Developing a Multi-Agent Fuzzy-based Control Architecture

for Autonomous Mobile Manipulators

Mohamed Ayoub Messous

, Abdelfetah Hentout

, Saliha Oukid

and Brahim Bouzouia

Division of Computer-Integrated Manufacturing and Robotics (DPR),

Centre for Development of Advanced Technologies (CDTA), BP 17, Baba Hassen, Algiers 16303, Algeria

Research Laboratory for the Development of Computerized Systems (LRDSI), University Saad Dahleb of Blida (USDB),

BP 270, Blida 09000, Algeria

Keywords: Multi-Agent Systems, Fuzzy Logic, Mobile Manipulators, Control Architectures.

Abstract: Controlling a robotic system, while reaching a certain degree of autonomy and complexity, is carried by the

establishment of its control architecture. The control process is intended through achieving general goals

and/or reacting to changes of the environment. An autonomous robot is required to meet some design

specifications and behavior requirements: its reactivity to environment change, its reliability and its fault-

tolerance, etc. However, A control architecture of a robot must ensure that the robot will achieve, in real-

time, its tasks despite all the constraints. The control is required to be reactively fast but also thorough,

while maintaining some properties such as stability and robustness. The main objective we are intending to

achieve is to design our own approach for autonomous control of mobile manipulators. The expected

approach is meant to be thorough and generic as possible. It should offer a real-time reactive response,

while maintaining a fault-tolerance capabilities and a robust control.

1 STAGE OF THE RESEARCH

Throughout the literature, there has been a

substantial effort devoted to autonomous control of

mobile manipulators. The resulted works brought up

different control approaches based on different

techniques and practices.

As a part of a research project intended for the

control and supervision of mobile manipulators for

the telerobotics in a constrained environment, we

are working on the development of a novel generic

architecture for autonomous control of such robots.

The suggested approach is based on Fuzzy-Logic

Reasoning for the cognitive part. The

implementation is done through a Multi-agent

System scheme.

2 OUTLINE OF OBJECTIVES

Controlling a robotic system (mobile, manipulator or

mobile manipulators), while reaching a certain

degree of autonomy and complexity, is carried by

the establishment of its control architecture. The

control process is intended through achieving

general goals and/or reacting to the changes of the

environment.

An autonomous robot is required to meet some

design specifications and behavior requirements

such as reactivity to environment changes, reliability

and fault-tolerance, etc. However, a control

architecture of a robot must ensure that the robot

will achieve, in real-time, its tasks despite all the

internal and external constraints. The control is

required to be reactively fast but also thorough,

while maintaining some properties (stability,

robustness, etc.).

The main objective we are intending to achieve

is to design our own approach for autonomous

control of mobile manipulators. The expected

approach is meant to be thorough and generic as

possible. It should offer a real-time reactive

response, while maintaining a fault-tolerance

capabilities and a robust control.

3 RESEARCH PROBLEM

A mobile manipulator is constituted by uniting two

disjointed mechanical sub-systems, the mobile base

Messous M., Hentout A., Oukid S. and Bouzouia B. (2013).

Developing a Multi-Agent Fuzzy-based Control Architecture for Autonomous Mobile Manipulators.

In Doctoral Consortium, pages 15-22

DOI: 10.5220/0004638000150022

 SciTePress

and the manipulator. Each one has its own features

and offers different capabilities. The coupling of

these two heterogeneous parts has the benefit of

combining the mobile base locomotion with the

manipulation capabilities of the manipulator. The

resulted system will have an enlarged accessible

workspace and numerous new features. However,

this alliance increases the complexity of the control

process and path planning, especially for tasks that

require an imposed trajectory for the end-effector of

the robot. In such a case, infinity of configurations

for both mobile base and manipulator are possible.

4 STATE OF THE ART

A control architecture design should meet defined

requirements. The existing research experience, so

far, seems to have not figured out a definitive

paradigm for the distribution and/or coordination of

the functionalities required for all the autonomous

robots (Medeiros, 1998).

In the literature, numerous studies have focused

on proposing control architectures for mobile

manipulators. Each proposition gives a specific way

to solve the control problem. Some approaches

provide a total decoupling between the different sub-

systems constituting a mobile manipulator. While

performing a task that requires the involvement of

the two sub-systems, the control process is carried

out sequentially among the two disjoined entities.

There are, also, models for synchronizing the control

of the mobile base and the manipulator.

To achieve the operations and the tasks that the

robot must perform, each designer uses his own

approach. This latter involves the deployment and

the structuration of the internal functions to achieve

the final objectives assigned by a higher-level

operator.

In the following section, we present a review of

the main control approaches for autonomous robots,

discuss their major properties and propose a

classification into two different classes depending on

the techniques used for controlling the robot.

4.1 Classical Control Approaches

Considered as classical, this first class is based on

the study of mathematical models for both

mechanical sub-systems (manipulator and mobile

base). Controlling a mobile manipulator consists of

computing the motion of the manipulator joints and

that of the mobile base. For this aim, the study of

both Direct and Inverse Kinematic Models of the

robot is needed.

Using the mathematical models to control mobile

manipulators produces good and accurate results,

and offers a fairly exact control for repetitive tasks

in controlled and known environments (industrial

robotics for example). In this case, when the robot is

required to repeat a trajectory thousands of times,

very complicated computation of these models is

done, in most cases, off-line with the ability to

optimize path covered, energy consumption, time

spent, etc. while performing a task. However, some

robotic environments using mobile manipulators are

not completely known but, in contrast, evolutionary.

Which could result to a very poor performances in

real-time control, due to the computational resources

needed to come out with a response in a limited time

period. Besides, it must be noted that classical

approaches have the disadvantage of computing time

which is quite important depending on the high

number of degrees-of-freedom of the robot,

especially in frequently-changed and not completely

known environments. According to Brooks (Brooks,

1986), traditional robotics seems unable to deliver

real-time performances in a dynamic world.

Moreover, it should be noted that the methods

used for computing the Direct Kinematic Models

represented generic rules, whereas the computation

of the Inverse Kinematic Models were constructed

according to the mechanical structure of the robot.

Finally, these models don’t tolerate any change in

the mechanical structure of the robot (malfunction of

one or more joints of the manipulator for example)

without adding specific modes for failures treatment.

When a fault occurs, the possibility of offering a

minimum service until repairing the breakdown is an

important element of the Quality of Service

(Delarue, 2007).

4.2 Multi-Agent Fuzzy-based Control

Approaches

When working with complex problems with large

dimensions, the resolution of control problems for

mobile manipulators is very difficult using

traditional mathematical models. Several approaches

have described the process allowing the end-effector

of such a robot to reach a Cartesian position in its

workspace, without using the Inverse Kinematic

Model or differential-equation solvers. Some of

these works can be found in (Duhaut, 1999); (Erden,

2004); (Colle, 2006), etc.

ICINCO2013-DoctoralConsortium

4.2.1 General Principle

Multi-agent and Distributed Artificial Intelligence

Techniques offer simple solutions (Colle, 2006).

Each joint is implemented as an agent. Every agent

tries to align the position of the end-effector of the

robot with that of the target, while being

independent of the motions and positions of the

other joints, and with no prior knowledge of the

actions of the other agents. By acting recursively,

the other agents controlling the other joints try to do

the same job. A global behavior can emerge,

consequently, from all the local agents which will

satisfy the desired objective (reaching the desired

position).

In contrast to classical control approaches, Multi-

agent Approaches offer methods that use the agent

paradigm by proposing a decomposition of the robot

control into a set of distinct agents. The Multi-agent

Approaches benefit of all the advantages of

distributed problem solving. The Multi-agent System

perspective made it possible to consider the

architecture as a compound of simpler modules,

which gave way to easier design of the whole

system. In addition, the need for massy

mathematical models, Inverse Kinematic Models and

differential-equation-solutions is overcome.

Therefore, there is a considerable decrease in design

effort and computation time compared to classical

approaches. Moreover, with such a usage of Multi-

agent Systems, the control architecture is more

flexible to be applied to any robot (Erden, 2004).

To solve a complex problem, a Multi-Agent

System can emerge a global behavior using several

agents. Each of these latter has a confined

knowledge and limited actions ability. Fuzzy Logic,

Neural Networks and/or Genetic Algorithms

associated with Multi-agent Systems, can provide

high-level control for complex systems

(Tournassoud, 1992); (Guessoum, 1997).

4.2.2 Fuzzy-Logic Control

Fuzzy Logic is a mathematical formulation that

copes with uncertainty in information (Klir and

Folger, 1992). Fuzzy control has proven to be a

successful control approach to many complex non-

linear systems or even non-analytic ones. It has been

suggested as an alternative approach to conventional

(classical) control techniques in many situations

(Precup and Hellendoornr, 2011). Such a control is

characterized by the use of linguistic rules to

manipulate and implement human knowledge in

control systems so as to handle the uncertainty

present in the environment (Passino and Yurkovich,

1998).

4.2.3 Advantages of Multi-Agent

Heuristic-based Control Approaches

Multi-agent Heuristic-based Control Approaches

don’t require the precise mathematical model of the

robot to be controlled. However, if the model exists,

it can be used for the simulation and for the test of

the control strategy. The main advantages of such

controllers are given as follows (Godjevac, 1995);

(Singh, 2008):

 No need to have a mathematical model of the

robot (data is either unavailable, incomplete or

the process is too complex).

 It is possible to implement expert-human

knowledge and experience using comprehensible

linguistic rules.

 Thanks to dedicated processors, it is possible to

control fast processes.

 Such techniques allow building robust and

smooth controllers starting from heuristic

knowledge and qualitative models.

 These approaches allow considering imprecise,

vague and unreliable information; and

integrating symbolic reasoning and numeric

processing in the same framework.

Throughout our literature study, the objective of all

the reviewed works we referred to was to test the

performances of the multi-agent heuristic-based

design of control approaches in simulation.

Moreover, all the studies were applied on a simple

case of a service mobile manipulator undertaking a

specific task in two-dimension workspace.

Unfortunately, to our knowledge, no works were

done on real physical robots in three-dimension

space.

5 METHODOLOGY

The design phase of a control architecture is the

most strategic one for the development of a

controller. However, it is necessary to understand

and to identify the needs to design and implement

the approach in a proper manner. With the

increasing complexity of architectures, using a

development methodology is very necessary.

Nevertheless, the absence of such a methodology

covering the whole life cycle of a Multi-agent

System makes the task very difficult and complicated

(Hentout, 2008). The Multi-agent Approach poses,

DevelopingaMulti-AgentFuzzy-basedControlArchitectureforAutonomousMobileManipulators

Figure 1: Structure of the Multi-Agent System package.

moreover, the problem of the management and

control of agents and their shared resources. A

constraint that should not be overlooked when

designing such a control system is the lack of

information. This lack is mainly due to the

measurement errors delivered by the physical

sensors of the robot.

The combined use of Fuzzy-Logic Reasoning

with Multi-agent Approach has the benefit of

offering a Distributed Fuzzy Control System with

smaller fuzzy sub-systems instead of one big

centralized fuzzy system. Therefore, adopting an

architecture/approach is a technological problem for

the designer, who is supposed to consider the

required degrees of reactivity, intelligent behavior

and the related implementation cost. The type of the

robotic system, its technical specifications and the

required level of autonomy are the most important

constraints while designing a novel approach.

Typically, the differences are the ease with which

approaches can be developed, and the efficiency

with which tasks can be achieved.

The control architecture we are developing is

meant to be as generic as possible. It is intended to

fulfill a list of requirements and specifications

related to autonomous control of mobile

manipulators. Some of which are robustness and

fault-tolerance, programmability, extensibility and

scalability:

Robustness is proven through the ability of the

fuzzy-logic reasoning to handle imperfect inputs.

 Fault tolerance is guaranteed with the aptitude,

of the used Multi-agent System, to handle

unexpected events and sudden malfunctions.

These latter will be countered via the possibility

of offering a minimum service functionalities in

case of one or more joints of the manipulator

break down.

 Thanks to a scalable structure and a modular

design, new features can be developed and added

progressively without modifying the existing

ones.

5.1 Presentation of the Control

Architecture

Our work is dedicated to develop a novel control

architecture for autonomous mobile manipulators

using a Multi-agent fuzzy-based control approach.

This architecture is responsible for controlling the

robot while sharing the control of the heterogeneous

sub-systems.

The implementation of the control architecture is

being done by using Java Netbeans 7.0.

Furthermore, for the implementation of the fuzzy-

based part, we are using the open source library for

fuzzy systems jFuzzyLogic developed by (Cingolani

and Alcala-Fdez, 2012) and available from

jfuzzylogic.sourceforge.net.

Choosing the suggested model is justified by the

generic nature of the proposed agent models and by

the possibility of integrating the whole in a modular,

more general, robotic system.

We refer to each agent as a separate Thread. In

addition, we propose the use of two distinguished

types of agents as showed in Figure 1 (i)

SystemAgents and (ii) ControlAgents. Both of which

has a distinctive purpose. More details are given in

the following sub-sections.

5.1.1 System Agents

This first type of agents is more complex than the

ControlAgent

ICINCO2013-DoctoralConsortium

second one. The agents of this type inherit from the

SystemAgent class. They all belong to the same

package, SystemAgents, and have the same outline

structure. They are meant for the processing and

tuning of data issued from the different sensors

equipping the robot.

Such agents may comprise several features

regarding the control process. The modules we

propose are (i) Module of Vision, (ii) Module of

Localization of the robot in its environment and (iii)

Module of Localization of the targets. However, this

list of modules is not thorough. Other modules may

be proposed and integrated progressively into the

control architecture.

5.1.2 Control Agents

Inheriting from the ControlAgent class, the agents of

this type are regrouped in the RobotAgents package.

As its name suggests, a ControlAgent is dedicated to

the control process itself. Each agent is assigned to

control one mechanical sub-system, which can be

either the mobile base or one of the joints

constituting the manipulator.

Figure 2: Composition of the ControlAgents.

The set of control agents are brought together to

reach a high-level goal. An emergent behavior is

arisen (using reactive agents with poor own world

knowledge) which overall can solve a higher-level

problem than each agent, by itself, can achieve. The

composition of the ControlAgents is given in Figure

Through our study, we are more interested by

this type of agents. Such an agent produces the

control instructions directed to its respective

controlled sub-system (mobile base or manipulator).

The cognition part is ensured by using fuzzy-based

modules, called Fuzzy Controllers.

A fuzzy controller is the entity offering the

intelligence of the system. It is based upon a Fuzzy

Inference System, which compute the entries of a

fuzzified values of linguistic variables through a set

of fuzzy-logic rules. A fuzzy controller produces, as

an output, the deffuzified values for the

corresponding velocity (rotation and translation

velocity of the mobile base, angular velocities of the

joints of the manipulator).

A ControlAgent is designed using one or more

fuzzy controllers.

5.1.2.1 BaseAgent

Figure 3 shows a synoptic structure for the

BaseAgent where:

 (X

, Y

, Z

): it represents the operational

coordinates of the target to be reached by the

end-effector of the robot.

 (X

, Y

, Z

): it is the current position of the end-

effector of the robot.

 (Dist

, …, Dist

): they represent the distances

separating the mobile base to obstacles (1, …, n)

present in its environment.

 E_angle: it represents the computed angular error

between the current position of the end-effector

and the desired target to be reached.

 E_distance: it is the distance error between the

current position of the end-effector and the final

target.

BaseAgent is composed of two fuzzy controllers (i)

RotationFC and (ii) TranslationFC. They control,

respectively, the rotation and translation velocities of

the mobile base.

The intended scheme is considered as a generic

structure of a higher-level abstraction, which would

be applicable to most mobile base systems.

However, a lower-level middle-layer will be

implemented for each specific mechanical structure

of a mobile base. It should be declared that obstacles

avoidance capabilities will be implemented and

fitted into the BaseAgent.

5.1.2.2 JointAgent

JointAgent is a reactive agent dedicated for

controlling one of the manipulator joints. Each

JointAgent is composed of one fuzzy controller that

assures the velocity control output of the

corresponding joint. Figure 4 gives a synoptic

ControlAgent

DevelopingaMulti-AgentFuzzy-basedControlArchitectureforAutonomousMobileManipulators

Figure 3: A synoptic scheme for the internal structure of the BaseAgent.

Figure 4: Inner synoptic structure for the JointAgent.

scheme for this kind of ControlAgents where:

 (Joint

, …, Joint

): they are the other joints of the

manipulator related to the current controlled

joint. They are defined as the joints that affect

the position of the end-effector in a mutual way,

i.e., the movement of one of the related joints

influences, more or less, the traveled path of the

end-effector in the same way.

5.1.3 Fuzzy Controllers

The two following figures (Figure 5 and 6) illustrate

some examples of the used Fuzzy Membership

Functions of the different controllers of the

BaseAgent:

 In figure 5, the two first Fuzzy Membership

Functions represent the entries of the

RotationFuzzyController. They are respectively

(i) E_angle and (ii) E_distance. The third

function is the deffuzified output which

represents the Rotation Velocity value of the

BaseAgent.

Figure 5: Fuzzy membership functions of the

RotationFuzzyController for the BaseAgent.

ICINCO2013-DoctoralConsortium

RULE 1: IF E_angle IS left_big AND E_distance IS far THEN

RotationVelocity IS fast_left;

RULE 2: IF E_angle IS left_big AND E_distance IS near THEN

RotationVelocity IS slow_left;

RULE 3: IF E_angle IS right_big AND E_distance IS far THEN

RotationVelocity IS fast_right;

RULE 4: IF E_angle IS right_big AND E_distance IS near THEN

RotationVelocity IS slow_right;

RULE 5: IF E_angle IS zero THEN

RotationVelocity IS null;

RULE 6: IF E_distance IS very_near THEN

RotationVelocity IS null;

RULE 7: IF E_angle IS left_small AND E_distance IS far THEN

RotationVelocity IS slow_left;

RULE 8: IF E_angle IS right_small AND E_distance IS far THEN

RotationVelocity IS slow_right;

Figure 8: Rules-Block used for the RotationFuzzyController.

 Likewise, Figure 6 represents the two inputs (i)

E_angle and (ii) E_distance, and the output,

Translation Velocity, for the

TranslationFuzzyController of the BaseAgent.

Figure 6: Fuzzy membership functions of the

TranslationFuzzyController for the BaseAgent.

5.2 Presentation of the Control

Approach

The objective set for the ControlAgent is to move

the end-effector of the robot as close as possible to

the desired goal, while considering that all of the

other ones as stationary.

The Control Agents, in our approach, implement

this simple reactive behavior in a parallel way based

on their local knowledge. They are autonomous and

independent with, as a criterion, a local distance

minimization. A global emerging behavior will arise

satisfying the main objective.

Every single ControlAgent admits as an entry the

computed value of the current position of the end-

effector of the robot computed by using the Direct

Kinematic Model of the manipulator. Each agent,

also, needs the current position of the target in order

to compute E_distance and E_angle. Furthermore,

each JointAgent requires, besides the current

position of the controlled articulation and the instant

positions of all its related joints.

In figure 7, each agent is supposed to know the

position of the end-effector and the influence of its

movement according to the target position in a three-

dimensional workspace.

Figure 7: Four Parallel JointAgent extent (Delarue, 2007).

The position of the end-effector, in our case, is

computed by using the Direct Kinematic Model.

However, our approach will, also, work with other

external modules for computing the instant positions

of the end-effector by using, for example, cameras,

laser measurement sensors, etc.

The cognitive effort throughout the proposed

approach is contained within the fuzzy-logic rules,

which are inserted in the fuzzy inference system of

each fuzzy controller. The following figure (Figure

8) shows a narrow example from the set of rules

used in the RotationFuzzyController.

DevelopingaMulti-AgentFuzzy-basedControlArchitectureforAutonomousMobileManipulators

6 EXPECTED OUTCOME

Based on Artificial Intelligence techniques used for

problem solving, we proposed a novel control

architecture for autonomous mobile manipulators.

The control process is mainly distributed on several

concurrent agents, with independent behaviors,

combining reactive and deliberative capacities. This

class provides an alternative to the use of

mathematical models to control such robots. It offers

results that approximate human behaviors, and

improves tolerance to certain faults and mechanical

failures. Throughout this paper, we have reviewed

some recent research works which proposed

interesting models for the control of autonomous

mobile manipulators.

The future works tends to achieve a thorough

testing for the proposed approach in different

scenarios. In addition, a comparison of the obtained

results should be made with the existent control

architectures.

REFERENCES

Brooks, R. A., 1986. “A robust layered control system for

a mobile robot”, IEEE Journal of Robotics and

Automation RA, 2(1), pp. 14-23.

Cingolani, P., Alcala-Fdez, J., 10-15 June 2012.

“jFuzzyLogic: A Robust and Flexible Fuzzy-Logic

Inference System Language Implementation”, The

IEEE World Congress on Computational Intelligence

(WCCI 2012), Brisbane, Australia, pp. 1090-1097.

Colle, E., Nait-Chabane, K., Delarue, S., Hoppenot, P., 7-9

June 2006. “ARPH: Comparaison d’une méthode

classique et d’une méthode utilisant la coopération

hommemachine pour exploiter la redondance de

l’assistant robotisé”. The 4

Conference

HANDICAP2006, France.

Delarue, S., Hoppenot, Ph., Colle, E., 9-12 May 2007. “A

Multi Agent Controller for a Mobile Arm

Manipulator”. The 4

International Conference on

Informatics in Control, Automation and Robotics

(ICINCO2007), Angers, France.

Duhaut, D., 1999. “Distributed Algorithm For High

Control Robotics Structures”. International

Conference on Artificial Intelligence, Vol. 1, pp 45-50.

Erden, M. S., Leblebicioglu, K., Halici, U. 2004. “Multi-

agent System-Based Fuzzy Controller Design with

Genetic Tuning for a Mobile Manipulator Robot in the

Hand Over Task”, Journal of Intelligent and Robotic

Systems, 39(3), pp. 287-306.

Guessoum, Z., April 1997. “A Hybrid Agent Model: a

Reactive and Cognitive Behavior”. The 3

International Symposium on Autonomous

Decentralized Systems (ISADS’97), Germany, pp. 25-

32.

Godjevac, J., February 1995. “Comparative study of fuzzy

control, neural network control and neuro-fuzzy

control”, Technical report n° 103/95, Federal

Polytechnic School of Lausanne, Computer Science

Department.

Hentout, A., Bouzouia, B., Toukal, Z., 09-12 March 2008.

“Modeling of Agent-based Architecture for Driving

Mobile Manipulator Robots”. The International

Conference on Distributed Human-Machine Systems

(DHMS 2008), Greece.

Klir, G. J., Folger, T. A., 1992. “Fuzzy Sets, Uncertainty,

and Information”, Upper Saddle River, NJ: Prentice-

Hall.

Medeiros, A. A. D., April 1998. “A Survey of Control

Architectures for Autonomous Mobile Robots”,

Journal of the Brazilian Computer Society, 4(3),

Campinas.

Precup, R.-E., Hellendoorn, H., 2011. “A survey on

industrial applications of fuzzy control”, Computers in

Industry, 62(1), pp. 213-226.

Passino, K. M., Yurkovich, S., “Fuzzy Control”, Menlo

Park, CA: Addison-Wesley, 1998.

Singh, M. K., Parhi, D. R., Bhowmik, S., Kashyap, S. K.,

1-6 October 2008. “Intelligent Controller for Mobile

Robot: Fuzzy Logic Approach”, The 12

International

Conference of International Association for Computer

Methods and Advances in Geomechanics (IACMAG),

Goa, India.

Tournassoud, P., September 1992. “Planification et

contrôle en robotique: Application aux robots mobiles

et manipulateurs”, Hermes Sciences Publications.

ICINCO2013-DoctoralConsortium