A Method of Conceptual Modelling for Realistic Training Scenarios

Inna Shvartsman

, Kuldar Taveter

and Merik Meriste

Tallinn University of Technology, Department of Informatics, Raja 15, 12618 Tallinn, Estonia

Research Lab for Proactive Technologies, Tallinn University of Technology, Raja 15, 12618 Tallinn, Estonia

Keywords: Agent-oriented Modelling, Socio-technical Systems, Simulations.

Abstract: Training is crucial for improvement of the capabilities of both military and non-military personnel. In this

paper, we argue for the need of conceptual modelling for the creation of training scenarios. This research

proposes a particular method for developing training scenarios for complex domains based on agent–

oriented modelling. The advantage of agent-oriented modelling is that it enables to describe a problem

domain from three balanced and interrelated aspects – interaction, information, and behaviour, and at three

abstraction layers: analysis, design, and simulation. Thus, we can obtain agent-directed simulations for

elaborating selected aspects of emergent behaviour to support development of practical training scenarios in

a partially known environment.

1 INTRODUCTION

Training is of critical importance in preparing both

militaryand non-military personnel for peacekeeping

missions. Composing practical scenarios for training

to cope in an environment with several potential

asymmetric threats is far from trivial. In this study

we rely on realistic case studies.

A tabletop exercise is an exercise that is designed

to test the ability of a group to respond to a situation.

In a typical tabletop exercise, a facilitator creates a

setting in which each player plays a predefined role.

The facilitator also describes the situation to be

responded to and the physical environment.Training

is often conducted as a role play in the form of a

tabletop exercise. In training exercises of this kind

predefined training scenarios are used. In the

exercise, the participants act out the scenario,

whereby the scenario changes in response to their

actions and random factors, which may be

determined by rolling dice or drawing cards. This

implies that the results of a tabletop exercise cannot

be predicted as they always emerge from conducting

the concrete exercise.

Computer-based training opens new possibilities

for conducting tabletop exercises. The method we

use for computer-based training is based on

representing individuals or groups of individuals as

agents, where an agent is defined as an entity that is

(a) reactive; (b) proactive; (c) social; and (d) situated

in some environment. The particular training

scenario emerges from agents' activities, thus - by

tuning agent's behaviour one can tune the simulation

behaviour.

A clear advantage of this method is that

training doesnot depend on the number of trainees

because some roles in training can be played by

software agents and one or more roles – by trainees.

Moreover, with this kind of training, we can also

explore experiments with different psychological

profiles of trainees played by software agents. As a

result, we can create many different team setups and

embed a trainee or trainees in them. There is no

tabletop exercise that can provide this kind of

experience. In a tabletop exercise, the goals to be

achieved by the scenario, as well as the activities to

be performed by different participants and the

criteria for evaluating these activities should be

clearly outlined. As it has been pointed out in

(Vattam et al., 2011), preparation of training

scenarios rooted in agent-based simulation, starting

from agents themselves, requires a proper

conceptualization of the problem domain at hand.

This paper proposes a method of conceptual

modelling for computer-based training, where the

training scenarios are created based on a set of

conceptual models. The method is overviewed in

Section 2. The process of conceptual modelling is

described in Section 3 by elaborating an example

from (Shvartsman et al., 2010). Finally, conclusions

are presented in Section 4.

443

Shvartsman I., Taveter K. and Meriste M..

A Method of Conceptual Modelling for Realistic Training Scenarios.

DOI: 10.5220/0004333404430446

In Proceedings of the 5th International Conference on Agents and Artiﬁcial Intelligence (ICAART-2013), pages 443-446

ISBN: 978-989-8565-38-9

 2013 SCITEPRESS (Science and Technology Publications, Lda.)

2 THE METHOD

The method that we use for conceptual modelling

for realistic training scenarios is agent-oriented

modelling (Sterling and Taveter, 2009). We prefer

this method over other alternatives because it

straightforwardly enables to model a system of goals

to be achieved by a training scenario, as well as the

activities to be performed by players in the scenario

different participants and the criteria for evaluating

these activities should be clearly outlined. We also

have a lot of positive anecdotal experience of

applying AOM in the related domain, where AOM

was successfully applied for developing a training

resource to teach secondary school students to

respect people with Asperger’s Syndrome.

Agent-oriented modelling is an approach for

modelling and simulating the behaviours of complex

socio-technical systems where a problem domain is

first conceptualized in terms of the goals to be

achieved by the system, the roles required for

achieving them, and the domain entities embodying

the required knowledge. The roles are thereafter

mapped to the agents playing the roles, the goals – to

the activities performed by the agents, and the

domain entities – to the items of knowledge held by

the agents. As we are concerned with “human-in-

the-loop” simulations, the term “agent” subsumes

both human agents and man-made agents –

softwareagents simulating humans. Conceptually,

we consider models as abstractions reducing the

complexity of a system for better understanding of

the system’s particular aspects and their impact on

its behaviour.

The types of models proposed by agent-oriented

modelling (AOM) are represented in Table 1. In

addition to representing for each model the

abstraction layer (analysis, design, or simulation),

Table 1 maps each model to the vertical viewpoint

aspect of interaction, information, or behaviour.

Each cell in the table represents a specific viewpoint.

We will next give an overview of agent-oriented

models relevant for understanding this article

proceeding by viewpoints. These models are

distinguished by using a bold font in Table 1.

From the viewpoint of behaviour analysis, a goal

model can be considered as a container of three

components: goals, quality goals, and roles (Sterling

and Taveter, 2009). A goal is a representation of a

functional requirement for the simulation system,

describing the phenomenon or process to be

simulated. A quality goal, as its name implies, is a

non-functional or quality requirement of the system.

Goals and quality goals can be further decomposed

into smaller related subgoals and subquality goals.

The hierarchical structure is to show that the

subcomponent is an aspect of the top-level

component. Goal models also determine roles that

are capacities or positions that agents playing the

roles need to contribute to achieving the goals. The

notation for representing goals and roles is shown in

Table 2 (Sterling and Taveter, 2009).

From the viewpoint of interaction analysis, the

properties of roles are expressed by role models. A

role model describes the role in terms of the

responsibilities and constraints pertaining to the

agent(s) playing the role.

From the viewpoint of interaction design,

interaction models represent interaction patterns

between agents of the given types. They are based

on responsibilities defined for the corresponding

roles. In this paper, we represent interaction models

by means of action events and non-action events. An

action event is an event that is caused by the action

of an agent, like sending a message or starting a

machine. An action event can thus be viewed as a

coin with two sides: an action for the performing

agent and an event for the perceiving agent. A

message is a special type of action event—

communicative action event—that is caused by the

sending agent and perceived by the receiving agent.

On the other hand, there are non-action events that

are not caused by actions. Non-action events include

exogenous events. An exogenous event is a kind of

event whose creating agent we are not interested in.

Finally, from the viewpoint of behaviour design

behaviour models describe the behaviours of

individual agents (Sterling and Taveter, 2009).

Table 1: The Model Types of Agent-Oriented Modelling.

Viewpoint aspect

Abstraction

layer

Interaction Information Behaviour

Analysis

Role

models and

organization

model

Domain

model

Goal models

and

motivational

scenarios

Design

Agent

models and

interaction

models

Knowledge

models

Scenarios and

behaviour

models

Simulation Platform-specific models

3 CONCEPTUAL MODELLING

In this section we show how a trainingscenario that

ICAART2013-InternationalConferenceonAgentsandArtificialIntelligence

444

has been used, assessed, and elaborated in

numerous psychological experiments (Parmak et al.,

2010) can be conceptually modelled by AOM for

computer-based simulations with emergent

behaviour. The firstmodel to be created is the goal

model that determines the overall purpose of the

simulation and its subgoals. This model serves to

discuss the purpose of the simulation with all the

stakeholders involved: military commanders and

experts, trainers, trainees, adventure games’experts,

etc. As is reflected by Figure 1, the overall purpose

of the simulation is to evacuate the building.

Achieving the purpose can be divided into the

following subgoals, each of which represents a

particular aspect of the evacuation: penetrate into the

building, help the injured, ensure safety inside,

ensure safety outside, and collect and pass

information. Each subgoal can, in turn, be divided

into third-level subgoals. Figure 1 represents the

refined subgoals for the “Help the injured” subgoal.

For clarity, the other subgoals are elaborated in

separate figures which we do not present here

because of space constraints. Achieving a goal may

be characterized by a quality goal which in the given

context represents the criteria for evaluating the

extent to which the goal in the simulation has been

achieved. The goal model also shows the roles that

are required for achieving the goals of the simulation

scenario. The roles are separately modelled further

on in this section.

Figure 1: The goal model for the urban operation.

As was described in Section 3, the roles are

described in terms of the responsibilities and

constraints applying to the agents that will perform

the roles. Because of the scope of this paper, we

present in Table 2 only the model of the roles

External Safeguard.

Having defined the goals for the scenario to be

simulated and the roles comprised by the scenario,

we can start designing simulations in such a way that

any role in the simulation system could be

Table 2: The Role Model of External Safeguard.

Role name External Safeguard

Description The role of the external safeguard

of the building during the operation

Responsibilities Ensure safety outside the building

Inform the Communication

Responsible about any potential

threats

Receive the injured from the

Internal Safeguard along with the

instructions

Inform the Communication

Responsible about the injured

received and the instructions

Constraints Quick, efficient, informed, and

helpful behaviour

performed by either a human agent or a software

agent. This enables to perform training simulations

in teams of any size and evaluate the performance of

individual human agents. We illustrate platform-

independent design by presenting in Figure 2 an

interaction model for the scenario. The interaction

model depicted in Figure 2 includes the roles of

three purposeful agents – External Safeguard,

Internal Safeguard, and Communication Responsible

– whose goals comply with the goals set for the

simulation scenario by the goal and role models. In

addition, the interactions involve the role Physician

that is not represented in this figure. Corresponding

to the notation represented in Figure 1 and according

to the explanations provided in Section 3, the

interaction model represents the interactions

between agents performing the above-mentioned

roles as action events. In addition, the interaction

model includes two non-action events representing

the cave-in and appearance of strangers.

Distinguishing between action events and non-action

events is crucial in the simulation of military

operations. We have decided to model the non-

action events as exogenous events because both of

them are generated by the simulation environment.

Please note that the notation used in Figure 2 does

not prescribe any order for the occurrence of events.

The behaviour of an agent playing the role of

External Safeguard can be described by rules, such

as the following ones:

ON Strangers THEN Inform Communication

Responsible

ON Strangers THEN Wait N Sec; Inform

Communication Responsible

ON Strangers THEN Engage Strangers

ON Strangers THEN Take Strangers

Hostage

ON Strangers THEN Interrogate and

Search Strangers

AMethodofConceptualModellingforRealisticTrainingScenarios

445

Figure 2: Interaction model for the scenario to be simulated.

The first two and the last three rules presented

above are to be applied as alternatives rather than in

parallel. For example, upon the appearance of

strangers, an agent performing the External

Safeguard role may inform another agent playing the

Communication Responsible role right away or only

after waiting for a specified number of seconds.

Similarly, the same agent may react to the

appearance of strangers in one of the following three

ways: (a) engage strangers right away, that is, open

fire at them; (b) take strangers hostage “just in case”;

What alternatives are chosen also depends on the

psychological profiles of the simulated agents. How

the profile can be represented and how it influences

agent behaviour depends on platform-specific

design, that is, on the agent architecture and

platform chosen. This will be described in our future

papers.

4 CONCLUSIONS

Conceptual modelling is of crucial importance for

developing appropriate training scenarios for

complex social processes such as addressing

asymmetric threats in a city environment or winning

hearts and minds. We demonstrated how a training

scenario can be developed by means of AOM. In

these kinds of training scenarios, emergent

behaviour can occur at least in one of the following

two ways: (a) through different latencies of

simulated exogenous events; (b) through different

alternative behaviours of participating agents and the

combinations of their behaviours and interactions. In

our future work, we will demonstrate how a family

of training scenarios can be generated from

conceptual models based on agents’ behavioural and

interaction patterns.

ACKNOWLEDGEMENTS

This research was supported by European Social

Fund´s Doctoral Studies and Internationalization

Programme DoRa.

REFERENCES

Sterling, L., Taveter, K., 2009. The art of agent-oriented

modeling, MIT Press. London.

Shvartsman, I., Taveter, K., Parmak,M., Meriste, M., 2010.

Agent-oriented modelling for simulation of complex

environments. In IMCSIT 2010, The International

Multiconference on Computer Science and

Information Technology, IEEE Computer Society, 209-

216.

Vattam, S. S., Goel, A. K., Rugaber, S., 2011. Behavior

patterns: Bridging conceptual models and agent-based

simulations in interactive learning environments. In

IEEE 11th International Conference on Advanced

Learning Technologies, 139-141.

Parmak, M., Mylle, J. J. C., Euwema, M. C., 2010.

Personality and the perception of situational structuredness

in a military environment: Seeking and enjoying

sensation versus structure as a soldier. Journal of

Applied Social Psychology (accepted).

ICAART2013-InternationalConferenceonAgentsandArtificialIntelligence

446