A MULTI-AGENT COLLABORATIVE CONTROL ARCHITECTURE

WITH FUZZY ADJUSTMENT FOR A MOBILE ROBOT

Bianca Innocenti, Beatriz L

opez, Joaquim Salvi

Institute of Informatics and Applications

University of Girona - Spain

Keywords:

Robot design, control architecture, mobile robots.

Abstract:

One of the current challenges of control research is to make systems capable of showing intelligent responses

to changing circumstances. To address this task, more complex systems are being developed. However, it

is technologically difﬁcult and potentially dangerous to build complex systems that are controlled in a com-

pletely centralized way. One approach to building decentralization systems is using multi-agent technology for

building control architectures. But it seems risky to recursively extend using multi-agent systems to develop

part of the system, such as a single behaviour.One alternative approach is to use collaborative control to deploy

speciﬁc (low level) behaviours, so that several controllers are combined in a single agent of the multi-agent

architecture in order to achieve the wanted behaviour. This paper presents a collaborative controller applied to

the goto behaviour. The experiments were carried out using a Pioneer mobile robot.

1 INTRODUCTION

One of the current challenges of control is to make

systems capable of showing highly ﬂexible and intel-

ligent responses to changing circumstances. Artiﬁcial

Intelligence provides learning and adaptation meth-

ods, as well as decision making techniques to achieve

these control properties. However, it is technologi-

cally difﬁcult and potentially dangerous to build com-

plex systems that are controlled in a completely cen-

tralized way (Murray et al., 2003).

In this line, (Rosenblatt, 1997) built an architecture

composed of distributed, independent, asynchronous

decision-making behaviours that were coordinated by

a central arbiter. The overall behaviour of the system

is rational, coherent and goal-oriented while preserv-

ing real-time responsiveness to its immediate physical

environment. In (Bryson, 2001) several architectures

are analyzed. The advantages of this architecture are

the following: they facilitate their development and

lead to the evolutionary creation of robust systems

of incrementally greater capabilities. In these archi-

tectures, each behaviour is implemented by a module

with communication abilities.

Recent advances in Multi-Agent Systems (MAS)

have inspired researchers to go one abstract level fur-

ther in implementing the architectures, in which mo-

dules are replaced by agents.

Nevertheless, behaviours considered in these archi-

tectures are not simple. For example, a goto behavi-

our, in a free-obstacle path, should take into account

if the target point is close or far away from the current

robot position.

In order to tackle the design of each behaviours,

two approaches can be followed. On one hand, each

behaviour can be implemented as a MAS again, as-

suming the risk that the robot could not be reactive

enough to avoid obstacles when it moves too fast or

the obstacles are mobile. And on the other hand, it is

possible to take advantage of collaborative control to

make up a single behaviour, combining several con-

trollers in a single agent.

Integrating both research lines (the multi-agent ap-

proach and collaborative control) we get as a result a

MAS architecture with collaborative controllers. Col-

laborative control is applied to design and develop

a single behaviour, while the overall robot architec-

ture is based on a MAS where each agent represents

each behaviour. In this paper we give a detailed ex-

planation of how the collaborative control approach

based on Fuzzy Logic (Klir and Folger, 1992) is im-

plemented in a single agent. Details on the MAS ap-

proach can be found in (Innocenti et al., 2006).

This paper is organized as follows. In Section 2,

523

Innocenti B., López B. and Salvi J. (2006).

A MULTI-AGENT COLLABORATIVE CONTROL ARCHITECTURE WITH FUZZY ADJUSTMENT FOR A MOBILE ROBOT.

In Proceedings of the Third International Conference on Informatics in Control, Automation and Robotics, pages 523-526

DOI: 10.5220/0001214005230526

 SciTePress

the related work is presented. Then, in Section 3 the

MAS architecture is described while the collaborative

control proposal is given in Section 4. In Section 5

the results are shown. Finally, some conclusions and

future work are drawn in Section 6.

2 RELATED WORK

As stated above, our approach concerns MAS and col-

laborative control. Collaborative control has a general

meaning, so, each time an algorithm to control a com-

plex task is deﬁned, the idea of collaboration in con-

trol is introduced. Thus, any development of complex

systems with MAS can be considered as a collabora-

tive approach.

On one hand, there are several architectures built as

multi-agent systems to control a single robot, as for

example (Neves and Oliveira, 1997), (Busquets et al.,

2003), (Giorgini et al., 2002) or (Ros et al., 2005).

Most of them present a centralized behaviour coordi-

nation and each behaviour has only one controller to

set the desired outputs.

On the other hand, there are some works related to

collaborative control in robots ((Goldberg and Chen,

2001), (Figueras et al., 2002), (Gerkey et al., 2002)).

In the latter, collaborative control is obtained by re-

lying on the physical dynamics of the robot’s actua-

tor. Due to their nature, motors temporally average

their inputs, so Gerkey and colleagues propose that

a population of non-communicating controllers drive

the robot by interleaving commands to them. The re-

sultant robot motion is then achieved as a superposi-

tion of the different control signals.

In accordance with (Safﬁotti, 1997), our hypothesis

is that, instead of superposing measures, higher deci-

sion making procedures can be used to coordinate the

different controllers. Particularly, we propose using

Fuzzy Logic to model the control actions provided

by heterogeneous controllers and to decide, accord-

ing to the robot motion dynamics, which combination

of control actions have to be executed at a given time.

In addition, the output of the collaborative con-

troller is the output of a single behaviour, instead of

being directly connected to the robot actuators. The

output of the single behaviour (agent) is coordinated

in a multi-agent architecture to decide the next robot

action.

3 MAS ARCHITECTURE

In our MAS architecture, agents can be grouped

into perception, behavioural, actuator and deliberative

agents. Perception agents obtain information about

the environment and about the internal conditions of

the robot; behavioural agents carry out speciﬁc ac-

tions, such as avoiding obstacles; deliberative agents

implement high-level tasks such as planning; and ac-

tuator agents are in charge of controlling the linear

and angular speed of the robot interacting directly

with motors.

The goto agent is a behavioural agent which is re-

sponsible for driving the robot to the target position

at different speeds. Other behavioural agents, are the

avoid agent, responsible for avoiding obstacles and

the goThrough agent, that is in charge of driving the

robot through narrow places like doors. All the agents

coordinate their behaviours by means of a distributed

protocol in order to assure that no conﬂicting actions

are sent to the robot motors (see (Innocenti et al.,

2006) for a detailed explanation).

4 COLLABORATIVE CONTROL

In this section we present our collaborative control

method based on combining multiple controllers us-

ing Fuzzy Logic to implement the behaviour of the

goto agent.

Instead of developing only one quite elaborated

controller, we design several controllers to cope with

different control aspects separately and join their ac-

tions in order to obtain the complex behaviour of the

goto agent.

Our starting point is the aggregation function pro-

posed in (Gerkey et al., 2002), that we extend by

adding weights corresponding to the relevance of each

controller according to the current context. Therefore,

the desired speed calculation is deﬁned as:

Ω =

i=1

· w

t=1

(1)

where η

is the requested wheel speed over time, Ω

the ﬁnal wheel speed, n the number of controllers and

the weights that satisfy

t=1

= 1. By using

weights, it is possible, to give more or less importance

to the controllers.

In order to determine the weights, we propose to

introduce knowledge about the environment, such as

whether the robot is near the destination point or not.

Depending on this information, we can combine two

position controllers: one that is fast and the other that

is accurate. When the robot is far away from the des-

tination point, we can stress the fast controller; when

the robot gets closer to its destination, we can give

more importance to the commands provided by the

accurate one.

Fuzzy terms such as far or close can be modelled

by fuzzy sets, in terms of the distance left (d) to

the destination point. d is deﬁned as d = (d

max

−

rec

)/d

max

where d

max

is the distance between the

ICINCO 2006 - ROBOTICS AND AUTOMATION

524

initial and ﬁnal coordinates, and d

rec

, the distance

from the initial to the current coordinates. Thus, the

fuzzy set close is deﬁned as:

(d) =











1 d ≤ min

(−d+max)

(max−min)

min < d < max

0 d ≥ max











(2)

where min and max parameters have been tuned em-

pirically.

According to this deﬁnition and depending on the

robot’s movement, the distance to the destination

point can be nonlinear along time, making the fuzzy

set also nonlinear.

On the other hand, the fuzzy set far is deﬁned as:

(d) = 1 − µ

(d); (3)

Based on the above fuzzy set deﬁnition, we can re-

solve the fuzzy concurrent control adjustment as:

fast

= µ

(d)

slow

= µ

(d)

(4)

where w

fast

corresponds to the weight of the faster

controller and w

slow

to the accurate one.

Note that the way we use the fuzzy values deter-

mines, according to equation 1, the relevance of the

commands of the different controllers. According to

equation 4, both controllers collaborate in an inter-

mediate situation, that is, in the sloping part of the

fuzzy sets. Therefore, there is no abrupt change in the

control, but progressively one controller has less in-

ﬂuence in the ﬁnal decision while the other one takes

control (fuzzy adjustment).

5 RESULTS

We have implemented the MAS architecture in C++

adhoc MAS platform due to communication con-

straints. The experiment shown in this paper focus on

the implementation of goto agent as a collaborative

controller, so we assure that they were performed in

obstacle-free paths meaning that the goto agent does

not need to coordinate its desired actions with the rest

of the behavioural agents in the MAS architecture.

All the experiments were carried out with our own

model (Innocenti et al., 2004) of the commercial robot

Pioneer 2DX of ActivMedia Robotics.

In order to design the goto agent, we choose to im-

plement two different controllers and mix both con-

trol vectors by means of fuzzy weights. One position

controller is very fast but does not arrive exactly to

the set-point while the other is accurate and reaches

the desired input. The output of the controllers are

the desired linear and angular speeds. The best accu-

rate controller achieved has a settling time of at least

twice the time taken by the fast controller.

The control loop of the proposed concurrent con-

trol is shown in Fig. 1. The fuzzy concurrent control

adjustment block is in charge of mixing the desired

speeds of the controllers in order to change progres-

sively from one controller to the other.

Figure 1: Block diagram of collaborative control loop.

Fig. 2 shows the response of the whole system using

the collaborative control (1)) for the initial position

and heading of (x

, y

, θ

) = (0, 0, 0) and the desired

set-point of (x

, y

, θ

) = (−1, 5, 0).

0 5 10 15 20 25

−1

−0.5

0.5

x (m)

Time (s)

0 5 10 15 20 25

y (m)

Time (s)

0 5 10 15 20 25

−0.2

0.2

0.4

0.6

theta (rad)

Time (s)

−1 −0.5 0 0.5

y (m)

x (m)

0 5 10 15 20 25

−200

−100

100

200

Linear speed (m/s)

Time(s)

0 5 10 15 20 25

−10

Angular speed (rad/s)

Time (s)

Figure 2: Response of the fuzzy concurrent control.

One interesting feature of this concurrent controller

is that it works better than the controllers separately,

especially for the unreachable states produced by the

non-linear nature of the robot model. This behaviour

can be seen in Fig. 3, where the graphics represents

the response for the accurate controller, the fast one,

and the collaborative approach respectively.

A MULTI-AGENT COLLABORATIVE CONTROL ARCHITECTURE WITH FUZZY ADJUSTMENT FOR A MOBILE

ROBOT

525

−2.5 −2 −1.5 −1 −0.5 0 0.5

−1

y (m)

x (m)

0 5 10 15 20 25

−2

−1

theta (rad)

Time (s)

−3 −2 −1 0 1

−1

y (m)

x (m)

0 5 10 15 20 25

−2

−1

theta (rad)

Time (s)

−2 −1.5 −1 −0.5 0 0.5

−1

y (m)

x (m)

0 5 10 15 20 25

−2

−1

theta (rad)

Time (s)

Figure 3: Comparison of the response of the controllers; a)

the slow controller, b) the fast controller and c) the concur-

rent control.

6 CONCLUSIONS

In this paper we present a robot collaborative control

architecture based on integrating recent advances in

multi-agent systems and collaborative control. In par-

ticular, we focus on the design of a single agent, the

goto agent, based on a fuzzy adjustment of two po-

sition controllers. This approach tries to introduce

higher knowledge into the decision making process

of the control system. We propose modelling the rel-

evance of the controllers as a fuzzy set, considering

the distance travelled by the robot.

To test our method, we have designed the goto

agent of the MAS architecture with the proposed

collaborative controller. We have performed several

experiments to evaluate the responsiveness and efﬁ-

ciency of our architecture. With the fuzzy collabora-

tive control, in which both controllers are combined,

the response of the controlled system for several set-

points is faster than the response of the accurate con-

troller and more accurate than the response produced

by the faster controller. Furthermore, it works for

some of the unreachable set-points of the previous ex-

periments (isolated controllers).

As further work, we are planning to extend our ap-

proach to n controllers. In addition, we are also ex-

ploring the extension of the collaborative control to

other agents, such as the goThrough agent, which

is responsible for driving the robot through narrow

spaces, such as corridors or doors.

ACKNOWLEDGEMENTS

This work has been partially supported by the Spanish

MEC Project TIN2004-06354-C02-02 and DURSI-

AGAUR 00296SGR.

REFERENCES

Bryson, J. (2001). Intelligence by Design: Principles of

Modularity and Coordination for Engineering Com-

plex Adaptive Agents. PhD thesis, Massachusetts In-

stitute of Technology.

Busquets, D., Sierra, C., and L

opez de M

antaras, R. (2003).

A multiagent approach to qualitative landmark-based

navigation. Autonomous Robots, 15:129 – 154.

Figueras, A., Colomer, J., and De la Rosa, J. (2002). Super-

vision of heterogeneous controllers for a mobile robot.

In The XV World Congress IFAC.

Gerkey, B., Mataric, M., and Sukhatme, G. (2002). Ex-

ploiting pphysical dynamics for concurrent control of

a mobile robot. Proceedings ICRA ’02. IEEE In-

ternational Conference on Robotics and Automation,

4:3467 – 3472.

Giorgini, P., Kolp, M., and Mylopoulos, J. (2002).

Socio-intentional architectures for multi-agent sys-

tems: The mobile robot control case. Proceedings

of the Fourth International Bi-Conference Workshop

on Agent-Oriented Information Systems (AOIS-02) at

CAiSE2002, Toronto, Canada.

Goldberg, K. and Chen, B. (2001). Collaborative control of

robot motion: robustness to error. In Proceedings of

the 2001 IEEE/RSJ International Conference on Intel-

ligent Robots and Systems, pages 655–660.

Innocenti, B., L

opez, B., and Salvi, J. (2006). How MAS

support distributed robot control. International Sym-

posium of Robotics (ISR).

Innocenti, B., Ridao, P., Gascons, N., El-Fakdi, A., L

opez,

B., and Salvi, J. (2004). Dynamical model param-

eters identiﬁcation of a wheleed mobile robot. 5th

IFAC/EURON Symposium on Intelligent Autonomous

Vehicles (preprints).

Klir, G. J. and Folger, T. A. (1992). Fuzzy Sets, Uncertainty,

and Information. Prentice Hall.

Murray, R.,

Astr

om, K., Boyd, S., Brockett, R., and

Stein, G. (2003). Future directions in control in an

information-rich world. IEEE Control Systems Maga-

zine, 23, issue 2:20 – 33.

Neves, M. C. and Oliveira, E. (1997). A multi-agent ap-

proach for a mobile robot control system. Proceed-

ings of Workshop on ”Multi-Agent Systems:Theory

and Applications” (MASTA’97 - EPPIA’97) - Coim-

bra -Portugal, pages 1 – 14.

Ros, R., de Mantaras, R. L., Sierra, C., and Arcos, J. L.

(2005). A cbr system for autonomous robot naviga-

tion. Frontiers in Artiﬁcial Intelligence and Applica-

tions 131, IOS Press, pages 299–306.

Rosenblatt, J. K. (1997). DAMN: A Distributed Architecture

for Mobile Navigation. PhD thesis, Robotics Institute

at Carnegie Mellon University.

Safﬁotti, A. (1997). The uses of fuzzy logic in autonomous

robot navigation. Soft Computing, 1(4):180 – 197.

ICINCO 2006 - ROBOTICS AND AUTOMATION

526