, as well as weights, denoting their degree

of importance.

The IVA in the considered scenario is modeled to

be a bearer of self-transcendence values, i.e., to have

a self-transcendence need. According to (Maslow,

1954, Liu, 2022), it means concern for the well-being

of others, empathy (Paiva, et. al. 2017), socially

engaging emotions (Liu, 2022, Paiva, 2017), and

prosocial behavior (Liu, 2022, Paiva, 2017, Gratch,

2004). To realize this need, two goals are set in front

of the IVA. The first one is to accompany the visitors

to pre-set locations in the electrical substation. The

second goal is to ensure the safety of the visitors.

According to the scenario, at a randomly chosen

moment after the tour starts and before it ends, an

explosion occurs, requiring the evacuation of the

visitors. This means that the IVA has started actions

to achieve the first goal – visiting a sequence of

designated training locations. Before achieving it,

however, it has to stop the actions for achieving it and

undertake actions to achieve the second goal –

evacuation to save the students from the explosion.

Consequently, not the manner of achieving one

goal is managed here, but the pursuit of one goal must

be stopped and actions for the achievement of another

goal must be initiated. If we can talk about

compromises, the compromise here is to leave the

first goal unachieved to achieve the second one. For

the specific scenario, the compromise is that the

planned places for training remain unvisited to

guarantee the students’ safety.

3.1 Self-Transcendence Need and

Resulted Behavior Modeling

To realize a change of goals, the IVA must have a

Self-Transcendence Need and knowledge of the

hierarchy of all needs. In his theory of personality

motivation, Maslow defined a pyramid that illustrates

the hierarchy of needs (Maslow, 1954). For the

purposes of the experiments in this paper, only the

possibilities for the IVA to take care of user safety

and knowledge transfer are modeled.

According to Maslow's theory (Maslow, 1954),

the place of each need in the hierarchy can change

depending on the degree of their satisfaction. When,

for example, there is an explosion, great uncertainty

arises and this need is given the highest priority. The

most important thing is to protect the visitors’ life and

health. And when there is no danger, the transfer of

knowledge is valued as the most important again.

According to the theory of Ortony, Clore, and Collins

(OCC model) (Ortony, et. al. 1988), the occurring

events receive a cognitive appraisal, giving rise to

emotions. The emotions are "valenced reactions".

Following this cognitive approach, which explains

the emergence of emotions, it is assumed that the

explosion receives a negative appraisal and the state

of safety - a positive one.

Virtual Agent Behavior Modeling in Case of a Risky Situation in a Virtual Electrical Substation

191

A module of the environment called "critic"

defines the priorities of the needs, the cognitive

appraisal to be given to the events, which occur in the

task environment, and the emotions, caused by these

appraisals (Russel and Norvig, 2009). The IVA's

priorities and emotions are associated with taking

specific actions. The rules defined in the "Critic"

module cannot be changed by the IVA. These rules

tell the agent what is right and what is wrong. The

rules, determining the agent's priorities, emotions,

and actions, which correspond to them, can be briefly

described by the pseudo-code, given here:

Function Environment State (percept) returns an action

if environment _state = = exploded then

priority  Update Highest priority (

safety achieving)

evaluation  Assigning evaluation (negative)

emotion  Assigning evaluation (fear)

return action evacuation;

else if environment _state = = safety then

priority  Update Highest priority (teaching)

evaluation  Assigning evaluation (positive)

emotion  Assigning evaluation (pride)

return action next learning place;

According to this pseudo-code, if there is an

explosion, ensuring safety becomes the highest

priority for the agent. The explosion receives a

negative cognitive appraisal and as a result, the

emotions of fear and empathy arise, and immediate

action is taken to evacuate the visitors out of the

electrical substation. If there is no explosion and the

situation is safe, the most important priority for the

IVA is to be useful to users. Safety receives a positive

cognitive appraisal.

This appraisal gives rise to the emotions of joy,

and the opportunity to share knowledge evokes pride.

3.2 Expert Social Power Strategy

Modeling

The simulated agent is in the role of an electrician

who accompanies the students on their visit to the

virtual electrical substation. It is necessary for the

students to follow the agent’s instructions especially

when a critical situation arises. A number of studies

(French and Raven, 1959, Pereira, et.al. 2016,

Hashemian, et.al. 2018) prove that social power and

the use of a social power strategy have an impact on

social interaction.

It is expected that if the IVA uses a social power

strategy it will have a greater impact on the students

and they will follow it to a greater extent. According

to the Theory of Social Power proposed by French

and Raven, there are five approaches to realizing

social power: Reward Social Power; Coercive Social

Power; Expert Social Power; Legitimate Social

Power; Referent Social Power. It is also important

what type of Power Resources the agent has at its

disposal. The modeled IVA is in the role of an

electrician, which is one of its Power Resources. It

also uses Expert Social Power because it has

knowledge of the task environment, training

locations, and evacuation routes.

As a social power strategy, the agent uses phrases

like: “Let's go. I will take you out by the shortest and

safest route. I know the electrical substation very

well."; “You will enjoy the following video tutorial.

You must see it. It's just for you”.

It is assumed that this role, the knowledge, and the

used social power strategy will help the IVA to gain

the trust of the visitors. The obtained results are

discussed in section 5.

3.3 Modeling the Agent Type, the

Performance Measure, the Task

Environment, the Actuators and

the Sensors

According to (Russel and Norvig, 2009), intelligent

agent modeling and modeling the task environment

are directly related to each other. Therefore, they are

considered in this section.

The world of the virtual agent - an electrician is

simplified at the maximum. Like a simple reflexive

agent, it chooses its actions based on the current

perception and does not store a history of perceptions.

As a result of the environment monitoring, the

IVA can dynamically change its goal. This enables

the agent with the ability to cope with the uncertainty

and stochasticity of the task environment.

The agent undertakes a visit to the locations,

designated for conducting the training process. The

agent has autonomy when learning to find the shortest

safe evacuation route or the shortest route to a given

goal to conduct a video tutorial with the visitors. A Q-

learning reinforcement algorithm is applied (Sutton

and Barto, 1998).

In summary, it can be said that the modeled agent

is a learning agent, which is utility-based, has goals,

and responds to the immediately observable

characteristics of the task environment.

The task environment is known to the virtual

agent – the electrician. It can lead the students –

ICAART 2023 - 15th International Conference on Agents and Artiﬁcial Intelligence

192

visitors of the virtual electrical substation to all

intended goals; in case of an explosion, it chooses the

shortest safe route of evacuation.

For the purposes of this experiment, the task

environment is assumed to be fully observable.

The working environment is considered episodic.

For each episode, the agent perceives the state of the

environment and takes an action.

The other modeled moving-on-the-scene virtual

electricians in this experiment are only considered as

objects. They neither cooperate nor compete with

each other. They do not take part in the experiment

and do not affect the behavior of the intelligent virtual

agent – an electrician under consideration. It means

that the modeled environment is a single-agent one.

For the conducted experiments, a dynamic,

uncertain, and stochastic environment was modeled.

This is achieved by limiting the abilities of the IVA.

For example, the future state of the environment does

not depend on the actions of the virtual agent – an

electrician. The IVA is not capable of predicting or

taking actions to prevent a critical situation. The

environment is dynamic, as it can change while the

agent shows the objects in the electrical substation to

the students. For example, a storm may start, and an

accident or an explosion requiring evacuation may

occur. Consequently, the environment is also

uncertain.

Table 1: The model of the working environment.

Working environment

Model of the

electrical

substation

Including generators, transformers,

circuit breakers, gauges, etc.

Other characters

Other characters, move on the

scene.

Users

Students, visiting the electrical

substation.

Models

Model of a storm; Model of an

explosion; Model of accidents such

as short circuit occurrence,

electrical sparks, arcing etc.

Input module

Camera, microphone, keyboard,

and motion controllers;

Output module

Visualization of the scene, of the

IVA, of the occurring events.

Module "Critic"

Rules, define when and which

priority is most important; which

event receives what rating; what

emotion it evokes; what action to

take.

Table 2: The model of a Utility-Based Intelligent Virtual

Agent.

Utility-Based Intelligent Virtual Agent

Motivation

(priorities)

Need to provide safety and to pass on

knowledge;

Emotions and

empathy

Joy and anxiety

Social skills

Gait, facial expressions, gestures,

TTS, Speech Recognition; Social

power strategy.

Goals

Two goals correspond to the agent’s

needs, which can also be in conflict.

Resolving a

conflict

situation

Resolving a conflict situation by

changing the priorities and thence by

changing the goal.

Behavior

Patrolling behavior; behavior aimed

at finding out the shortest route to the

training and evacuation locations

Ability to learn

The Reinforcement Learning

Algorithm uses a reward model;

environment model; dangerous

places avoidance model.

3D model of an

IVA,

Including movements, mimics, and

gestures animation, TTS, and speech

recognition.

Knowledge

Knowledge of the environment; the

occurred accidents; of the evacuation

route; the places where a video

tutorial can be seen; the

characteristics of the objects in the

electrical substation and knowledge

of the signs of danger of an accident.

Knowledge of the needs (priorities),

the emotions they give rise to, and the

corresponding actions.

The model of the working environment and the

model of an IVA, are given in Table 1 and Table 2

respectively. The IVA and the working environment

interact continuously.

4 MODELING THE

APPEARANCE OF THE IVA,

ITS ROLE, AND THE SCENE.

4.1 Unity Game Engine Using

Unity 3D real-time development platform is used for

modeling and control of 3D interactive applications

for VR, MR, and AR (Unity Documentation, 2022).

The applications, developed with the help of the

Unity 3D Game Engine, allow for conducting

experiments in the field of artificial intelligence,

Virtual Agent Behavior Modeling in Case of a Risky Situation in a Virtual Electrical Substation

193

robotics, machine learning, and artificial intelligence

for games (Craighead, et.al, 2008, Craighead, et.al,

2008a, Buyuksaliha, et.al, 2017, Wang, et.al, 2020,

Wang, et.al, 2019, Weitnauer, 2008).

The applied programs, developed with the help of

Unity consist of scenes, game objects, and scripts,

written in the programming languages Python or C#

using the programming environment Visual Studio.

NET. The developers can add resources from the

Unity Asset Store with the help of the Package

Manager. A valuable option is to use sensors in the

developed applications.

The great capabilities of the Unity Game Engine,

its popularity, and the recognition that it is the fast

way to implement a virtual reality application were

the reasons why it was chosen for the implementation

of the experimental setup in this study.

4.2 Modeling the Appearance of the

IVA and Its Role

The IVA is modeled at full height. The 3D character

Pete from the site (Mixamo) was chosen because he

is dressed in work clothes and is suitable for the role

of an electrician to accompany the students, visiting

the virtual electrical substation (Figure 1). Again

from (Mixamo) several animations of walking,

running, waving, pointing, and searching were chosen

to make the agent even more plausible. The package

for Unity SALSA LipSync (Crazy Minnow Studio) is

used to ensure synchronization of the agent's lip

movement with the pronounced speech and the

direction of the gaze toward the speaker. The agent

can pronounce any written text with the help of the

package for Unity RT-Voice (Crosstales), which uses

TTS voices, integrated into the application under

development.

An alternative suitable package for 3D modeling

of the IVA appearance is UMA2 - universal

multipurpose agents (UMA Steering Group), which

provides opportunities to construct a unique avatar by

changing the appearance of the agent - height, weight,

placement of eyes, cheekbones, lips, as well as

clothing and hairstyle.

4.3 Modeling the Scene

To model the scene, first of all, the standard Unity

Environment package is used, which allows placing

terrain, mountains, grass, and trees, to model a windy

area. Then the sky is placed using the Classic Skybox

package (Mgsvevo, 2015). The Storm Effects

package (WeBee3D, 2012) is used to model a storm

Figure 1: A model of a virtual electrical substation with a

visualized intelligent agent – an electrician.

with thunder, lightning, and the appropriate sounds.

The Package Pixel Art Particle System Pack

(Gonzalez, 2018) makes it possible to add electric

sparks, an electric arc, explosions, fire, smoke, and

other sound and visual effects to the project. The

substation model is based on the one, proposed in the

asset store package “Electrical substation – Power

Grid” (Kobra Game Studios, 2018). All object models

in this package such as generators, transformers,

circuit breakers, cables, etc. are authentic and fully

correspond to the real ones.

Figure 2: A scene in the electrical substation during a

critical situation caused by a heavy storm.

A scene in the electrical substation during a

critical situation caused by a heavy storm is given in

Figure 2.

5 EXPERIMENTAL SETTINGS

AND EXPERIMENTAL

RESULTS

5.1 Experimental Setting of the First

Experiment

One of the goals of the simulation is for the students

to experience as realistically as possible an extreme

situation caused by a heavy storm on the territory of

the virtual electrical substation. A disaster where the

ICAART 2023 - 15th International Conference on Agents and Artiﬁcial Intelligence

194

rain is torrential, the thunder is frequent and the

lightning is so close, is dangerous to see live. The goal

is for the students to be virtually present in a risky

environment, which they cannot visit live due to

danger to life. Accidents are modeled, such as the

occurrence of short circuits, electric sparks, arcing,

dangerous amplification of the sound emitted during

the operation of the devices, cracking of the ceramic

insulation, windings of sand and deposits on the

housings of the devices, leakage of protective oil from

the circuit breakers, overheating of the transformers

necessitating turning off.

In the first experiment, a virtual agent in the role

of an accompanying electrician, advises the students

to get into a safe, fully protected flying capsule to

explore the electrical substation in this risky situation,

passing as close as possible to each object. In this

realistically modeled risky environment, they can

choose to take the tour or not. What is studied is: the

preference of the students whether to tour the

electrical substation accompanied by an IVA or

independently; whether the experience is realistic and

whether it is useful for them to see this risky situation

up close while being completely safe.

5.2 Experimental Setting of the Second

Experiment

In the second experiment, the tour around the

electrical substation begins under normal conditions.

The virtual agent offers to guide the students to

specially placed screens where they will be able to

view video lessons. The agent explains that they will

see three video tutorials on screens, located in

different places on the grounds of the electrical

substation. The topics of the video lessons are, for

example, related to the principles of operation of the

electrical substation; the most frequent accidents; the

components of the transformer, generator, and circuit

breakers; the need for the tanks to be filled with non-

conductive oil, the fans, the ceramic insulation, the

gauges to be monitored; what accidents can be caused

by the storm and outdoor operation of the devices;

what is necessary to be observed by the electricians

during a tour and inspection of the facilities.

At a random moment after the tour around the

electrical substation begins and before the end of the

third video tutorial, an explosion occurs and the agent

initiates evacuation. The students can independently

explore the substation and watch the video tutorials.

They can follow the IVA and its advice and

recommendations or ignore it. What is investigated is

whether the students prefer to communicate with the

IVA.

5.3 Survey of the Students' Opinions

after the Conducted Experiments

Twenty first-year students at our university

participated in the experiment. After completing both

of the experiments, the students must answer yes, no,

or neutral to the following survey questions:

Regarding the simulated environment: Was the

experience truly realistic for you? Was the experience

extreme? Regarding the video lessons: Was the video

lesson more impactful because of the realistic setting?

Is it a good idea to study in a realistically modeled

environment? Did you remember and understand

more and better the presented learning material as a

result of the realistically modeled setting? Do you feel

better prepared to go to a real electrical substation

after the virtual reality experience? Would you like to

walk around the model electrical substation and

watch video tutorials for each device standing right in

front of it? Regarding the behavior of the intelligent

agent: Was the virtual agent realistic? Do you prefer

to follow an IVA when visiting the virtual electrical

substation? Do you prefer to explore the electrical

substation on your own? Did the IVA have sufficient

knowledge and social skills? In addition, students

were asked to rate on a ten-point scale the realism and

usefulness of the applied models and the confidence

in the modeled IVA.

Figure 3: Results of a student rating on a ten-point scale of

the realism of the applied models and confidence in the

modeled IVA.

The result of the research showed that the students

had a real, unique, and extremely exciting experience.

They appreciate the fact that they could not be present

in the electrical substation in a risky situation of a

heavy storm and electrical accidents. 80% of the

students expressed a preference to be accompanied by

an IVA during their visit to the virtual substation.

They shared that they enjoyed the flight capsule tour

and wanted to get as close as possible to any device

or sign of an accident. The students find it useful to

Virtual Agent Behavior Modeling in Case of a Risky Situation in a Virtual Electrical Substation

195

study in a realistic setting. They expressed a

preference for the first experiment, in which they

were present in the electrical substation during an

extreme situation created by a storm, something they

could not experience in the real world.

Figure 4: Results of a student rating on a ten-point scale of

the realism and usefulness of the applied models and

confidence in the modeled IVA.

After examining the survey data, it was found that

the realism ratings of the models corresponded very

closely with the IVA confidence rating (Figure 3).

They also correspond to utility ratings although to a

less extent. This is shown in Figure 4. It seems that it

can be concluded that there is a relationship between

the degree of realism of the representation of the

models and the assessment of their usefulness, as well

as confidence in them. But this claim requires a larger

study to be confirmed.

6 CONCLUSION AND

CONTRIBUTIONS

A model of a virtual electrical substation and a model

of a realistic IVA - an electrician have been proposed

in this paper. The modeled agent is a utility-based

learning agent, having goals, and responding to

immediately observable characteristics of the task

environment. The justification of the IVA behavior

through the theories of Maslow, (Maslow, 1954)

Ortony, Clore, Collins, (Ortony, et.al., 1988) French,

and Raven (French and Raven, 1959) is considered a

contribution.

Two scenarios of a risky situation development,

caused by a storm and an explosion, have been

proposed. The behavior of an IVA in these situations

has been investigated. The emergence of risk causes

a change in the priorities and, as a consequence, a

change in the goal of the IVA. The IVA behavior

model is based on Maslow's theory of personality

motivation. A contribution is that in the process of

implementation of the present model, the agent

terminates the execution of an initially chosen goal

before achieving it and undertakes the execution of

another goal in response to the change of priorities

and the risk that has arisen. The IVA modeling also

uses the theory of Ortony, Clore, and Collis,

according to which an event receives a cognitive

appraisal, giving rise to emotions. Based on this

theory, the emotions and the motivation of the IVA

are modeled. The IVA uses a social power strategy in

accordance with the French and Raven's theories.

To gain the users' trust, the IVAs are required to:

possess knowledge, and good social skills, use social

resources and a social power strategy, express

emotions, use appropriate gestures, gait, gaze

direction, recognize and deliver a speech, make

decisions in a critical situation, have goals and

priorities. All these functionalities help to achieve a

high degree of realism of the virtual agent.

The experiments, carried out with the students in

their work with the developed scenarios of a risky

situation occurring during their visit to a virtual

electrical substation are discussed. The results show

that the behavior of the IVA is perceived as correct

and realistic. 80% of the students who participated in

the experiment preferred an IVA to accompany them

on their visit to the electrical substation. They

consider that the use of Virtual Reality technology is

suitable for modeling risky situations to be observed

in a safe environment.

The study could be continued in the future in the

following direction: after the risk has passed the agent

could go on with achieving the unachieved original

goal.

Modeling other types of risky situations and/or

opportunities to prevent and predict them is also of

interest. The IVA will get greater autonomy and a

better performance measure. The students will get

more opportunities to interact with the environment

and take action to prevent accidents in the electrical

substation. Greater functionality of the IVAs is also

envisaged, such as explaining the actions to be taken

or explaining the status of objects from the electrical

substation; learning to manage the way to achieve

their goals so as to increase their performance

measure.

It will be interesting to model a multi-agent

environment in future experiments, where the agents

located in different places on the scene of the

electrical substation will exchange information about

the state of the objects in order to cooperate.

ICAART 2023 - 15th International Conference on Agents and Artiﬁcial Intelligence

196

ACKNOWLEDGMENTS

The authors gratefully acknowledge the financial

support provided within the Technical University of

Sofia, Research and Development Sector, Project for

Ph.D. students helping 222ПД0001-19 „New

algorithms and models for working of intelligent

agents assistants in a risky environment.“

REFERENCES

Flavián, C., Ibáñez-Sánchez, S., Orús C. (2019). The impact

of virtual augmented and mixed reality technologies on

the customer experience, Journal of Business Research,

ELSEVIER, Volume 100, Pages 547-560. https://

doi.org/10.1016/j.jbusres.2018.10.050

Gartner. (2018). Top 10 strategic technology trends for

2018.[Online].https://www.gartner.com/smarterwithga

rtner/gartner-top-10-strategic-technology-trends-for-

2018

Resnick, M. (2022). Building a Digital Future: The

Metaverse, Gartner IT Symposium/Xpo™ 2022

conference in Gold Coast, Australia.

Digital engineering and magic B2B technology company.

[Online]. https://www.digitalengineeringmagic.com/

vr-training/

Electric Power Research Institute (EPRI), Inc. (2001-2022).

[Online].https://www.epri.com/thought-

leadership/artificial-intelligence

Occupational Safety and Health Administration (OSHA).

(2022). Examining Fatal Shipyard Accidents Videos

Electrical Panel Repair Results in Electrocution - 1

Fatality,[Online].https://www.osha.gov/video/shipyard

-accidents/lockout-tagout-failure

saVRee. (2022). High Quality Engineering Courses.

https://courses.savree.com/. [Online].

Vanfretti, L., Bogodorova, T., Dorado-Rojas, S. (2020).

Power System Machine Learning Applications. From

Physics-Informed Learning for Decision Support to

Inference at the Edge for Control, Tutorial2, IEEE

International conference on communications, control,

and computing technologies for smart grids, IEEE

SmartGridComm, virtual conference. https://www.

youtube.com/watch?v=evkjow9j3Bg

Perez-Ramirez, Miguel., Arroyo-Figueroa G., Ayala, A.,

(2019). The use of a virtual reality training system to

improve technical skill in the maintenance of live-line

power distribution networks, Journal, Interactive

Learning Environments, 29(3):1-18, DOI:10.1080/

10494820.2019.1587636

Sier, S., Pourakbari-Kasmaei, M., Vyatkin, V., (2022). A

taxonomy of machine learning applications for virtual

power plants and home/building energy management

systems, journal, Automation in Construction, Vol.136,

Elsevier, https://doi.org/10.1016/j.autcon.2022.104174

Silva, A. C., Cardoso A., Lamounier E. junior, Barreto, C.

L. junior, (2021). Virtual Reality for Monitor and

Control of Electrical Substations, Anais da Academia

Brasileira de Ciências, 93(1): e20200267 DOI

10.1590/0001-3765202120200267

Memik, E., Nikolic, S., (2021). The Virtual reality electrical

substation field trip: Exploring student perceptions and

cognitive learning, STEM Education, 1 (1): 47–59,

Tanaka, E.H., Paludo, J.A., Bacchetti, R., Gadbem, E.V.,

Domingues, L.R., Cordeiro, C.S., Giraldi, O., Gallo,

G.A., Silva, A.M., & Cascone, M.H. (2017). Immersive

virtual training for substation electricians. 2017 IEEE

Virtual Reality (VR), 451-452.

Hernandez, Y., Pérez-Ramírez M., (2016). Architecture of

an Intelligent Training System based on Virtual

Environments for Electricity Distribution Substations,

Article in Research in Computing Science, DOI:

10.13053/rcs-129-1-7

Lee, B. P., Kao, E. Ch., Soo, V., (2006). Feeling

Ambivalent: A Model of Mixed Emotions for Virtual

Agents,

IVA, LNAI 4133, Springer, pp. 329-342.

Campos, H., Campos, J., Martinho, C., Paiva, A., (2012).

Virtual agents in conflict, IVA'12: Proceedings of the

12th international conference on Intelligent Virtual

Agents, Pages 105–111 https://doi.org/10.1007/978-3-

642-33197-8_11

Campos, H., (2012). CONFLICT: Agents in Conflict

Situations, Msc thesis, Instituto Superior Técnico

Basheer, G. S., Ahmad, M. Sh., Tang, A. Y. C. (2013). A

Framework for Conflict Resolution in Multi-agent

Systems, International Conference on Computational

Collective Intelligence, DOI:10.1007/978-3-642-

40495-5_20

Rao, A. S., Georgeff, M. P., BDI Agents: From Theory to

Practice, Proceedings of the First International

Conference on Multi-Agent Systems (ICMAS-95), San

Francisco, USA, June, 1995.

Kinny, D., & Georgeff, M.P. (1991). Commitment and

Effectiveness of Situated Agents. International Joint

Conference on Artificial Intelligence.

De Paz, et.al., (2011). A multiagent system for web-based

risk management in small and medium business.

Journal: Advances in intelligent and soft computing.

Carole Adam, Benoit Gaudou, (2016). BDI agents in social

simulations: a survey. Knowledge Engineering

Review, 31 (3), pp.207-238.

Puica, M. A., Florea, A. M., (2013). Emotional Belief-

Desire-Intention Agent Model: Previous Work and

Proposed Architecture. (IJARAI) International Journal

of Advanced Research in Artificial Intelligence,Vol.2.

Moffaert, K. V., (2016). Multi-Criteria Reinforcement

Learning for Sequential Decision Making Problems,

Dissertation for the degree of Doctor of Science:

Computer Science, Brussels University Press, ISBN

978 90 5718 094 1.

Natarajan, S., Tadepalli, P., (2005). Dinamic Preferences in

Multi-Criteria Reinforcement Learning. 22nd

International Conference on Machine Learning. Bonn,

Germany.

Budakova, D., Petrova-Dimitrova, V., Dakovski, L.,

(2020). Intelligent virtual agent, learning how to reach

a goal by making the least number of compromises, 9-

Virtual Agent Behavior Modeling in Case of a Risky Situation in a Virtual Electrical Substation

197

th international scientific conference "TechSys2020",

Plovdiv, Bulgaria, IOP Conf. Series: Materials Science

and Engineering 878 (2020) 012030, IOP Publishing.

Budakova, D., Petrova-Dimitrova, V., Dakovski, L.,

(2020a). Virtual agents, learning how to reach a goal by

making appropriate compromises, International

Scientific Conference, Computer Science’2020, 18-21

October, Velingrad, Bulgaria.

Maslow, A. H. (1954). Motivation and personality, (First

ed.). Harper & Row. ISBN 978-0-06-041987-5

Liu, P., Zhang, R., Li, D., (2022). The effect and

mechanisms of self-transcendence values on durable

happiness. Advances in Psychological Science, 30(3):

660-669.

Paiva, A., Leite, I., Boukricha, H., Wachsmuth, I., (2017).

Empathy in virtual agents and robots: A survey, ACM

Transactions on Interactive Intelligent Systems (TiiS) 7

(3), 1-40

Gratch, J., Marsella, S., A domain-independent framework

for modeling emotions. (2004). Journal of Cognitive

Systems Research 5(4) 269–306

Ortony, A., Clore, G. L., Collins, A., (1988). The Cognitive

Structure of Emotions. Cambridge University Press,

Cambridge, UK

Russell, S., Norvig, P., Artificial Intelligence, A Modern

Approach, (2009). Third Edition, Prentice Hall Series

in Artificial Intelligence, 2009, ISBN-13: 978-0-13-

604259-4, ISBN-10: 0-13-604259-7

Sutton, R., Barto, A., (1998). Reinforcement Learning: An

Introduction. MIT Press.

French, J. R. P., Jr., & Raven, B. (1959). The bases of social

power. In D. Cartwright (Ed.), Studies in social power

(pp. 150–167). Univer. Michigan.

Pereira, G., Prada, R., Santos, P. A., (2016). Integrating

social power into the decision-making of Cognitive

Agents, Article in Artificial Intelligence, Volume 241,

pp. 1-44, Elsevier, DOI: 10.1016/j.artint.2016.08.003

Hashemian, M., Prada, R., Santos, P. A., Mascarenhas, S.,

(2018). Enhancing Social Believability of Virtual

Agents using Social Power Dynamics, IVA ’18,

November 5–8, Sydney, NSW, Australia, ACM ISBN

978-1-4503-6013-5/18/11.,

https://doi.org/10.1145/3267851.3267902

Unity Documentation. (2022). Available: [Online].

https://docs.unity3d.com/Manual/index.html

Craighead, J., Burke J., Murphy R. (2008). Using the Unity

Game Engine to Develop SARGE: A Case Study,

Proceedings of the Simulation Workshop at the

International Conference on Intelligent Robots and

Systems (IROS 2008), 4552

Craighead, J. D., Gutierrez R., Burke J., Murphy R. R.,

(2008a). Validating The Search and Rescue Game

Environment As A Robot Simulator By Performing A

Simulated Anomaly Detection Task, IEEE/RSJ

International Conference on Intelligent Robots and

Systems, Acropolis Convention Centre, Nice, France.

Buyuksaliha, I., Bayburta, S., Buyuksaliha, G., Baskaracaa,

A.P., Karimb, H., and Rahman, A.A. (2017). 3d

modelling and visualization based on the unity game

engine – advantages and challenges, 4th International

GeoAdvances Workshop, Safranbolu, Karabuk, Turkey,

DOI:10.5194/isprs-annals-IV-4-W4-161-2017.

Wang, Z., Liao, X., Wang, C., Oswald, D., et. al., (2020).

Driver Behavior Modeling Using Game Engine and

Real Vehicle: A Learning-Based Approach, IEEE

Transactions on Intelligent Vehicles, vol. 5, no. 4.

Wang, Z., Wu, G., Boriboonsonsin, K., Barth, M., et. al.,

(2019). Cooperative ramp merging system: Agent-

based modeling and simulation using game engine, SAE

Int. J. Connected Autom.Veh., vol. 2, no. 2, 2019.

Weitnauer, E., Thomas, M. N., Rabe, F., and Kopp, S.,

(2008). Intelligent Agents Living in Social Virtual

Environments, Bringing Max into Second Life, IVA

2008: Intelligent Virtual Agents, pp 552–553.

Mixamo. Animate 3D characters for games, films, and

more. https://www.mixamo.com/#/ Available [Online]

Crazy Minnow Studio. SALSA-Simple Automated Lip

Sync Approximation. Unity Asset Store.

https://assetstore.unity.com/packages/tools/animation/

salsa-lipsync-suite-148442 Available [Online]

Crosstales. RT-Voice PRO. Unity Asset Store.

https://assetstore.unity.com/packages/tools/audio/rt-

voice-pro-41068 Available [Online].

UMA Steering Group. UMA2-Unity Multipurpose Avatar.

Unity Asset Store. Available [Online].

https://assetstore.unity.com/packages/3d/characters/um

a-2-unity-multipurpose-avatar-35611

Mgsvevo. (2015). Classic Skybox. Unity Asset Store.

Available [Online]. https://assetstore.unity.com/

packages/2d/textures-materials/sky/classic-skybox-

24923

WeBee3D. (2012). Storm Effects. Unity Asset Store.

[Online].https://assetstore.unity.com/packages/vfx/part

icles/environment/storm-effects-5048

Gonzalez J., (2018). Pixel Art Particle System Pack. Unity

Asset Store. Available [Online]

https://assetstore.unity.com/packages/vfx/particles/pixel-

art-particle-system-pack-80249

Kobra Game Studios. Electric substation (Power Grid).

Unity Asset Store. Available [Online] https://

assetstore.unity.com/packages/3d/props/industrial/elec

tric-substation-power-grid-133434#content.

ICAART 2023 - 15th International Conference on Agents and Artiﬁcial Intelligence

198