COMMUNICATION AT ONTOLOGICAL LEVEL IN

COOPERATIVE MOBILE ROBOTS SYSTEM

Lucia Vacariu, Mihai Chintoanu

Department of Computer Science, Technical University of Cluj Napoca, 26 Baritiu str, Cluj Napoca, Romania

Navigon Company Cluj Napoca Subsidiary, Cluj Napoca, Romania

Gheorghe Lazea, Octavian Cret

Department of Automation, Technical University of Cluj Napoca, Cluj Napoca, Romania

Department of Computer Science, Technical University of Cluj Napoca, Cluj Napoca, Romania

Keywords: Mobile robots, Multiagents system, Communication, Ontology, Cooperation.

Abstract: Mobile robot applications are faced with the problem of communicating large amounts of information,

whose structure and significance changes continuously. A traditional, layered style of communication

creates a cooperation problem as fix protocols and message meaning cannot mediate the dynamic, novel

types of behaviors mobile robots are acquiring in their environment. We propose by contrast, a non-

hierarchical communication control mechanism based on the software paradigm of multiagent systems that

have specialized ontologies. It is a communication at ontological level, which allows efficient changes in the

content of the messages among robots and a better adaptability and specialization to changes. The intended

application is a cooperative mobile robot system for monitoring, manipulating and cleaning in a

supermarket. The focus of the paper is to simulate a number of well-defined, controllable and repeatable

critical ontological situations encountered in this environment that validate the system cooperation tasks.

1 INTRODUCTION

Research in mobile robotics has achieved a level of

maturity that allows applications to move from

laboratory into public-level applications. This leads

necessarily to a situation where several robots,

perhaps with different goals and different internal

representations, must cooperate for achieving shared

and not reciprocally damaging goals (Arai, Pagello,

and Parker, 2002).

A robot control system based on the paradigm of

multiagent systems can provide rich, active entities

that can adapt, represent and communicate

appropriate abstractions (Vacariu, et al, 2004).

Communicating relevant information among

heterogeneous units not designed to cooperate in the

first place raises several challenges. The

communication protocols cannot have a fix

hierarchical organization as such solutions have

been shown to lack flexibility for expanding and

adapting knowledge and behavior for cooperating

robots. The precondition is that the communication

between robots is enforced by a number of

component multiagents that have an efficient

exchange of information based on a large spectrum

of ontologies.

It has been recognized that an efficient and

correct communication will improve the

characteristics and the working of whatever

multiagent system (Russell and Norvig, 2002).

In the design of multiagent systems, it is possible

to introduce an ontological level, where specific

concepts for different relevant domains of

applications are described (DiLeo, Jacob, and

DeLoach, 2002). This level can be an information

source for multiagents.

Starting from these premises, we want to

demonstrate that the use of communication at the

ontological level of information representation is a

reliable method to obtain a correct functioning of a

system of mobile robots acting in cooperation.

455

Vacariu L., Chintoanu M., Lazea G. and Cret O. (2007).

COMMUNICATION AT ONTOLOGICAL LEVEL IN COOPERATIVE MOBILE ROBOTS SYSTEM.

In Proceedings of the Fourth Inter national Conference on Informatics in Control, Automation and Robotics, pages 455-460

DOI: 10.5220/0001620904550460

 SciTePress

The paper is structured as follows: Section 2

shows briefly the benefits of introducing the

ontological level in the multiagent systems design,

to achieve the information exchange. Section 3

describes a cooperative mobile robots system for

monitoring, manipulating and cleaning in a

supermarket and explains the use of ontological

level in communication. Section 4 reports the test

conditions and results obtained in the simulation.

Section 5 presents conclusions and research ideas for

future work.

2 ONTOLOGY IN MULTIAGENT

SYSTEM DESIGN

Ontology denotes here a common vocabulary with

basic concepts and relationships of a domain (Noy

and McGuinness, 2001). The domain description

requires representations for types and properties of

objects, and for relations between them. Software

agents use ontologies and knowledge databases on

ontologies as information sources.

The ontology classes describe concepts of a

domain in a taxonomic hierarchy. The roles describe

classes’ properties and have different values for

different instances. They may be restricted or not to

a certain set of values. The knowledge database is

created from individual instances of classes, with

specific values for properties, and supplemental

restrictions.

The aim of ontology integration in a multiagent

system design is a description of information using

the ontological level of knowledge representation. A

common vocabulary of a domain allows definitions

of basic concepts and relations between them to be

included in the design process of a multiagent

system. This creates premises to obtain a new

system with new facilities, that is also more robust

and adaptable.

We use the framework of MultiAgent System

Engineering MASE (DeLoach, Wood, and

Sparkman, 2001) that starts from an initial set of

purposes and makes the analysis, design and

implementation of a functional multiagent system.

The ontology can be built during the analysis stage

and after that is being used to further create new

goals for the system. Purposes often involve

parameter transmissions and consequently the

ontology classes can be used as parameters.

Objects of the data model are specified as

parameters in inter-agent conversations. In role

refining and conversation building, that involves

meta-message transmissions, that includes the type

specification for transmitted parameters. The actions

can use information contained in the parameter

attributes because the types of parameters and

attributes of types are all known. The internal

variables representing purposes and conversations

can be standardized with respect to the system

ontology. The validity of conversations is

automatically verified using parameters and

variables.

Communication is achieved by inter-agent

conversation. A conversation defines the

coordination protocol between two agents and uses

two communication diagrams between classes: one

for the sender and the other one for the receiver. The

communication diagram between classes is made of

finite state machines, which define the states of

conversation between the two participant classes.

The sender always starts a conversation by

sending the first message. Any agent who receives a

message compares it with all the active

conversations from his list. If a match is found, the

required transition is made, and the new required

activities of the new state are achieved. If there is no

match, the agent compares the message with all the

other possible messages he might have with the

sender agent, and if it finds one that matches, it will

start a new conversation.

To be able to participate in conversations, the

agents use information from disposition diagrams

where the name, address and configurations of

agents and stations are saved.

Conversations have no blocking states; all the

states have valid transitions from where it is possible

to reach the final state.

The transaction syntax uses UML notation:

receive_message(arg1)[condition]/action^transmit_

message(arg2).

Robot systems made by autonomous mobile

robots execute missions in spatial environments. The

environment description using a spatial ontology

will highlight entities with respect to space. The

spatial ontology defines concepts used to specify

space, spatial elements and spatial relations.

Therefore, when we create the ontology for the

multiagent system used in multirobot system, we

insist on physical objects, on concepts that describe

the environment using spatial localization of the

components. The terms of the concept list are

organized in classes and attributes, and an initial

model data is elaborated. The necessary concepts of

the system are specified for purpose achievement.

The communication in multiagent systems uses then

the concepts of the developed ontology.

ICINCO 2007 - International Conference on Informatics in Control, Automation and Robotics

456

3 COOPERATIVE MOBILE

ROBOTS SYSTEM

To verify the effectiveness of using the

communication at the ontological level in a mobile

multirobot system we designed and implemented a

multiagent system. The mobile multirobot system

has the purpose to serve, in cooperation, a

supermarket. The goals the system must provide are

supervising the supermarket, manipulating objects of

different types and dimensions, cleaning and giving

alarm if necessary. Logging of all events and

situations is also necessary.

The multirobot system is cooperative. All the

robots have the goal to accomplish activities in

collaboration.

3.1 Multiagent System Design

The specification for the cooperative multiagent

system gives the following general objectives: to

supervise the supermarket; to move in the assigned

zone in the market; to identify forms or objects that

do not match the knowledge about the supervised

environment and notify theirs positions; to

manipulate objects (pick up, move and push); to

coordinate activities; to exchange information

(sending and receiving); to select the appropriate

agent with respect to activities (by competencies,

positions in space, costs); to clean, and to log events.

The system’s objectives are grouped in a

hierarchy, based on the importance and connections

between them (Figure 1).

Figure 1: Objectives hierarchy.

From the initial specifications of the system, one

defines the base scenarios for identifying the

communication ways, like obstacle movement, zone

cleaning, and intruder identification.

Using these cases, we build sequence diagrams

from where the initial roles are then identified.

In the multiagent system design we highlight the

defining elements of the domain described by the

ontology and the inter-agent communication

methodology based on using ontology concepts.

The taxonomy of the used ontology and part of

the concepts are taken from the SUMO (Suggested

Upper Merged Ontology) ontology (SUMO, 2006).

We insist here on physical entities, with necessary

add-ons for multirobot system applications. The

ontology has a larger information domain than

necessary for this particular application, and allows

simple updates. This facility will assure their usage

also in futures applications.

The ontology base concept is the Entity, the

central node of the hierarchy. In the tree we have

PhysicalEntity and AbstractEntity. We consider

every concept that may be looked like an entity with

spatial and temporal position as a PhysicalEntity. In

this category, a distinction is made between Object

and Process. The Object concept has complete

presence in any moment of his life. The ontology

tree is developed with concepts down to the

multirobot application agent’s level. Object can be

Group or Individual. We develop the Individual

concept with Region, Substance and

ConectedObject.

The objectives structures and sequence diagrams

were converted in roles and goals associated with

them. The model of roles includes roles, everyone

goals and also information about goals interactions

(Figure 2).

Figure 2: Roles model.

3.2 Multiagent System Conversations

From the role model, the multiagent system

implements different agent types: supervisor,

coordinator, transporter, cleaner, mobile agent, and

communication agent. The agent’s class diagram

highlights the roles and the conversations between

different types of agents. For every identified

COMMUNICATION AT ONTOLOGICAL LEVEL IN COOPERATIVE MOBILE ROBOTS SYSTEM

457

communication, two conversation diagrams are

necessary, each with a sequence of steps. For

instance, obstacle movement has conversation

diagrams for supervisor, coordinator, and

transporter. Every sent message has a name and

content. The content can be missing if the message

has only control role.

In our approach, the message content is always

an ontology object. By this kind of messages we can

transmit among agents any concept from the system

ontology. Each time when information about a

system object is exchanged, an ontology object will

be in fact transmitted. The minimum information of

content is the type, coordinates and dimensions of an

object.

In the multiagent system built to serve a

supermarket, the obstacles that must be moved can

be different objects from the ontology. For instance,

when the coordinator sends a message to the

transporter for moving an object, this object can be

taken from two conceptual categories, with the same

parent ConnectedObject. One concept is Artifact –

Product where the obstacle can be Desk, Table,

Chair, Case, Bin, and the other concept is

OrganicObject, that can be Human or Plant. The

transporter will send to the supervisor a message

with an object that is itself, TransporterAgent, and

has the coordinates, self-identifier, and other

information necessary for mission completion. In the

ontology taxonomy, TransporterAgent is part of the

MobileAgent concept, which identifies the mobile

robots used in mobile multirobot system, and comes

from Artifact – Device – Robot concepts.

The agents are instantiated and put into a

network diagram. A number, type and location

identify them. The built multiagent system is

dynamic. Therefore in every moment new agents

can be introduced. Agents are placed on the same

computer or in remote computers, based on their

association with mobile robots. The collective

communication is provided by broadcast

transmission. Every agent is a separate execution

thread and has one’s own port for sending and

receiving messages.

By using the method of information exchange at

the ontological level, it is possible to share

knowledge from the entire ontology between any

agents from the multiagent system. This improves

the capabilities of the system and assures a reliable

execution of the missions, a better adaptability and

specialization to changes.

4 SIMULATION CONDITIONS

AND RESULTS

The multiagent system implementation is made in

Java language, with IntelliJ IDEA support (IntelliJ

IDEA, 2006). Conversations structure uses

agentMom (Message Oriented Middleware),

component of agentTool (MASE developer)

(agentTool, 2005). The application uses

BroadcastServer interface, implemented by every

agent or agent component. An agent can share

different types of conversations. The agents and

conversations are in different threads of execution.

The agents run in parallel, independently on each

other. All conversations are made at the same time.

The ontology classes are instantiated and

transmitted in messages content. The system’s

agents have different knowledge from the ontology.

For instance, the supervisor agent knows all kinds of

objects used in conversations because it is the one

who identifies the objects and decides what type

they have. The coordinator has the same knowledge

like the supervisor because it makes the connection

with all the agents from the system. The transporter

and cleaner agents know only sub-trees of ontology,

those in correspondence with their activity areas.

The concepts used in conversations from

ontology are Case for obstacle, ConnectedObject for

dirt and Human for intruder.

Objects use length, width, and height

dimensions. The height is used to compute the

volume. The intruders don’t use dimensions because

only their position is important for the system. The

positions of objects are implemented in the Object

concept.

We implement a WorldInstance class with

information about the types and positions of agents

and objects, the minimum, maximum, and implicit

dimensions, and the work area. A part of this

information can be modified.

The multiagent system developed for the mobile

multirobot system has been tested in simulation

conditions. We built a simulation framework for sets

of activities. The agents are associated with robots

having different functionalities. Possible problems

of synchronization and sharing resources have been

solved.

Agents have windows associated with them.

These are placed in tabs in main window of the

application interface used for messages exchange

and actions. These messages show that the

ontological information is indeed exchanged. The

coordinator has two supplementary windows, to

manage the control of supervisors and transporters.

ICINCO 2007 - International Conference on Informatics in Control, Automation and Robotics

458

We chose to simulate a number of well-defined,

controllable and repeatable critical ontological

situations encountered in this environment. The

functional situations tested were: detection of

obstacles, dirt, and intruders in the supervised area,

moving detected obstacle, cleaning detected dirt,

moving supervisors in other areas, new transporter

or cleaner agent with better capacities insertion,

coordination of supervisors, coordination of

transporters and cleaners, alarm activation, and

actions logging.

In the environment described in Figure 3.a, the

system allows detection of two cases, Square and

Rectangle, by supervisors S0 and S1. S0 and S1

detect both the Square, but just one transporter is

sent to move it. The transporters will move those

cases that are the closest from them. The other case

and dirt located outside the supervisors areas won’t

be moved (Figure 3.b). To detect these case and dirt,

S1 supervisor is moving in another area (Figure 3.c).

After moving the case by T1 and cleaning the dirt by

C0, all cases are in the storehouse; transporters are

in the waiting area and the cleaner remains where he

finished his last action. The environment looks like

Figure 3.d.

Figure 3.a: Environment.

Figure 3.b: Square and Rectangle detection and moving.

Figure 3.c: Case and dirt detection.

Figure 3.d: Final simulation.

Agents’ windows show that the system, as

designed, provided all necessary information in

communications between coordinator, supervisors,

transporters and cleaners. Objects from ontology

were successfully exchanged and all agents correctly

finished their goals, as seen in the coordinator

window (Figure 4).

Figure 4: Coordinator window.

The action is not executed and it is logged as

failure, if some information differs for some agents

(especially the dimensions of objects). This implies

COMMUNICATION AT ONTOLOGICAL LEVEL IN COOPERATIVE MOBILE ROBOTS SYSTEM

459

that new agents with better capabilities must be

introduced to accomplish the goals.

One other environment situation (Figure 5.a.) has

the transporter T2 with larger capacity and an

intruder that will be detected and after that the

coordinator will start the alarm (Figure 5.b).

Figure 5.a: New agent and objects.

Figure 5.b: Intruder detection and alarm.

The dirt is not removed because the cleaner agent

C0 is filled with the previous dirt. The new big case

is successfully processed by T2.

The coordinator performs logging of all events

and situations arising, for later consultation. For

every action we record the time elapsed for its

execution, the object that was the goal of the action

together with its characteristics, the result of action,

and which agent executed the action. When an

action is not finished or it was instantaneously

accomplished, in the log file will be record only the

time of object registration.

5 CONCLUSIONS

The design of multiagent system is valid with

respect to the purposes and requests of mobile

multirobot system. All agents are able to accomplish

successfully the missions and actions assigned to

them. The system is robust and flexible.

Using the ontological information in inter-agent

communications simplifies the communication

process. Conversations made with ontological

information are oriented to describe the spatial

representation by concepts of the environment,

specific to robotic systems.

The system allows efficient changes in the

content of the messages among robots and proves

that heterogeneous agents, having dissimilar

knowledge can, by using ontology information,

exchange necessary information to accomplish

complex goals.

The system has been tested under simulated

conditions. The positive results are leading to our

next goal, to make tests under real environments,

with mobile robots.

We will also test our approach in other types of

missions for mobile multirobot systems.

REFERENCES

AgentTool, 2005.

http://macr.cis.ksu.edu/projects/agentTool/

Arai, T., Pagello, E., and Parker, L.E., 2002. Advances in

Multi-Robot Systems. In IEEE Transactions on

Robotics and Automation Magazine. Vol. 18, No. 5,

pp. 655-661.

DeLoach, S.A., Wood, M.F., and Sparkman C.H., 2001.

Multiagent Systems Engineering. In: The International

Journal of Software Engineering and Knowledge

Engineering. Vol. 11, No. 3, pp. 231-258.

DiLeo, J., Jacob, T., and DeLoach, S., 2002. Integrating

Ontologies into Multiagent Systems Engineering. In

AOIS-2002 Proceedings of Fourth International Bi-

Conference Workshop an Agent-Oriented Information

Systems. Bologna, Italy.

IntelliJ IDEA, 2006. http://www.jetbrains.com/idea

Noy, N.F., McGuinness, D.L., 2001. Ontology

Development 101: A Guide to Creating Your First

Ontology. Stanford Knowledge Systems Laboratory

Technical Report KSL-01-05 and Stanford Medical

Informatics Technical Report SMI-2001-0880.

Russell, S.J., and Norvig, P., 2002. Artificial Intelligence:

A Modern Approach, Prentice Hall, USA, 2

edition.

SUMO, 2006.

http://protege.stanford.edu/ontologies/sumoOntology/s

umo_ontology.html.

Văcariu, L., Blasko, P.C., Leţia, I.A., Fodor, G. and Creţ,

O.A., 2004. A Multiagent Cooperative Mobile

Robotics Approach for Search and Rescue Missions.

In IAV2004 Proceedings of the 5

IFAC/EURON

Symposium on Intelligent Autonomous Vehicles.

Lisbon, Portugal.

ICINCO 2007 - International Conference on Informatics in Control, Automation and Robotics

460