ON THE USE OF AUTOMATED GUIDED VEHICLES IN

FLEXIBLE MANUFACTURING SYSTEMS

Samia Maza

and Pierre Castagna

Centre de Recherche en Automatique de Nancy, UMR 7039, CNRS-Nancy university

2, avenue de la Forêt de Haye, 54506 Vandoeuvre lès Nancy, France

Institut de Recherche en Communications et Cybernétique de Nantes, UMR CNRS 6597

1, rue de La Noë, BP 92101 Nantes Cedex 3, France

Keywords: Manufacturing systems, automated guided vehicles, conflicts, routing, traffic management.

Abstract: In manufacturing systems, material transport plays a key role for production process efficiency. Because of

their advantages over other material handling systems such as conveyors and robots, AGVs are widely used

in flexible manufacturing systems. The scheduling of several AGVs in a non-conflicting manner is a

complicated problem, especially when the AGV system is bi-directional. In fact, many undesirable

situations may arise such as deadlocks and head-on conflicts if no efficient control policy is used to prevent

them. This paper presents the key issues to be addressed to efficiently employ these devices, and deal

particularly with the traffic management problem.

1 INTRODUCTION

Automatic Guided Vehicle Systems (AGVS) are one

of the most exciting and dynamic areas in material

handling today. But AGVS is really not so new.

Fifty years ago when AGVS was invented it was

then called driverless systems. Through the years,

advances in electronics have led to advances in

guided vehicles. Technological developments may

have given AGVS more flexibility and capability,

but market acceptance has really given AGVS the

application variety to allow it to expand into the

standard accepted material handling method it is

today.

Automated guided vehicles (AGVs) are material

handling devices used to transport products and

goods among the workstations and storage areas of a

manufacturing system. The basic functions of an

AGVS are:

 Navigation and Guidance allow the vehicle to

follow a predetermined route which is

optimized for the material flow pattern of a

given application

 Routing is the vehicle's ability to make

decisions along the guidance path in order to

select optimum routes to specific destinations

 Traffic Management is a system or vehicle

ability to avoid collisions with other vehicles

while at the same time maximizing vehicle

flow and therefore load movement throughout

the system.

 Load Transfer is the pickup and delivery

method for an AGVS system, which may be

simple or integrated with other subsystems.

 System Management is the method of system

control used to dictate system operation.

The goal of this paper is to present some

interesting problems related to the use of AGV

systems and a short overview of papers dealing with

those problems especially the routing problem. We

will present our research works and results

concerning the routing of bi-directional AGV

Systems.

In the second section, the key issues to be

adressed to efficiently use the AGVs are presented.

The approaches we have developed for the conflict

free routing of bi-directional AGVs will be briefly

presented in the third section. The simulation study

is presented in section four and the study we have

made for the compact disc manufacturer is presented

in section five. Section six is devoted to the

conclusions.

476

Maza S. and Castagna P. (2007).

ON THE USE OF AUTOMATED GUIDED VEHICLES IN FLEXIBLE MANUFACTURING SYSTEMS.

In Proceedings of the Fourth International Conference on Informatics in Control, Automation and Robotics, pages 476-483

DOI: 10.5220/0001637004760483

 SciTePress

2 DESIGN AND CONTROL OF

AN AGVS

An AGV system is a set of a cooperative driver-less

vehicles moving on the same manufacturing floor

and coordinated by a control system.

For a successful deployment of an AGVS, the

following key issues should be addressed

(Reveliotis, 2000):

2.1 The Flow Path Design

The manufacturing floor is specified by a set of

physical or virtual guide-paths. If the AGVs are

allowed to move only according to one direction,

they are called unidirectional, otherwise, i.e., if they

can move into the two directions, they are said to be

bi-directional. It has been shown that the bi-

directional AGVS can improve considerably the

performances of a manufacturing system (Egbelu

and Tanchoco, 1986). The design of the guide-path

is an important problem, i.e., the choice of the guide-

path configuration in order to minimise the travelled

distances. Many research works deal with such a

problem, the others deal with the determination of

the guide-path’s lanes direction for unidirectional

AGVSs (for example Gaskins and Tanchoco, 1987).

Such a problem is generally formalised as an integer

programming problem with distance as criterion to

optimise. The most known configurations are the

simple loop, multiple loops, tandem and

conventional configurations.

2.2 The Fleet Size and Vehicle’s

Capacity Determination

The AGVs can be classified according to their load

capacity into two categories: single and multiple

load vehicles. And one of the important problems to

be addressed when designing AGVS is the

determination of the AGVs number and their loading

capacity. The research works which deal with such

problem are numerous and can be classified into

three categories:

 Analytical methods;

 Simulating methods;

 Hybrid or mixed methods.

For example in (Egbelu and Tanchoco, 1987), four

analytical methods are proposed to determine the

minimum fleet size to satisfy the production needs.

These methods were tested for various dispatching

rules. Beamon and Chen (Beamon and Chen, 1998a)

reproach the traditional methods to not consider the

system reliability when determining the optimal fleet

size. To consider the reliability of the AGVs and the

guide-path intersections, they include in their

proposed approach, the vehicles and intersections

failure rates in order to calculate the minimum

number of AGVs needed. Beamon and Deshpande

(Beamon and Deshpande, 1998b) proposed an

approach to jointly optimise the fleet size the

vehicle’s load capacity (i.e., the load batch size).

The objective is to make the better trade-off between

those two criteria. Indeed, more great is the load

size; shorter will be the total travelled distances,

since one vehicle will make only one displacement

with many loads. However, the loading and

unloading time will be greater than for a single load.

In the same manner, when there is a great number of

AGVs, the system performances will be improved

until the optimal number. Beyond this optimal

number, the performances will be degraded since a

great number of vehicles increases the traffic

congestion and deadlocks.

 In (Castagna and Maza, 2004) we proposed a

simulation approach to determine the optimal

fleet size given a production horizon T, the

manufacturing ranges, the production rate, and

the guide-path.

Figure 1: the fleet size calculation procedure.

The basic idea of this approach is:

In the first step, the number of AGVs needed to

realise the specified number of products into the

specified makespan T is calculated.

Start

The production rate;

The manufacturing order;

The production horizon T;

The manufacturing ranges

Calculation of the fleet size N according

to the management policy of idle vehicles

Make a simulation test and evaluate the

real production time T

≤T ?

Considering the deadlocks and T

calculate a new fleet size N to satisfy the

constraint on the productin duration T:

N=N+[T

/T]+1

End

Yes

ON THE USE OF AUTOMATED GUIDED VEHICLES IN FLEXIBLE MANUFACTURING SYSTEMS

477

In the second step, a simulation study is conducted

to test the calculated number; and to refine it by

considering the achieved performances (see figure

1). Indeed, when the fleet size is calculated in the

first step, deadlocks are not considered. The

simulation allows the determination of the real

production time T

needed to accomplish all the

transportation missions. This new makespan is used

to determine the AGVs fleet size once again and is

tested by simulation. The optimal fleet size is

obtained when all the transportation tasks are

effectively achieved in the specified production

horizon T.

This study was made for various management

policies of idle AGVs and was compared to the

analytical approach proposed in (Egbelu and

Tanchoco, 1987).

2.3 The Dispatching Problem

To achieve a product or a job, it has to be routed on

several stations of the manufacturing system to

undergo some transformation operations. These are

those transitions which introduce the problem of

vehicle’s task assignment.

Indeed, when an operation is completed on one

workstation, the product makes a request for a

vehicle to be transported to the next station.

The dispatching problem consists in choosing one

request among several ones in a standby state, and

choosing one vehicle to be affected to that request.

This vehicle should be able to make the resulting

displacement.

The dispatching problem was developed in many

research papers, but their number is still smaller than

the one of the papers dealing with the scheduling

problem without transportation resources. For

example Blazewicz et al (Blazewicz et al, 1991)

propose an approach to search for a scheduling that

jointly considers the jobs and vehicles. Egbelu and

Tanchoco (Egbelu and Tanchoco, 1984) describe the

major vehicles’ dispatching rules for two special

cases: (a) when there is only one transporting

request and many idle vehicles, and (b) when there

is only one idle AGV and many jobs requesting a

vehicle.

Other dispatching cases were considered in (Albert,

1998).

2.4 The AGVs Routing and Traffic

Management Problem

The aim of routing AGVs is to find an optimal (e.g.

shortest possible time path) and feasible route for

every single AGV.

Actually, the routing decision includes three

aspects. Firstly, it should detect whether there exists

a route which could lead the vehicle from its origin

to the destination. Secondly, the route selected for an

AGV must be feasible, i.e., the route must be

congestion-, conflict- and deadlock-free (Taghaboni

and Tanchoco, 1995), etc. Thirdly, the route must be

optimal or at least partially optimal, e.g. minimize

idling runs of vehicles.

Indeed, AGVS are the seat of a great number of

undesirable situations, in particular when they are

bi-directional. Situations like conflicts and

deadlocks. Conflicts occur when for example two

AGVs are attempting to travel one lane at the same

direction but at different speeds, or into opposite

directions. Other conflicts occur when several

vehicles attempt to cross one intersection at the same

time (Figure 2.)

Figure 2: Example of conflicts between AGVs.

Deadlock is a well known problem in the resource

allocation systems and technological areas such as

computer operating systems, transportation and

automated manufacturing systems (Lawley and

Reveliotis, 1999).

A resource allocation system (RAS) consists of a

finite set of resources that must be allocated to

competing processes. The processes enter the

system, request, acquire, use, and release their

required resources, and then exit the system.

Figure 3: Example of deadlock situation in a resource

allocation graph for a RAS of 3 processes and 3 resources,

where R

is of capacity 2.

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

478

Many types of RASs are prone to deadlock, an

insidious halting condition in which there exists a set

of processes with every process in the set awaiting

the allocation of resources held by other processes in

the set (Figure 3).

Well known strategies for handling deadlock are

(1) prevention, (2) detection-resolution, and (3)

avoidance. Prevention restrains the request structure

of processes so that deadlock is impossible. Because

it limits process concurrency, prevention tends to be

overly restrictive and typically achieves poor

resource utilization.

Detection-resolution approaches allow deadlock

to occur and then concentrate on expedient

resolution. This approach achieves the greatest

flexibility in resource allocation at the cost of system

stoppage and resolution procedures, which may

involve aborting processes or the time consuming

transport and reshuffling of physical entities.

Avoidance uses current state information along with

knowledge of process request and release structures

to restrain the way resources are allocated so that

deadlock never occurs. Avoidance achieves a middle

ground in terms of allocation flexibility, being more

flexible than prevention but less flexible than

detection. It does not incur the cost of system

stoppage and resolution and thus is the preferred

method when the incremental increase in allocation

flexibility does not merit the cost of allowing

deadlock to occur. Dijskstra was pioneer in that field

and proposed a polynomial algorithm, known as the

banker algorithm, to resolve a sequential resource

allocation problem. A more complete discussion of

fundamental deadlock concepts can be found in most

books on computer operating systems, for example

see (Silberschatz and Peterson, 1991).

These few last years, many research works were

conducted to avoid deadlocks in automated

manufacturing systems, for example (Pia Fanti,

1997) and (Reveliotis, 1996).

An AGV system can be considered as a resource

allocation system, where the processes are the AGVs

and the resources are intersections and lanes of the

guide path. The problem of deadlock and conflict

free routing in AGVS will be developed in the next

section.

3 CONFLICT FREE ROUTING OF

BI-DIRECTIONAL AGVS

In our research work, we were especially interested

in conflict-free routing of bi-directional AGVS.

Several routing approaches were proposed in the

literature and can be classified into two categories:

(1) Predictive or planning methods: here the

conflicts are predicted off-line and vehicles’

paths are planned to avoid these conflicts and

to minimise a performance criterion see for

example (Krishnamurthy et al, 1993) and

(Oboth et al, 1998).

(2) Reactive or dynamic methods: here, an AGV

path is not planned and routing decisions are

made in a real time manner according to

system’s state. Such methods are always

qualified as zone dynamic control methods,

since the guide path is divided into non

overlapping zones considered as non-

sharable resources; see for example

(Reveliotis, 2000), and (Branislav, 2002).

The advantage of the first category of methods is

that the system’s performances (like the makespan

or travel time of AGVs) are a priori considered and

optimised. However, unlike the reactive methods,

the planning approaches are sensitive to

perturbations since the scheduling is made in a

deterministic way. Dynamic methods are very robust

but do not consider the performances optimisation

beyond a short horizon.

In order to have the advantages of the types of

methods, we proposed a new approach in (Maza ad

Castagna, 2005a, 2005b) based on a planning

method proposed in (Kim and Tanchoco, 1991).

This planning method is based on an algorithm

called cfstp (i.e. conflict free shortest time

procedure), which calculates for each AGV, having

a transporting mission the fastest route, considering

the traffic status, to reach its destination without

conflicts. To this end, intersections and

characteristic point of the guide path are modelled

by square areas, called nodes, which are considered

as non sharable resources. When an AGV moves on

this guide path, it reserves some nodes for a while

(see figure 4). This duration is called reserved time

window (noted r

for node n), where the node is

exclusively reserved by that vehicle. Other time

intervals where nodes are free are called free time

windows (noted f

for node m).

The cfstp calculates the shortest path on a

directed time windows graph, in which the vertices

represent the free time windows and the links model

the reachability between these time windows. The

ability to reach a time window from another one is

established by calling another algorithm called the

reachability test procedure. For two free time

windows f

and f

associated respectively to the

nodes n and m, this last procedure makes the

following reachability tests between them:

ON THE USE OF AUTOMATED GUIDED VEHICLES IN FLEXIBLE MANUFACTURING SYSTEMS

479

(1) Check for space feasibility, i.e., the existence

of a physical link relating m to n.

(2) Check for time feasibility, i.e., the node m is

reachable from the node n within its free

time window f

(3) Check for potential conflicts.

Figure 4: Example of time-windows table.

⎯ r

is the j

reserved time window of node i

⎯ f

is the k

free time window of node l.

When a mission is possible, the cfstp delivers for its

assigned AGV a set of nodes to be visited and the

arrival and exit times to those nodes to avoid

conflicts and minimise the travel time. As said

before, such method is sensitive to perturbations.

There are two types of contingencies: temporary

and permanent. We consider only the first type, such

as a slowing down in front of a fixed or a moving

obstacle, or a temporary stop on a lane or a node to

charge the battery, etc. In that case, the scheduled

arrival and exit times will not be respected and

consequently, there is no security guarantee for the

AGVs since collisions can occur.

To ensure the reliability of an AGVS in the

presence of interruptions while maintaining the

scheduled trajectories, a control architecture was

proposed in (Maza and Castagna, 2001, 2005a).

Figure 5: The AGV control architecture.

(1) the scheduling level delivers for each node i, an

ordered list, O

, of AGVs having to cross it in a growing

order of their arrival dates.

(2) the deadlock avoidance level operates in presence of

contingencies by respecting the predicted node’s crossing

order (RVWA) or by re-ordering the AGVs (RVRAA or

RVDA). It informs the 1

level about the current changes.

Indeed, a second level of real-time control was

added to the AGVs scheduling level which uses the

cfstp, in order to avoid deadlocks and conflicts when

needed (see Figure 5). First, the AGVs are scheduled

on the nodes of the guide-path in a non conflicting

manner while optimising the mission’s duration.

Then, the scheduled entry times to each node are used

to establish for each AGV, its own priority to cross

these nodes.

Three polynomial algorithms were proposed. The

first one based on static priorities, called RVWA (or

Robust Vehicle Waiting Algorithm). RVWA is based

on a theorem that says that if each AGV respects its

node crossing order, the property of non-conflict is

conserved.

The second algorithm based on dynamic

priorities, called RVRAA (or Robust Vehicle Routing

Ahead Algorithm) allows the rescheduling of the

AGVs on some nodes in order to improve the RVWA

which always induces unnecessary waiting of

vehicles to respect their crossing priorities (Maza and

Castagna, 2005a). The basic idea of RVRAA is to give

the AGV V which calls the algorithm the greatest

priority on some specified path [N,M], where N is the

node where V calls RVRAA, and M is the node where

V has the highest priority.

The third algorithm called RVDA (or Robust

Vehicle Delaying Algorithm), also based on dynamic

priorities, penalises the AGV which is late, say U, in

front of some other AGVs on a path [N,M]. N is the

node where the algorithm is called, and M is a node

which is calculated by the algorithm to insure that

the system will never reach an unsafe state (Maza

and Castagna, 2005b). An AGV state is called

unsafe if it can conduct the AGV system to a

deadlock state, i.e., it satisfies the necessary

condition for the occurrence of conflicts.

These three algorithms were tested and compared in

a simulation study which is developed in the next

section.

We have developed another approach for the

reactive conflict-free routing of the AGVs based on

multi-agent systems (Breton et al, 2006). The main

idea of this approach is to consider an AGV as a

reactive agent, whose goal is to reach a predefined

destination node without conflict with the moving

AGVs. In order to design the AGV-agent, the

odes

Scheduling level

On-line deadlock avoidance

level

(

)

(

)

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

480

Cassiopea Multi-Agent System (MAS) design

methodology is used (Collinot and Droguoul, 1996).

Figure 6: The application of the Cassiopea methodology to

design an AGV-agent of a deadlock-free AGV system.

This methodology defines an agent in five

incremental layers, considering the agent’s different

roles. This incremental construction has the

advantage of satisfying the principle of parsimony,

i.e., the definition of the agent can be stopped as

soon as the system completes its desired function.

The steps followed to design an AGV-agent are

given in figure 6. A gradient G

is calculated in each

step j to meet some specification (for example a

track follow-up). An AGV-agent will move to

minimise this gradient, i.e. according to the direction

where the gradient is minimal. This approach was

also compared in a simulation study to the predictive

approach described before. For more details see

(Breton et al, 2006).

4 SIMULATION OF BI-

DIRECTIONAL AGV SYSTEMS

To test the various approaches described before, we

have used the ARENA software to develop a new

template panel, which allows us to model the routing

of bi-directional AGVs. This function is actually not

included in ARENA package. Our template also

allows the modelling of real AGVSs, subject to

contingencies. This template panel is well described

in (Maza and Castagna, 2005a).

To check for the efficiency of our algorithms

RVWA, RVRAA and RVDA, we made some

simulation tests using our template. The AGVS

under study is composed of bi-directional mesh-like

guide-path of 45 nodes and 60 links and a fleet of 8

AGVs. Each simulation essay is a sequence of at

least 10 replications. In one replication, each AGV

has to realise a set of 100 missions randomly

generated. To approach reality, random failures of

AGVs are generated in the simulation model. They

are characterised by two parameters: the failure rate

τ and the mean time between failures MTBF.

Different simulations were done with various system

parameters in order to compare these algorithms and

bring out the situations where the use of one

algorithm is more appropriate than another (Maza

and Castagna, 2005b).

For example, by varying the failure rate and fixing

other parameters, it can be concluded that more

the failure rate is important, better will be the

makespan achieved by the algorithms RVRAA and

RVDA and that the RVDA gives the best results

(Figure 7. This can be explained by the fact that

the RVDA algorithm is more permissive than the

two other algorithms. More simulation results are

available in (Maza and Castagna, 2005b).

5 AN INDUSTRIAL

APPLICATION

We briefly present here an example of a simulation

study we have done for an industrialist MPO who

wanted to implement an AGV system. The goal of

this study was the prediction of the stores evolution

according to various AGVs management policies.

For more than fifty years now, MPO has been an

expert in the manufacture and replication of pre-

recorded media. Every year, 600 millions discs and

Level 1: Definition of G0

The track follow up and

obstacles avoidance

Level 2: Definition of

G1/ G1=G0+coeff

The move toward a destination

Level 3: Definition of

G2/ G2=G1+coeff

The move awa

when blocked

Level 4: Definition of

G3/ G3=G2+coeff

The move away to a node

Level 5:

Definition of

G4/ G4=G3+coeff

The conflict

prediction

The

individual

role of an

agent

The social

role of an

agent

5000

10000

15000

20000

25000

30000

0 5 10 20

30 40

Figure 7: The makespan evolution according to the failure

rate (τ=0, 5%, 10%, 20%, 30%, 40%).

ON THE USE OF AUTOMATED GUIDED VEHICLES IN FLEXIBLE MANUFACTURING SYSTEMS

481

150 millions printings elements come out of the

MPO’s plants.

This simulation project took place in the

production plant of Averton (France). The

production plant is composed of four workshops

(Figure 8):

The pressing shop: the CDs and DVDs are

obtained by injection of fluidized polycarbonate in a

press mold. Then, a fine layer of aluminum is

deposited in a vacuum, by pulverization, onto the

surface of the disc (Figure 8 (a)).

The printing shop: two printing principles are

utilized. The first one is silkscreen printing. This

technique consists of printing by means of a cloth

frame, which favors the flat decorations. The second

one is offset printing, by means of linked rolls which

is suitable for photos and illustrations in several hues

(Figure 8 (b)).

The packaging shop: the discs are packaged in

their boxes (Figure 8 (c)).

The storage zone: the store is utilized both to

store raw materials and to store the CDs waiting for

package (Figure 8 (d)).

All the transports of goods between the workshops

are realized using an AGV system.

The particularity of this system is that one AGV is

used for towing one or more non-powered carriers as

a train. The AGVs are moving along a unidirectional

loop guided path (Figure 9).

The simulation study is made with a couple of

objectives. The first one is to determine the rule to

be applied to load the carriers. We consider the two

following rules:

R1: the products resulting from different

production orders can be mixed on the same carrier.

R2: one carrier holds only products of the same

production order.

The second objective is to locate and to size the

waiting areas of the carriers. Indeed, it is necessary

to place one or more garages near each workshop

where the carriers can wait for a tow.

The simulator we developed shows how the 600

carriers are distributed between the different storage

areas according the applied production rules. The

evolution of the total number of the waiting carriers

in each storage area according to time is shown on

Figure 10.

6 CONCLUSION

In this paper, we presented some important problems

to be considered when employing AGV systems.

Some of these problems concern the design aspect,

the others the piloting or control aspect. In our

research work, we were concerned with the control

aspect, particularly with the conflict-free routing of

bi-directional AGVSs.

This paper recalls the principle of the approaches we

developed for reactive routing of bi-directional

AGVs and gives some simulation results. We also

briefly presented a simulation study we have done to

help an industrialist to make decisions at the design

stage of their AGV system and also at the piloting

stage. Since this application is confidential, we could

not give more information on it.

REFERENCES

Albert P., 1998. Pilotage de véhicules autoguidés dans

un système automatisé de production. PhD thesis –

University of Nantes.

Beamon, B.M, Chen, V.C.P., 1998a. Performability-

based fleet sizing in a material handling system.

International Journal of Advanced Manufacturing

Technology, Vol.14, No.6, pp.441-449.

Beamon, B.M., Deshpande, A.N., 1998b. A

mathematical programming approach to

simultaneous unit-load and fleet size optimisation

in material handling systems design. International

Journal of Advanced Manufacturing Technology,

Vol.14, No.11, pp.858-863.

Blazewicz, J., Eiselt, H., Finke, G., Laporte, G.,

Wegrlarz, J., 1991. Scheduling tasks and vehicles in

Figure 9: The transportation loop with the stations.

Storage zone

Printing Shop

Packaging Shop

Pressing Shop

Figure 8: The MPO Production facilities.

(a)

(b)

(c)

(d)

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

482

Figure 10: Example of time evolution of the carriers’ number in each waiting area.

600

Time

uantit

waitin

carriers

10000000.0

a flexible manufacturing system. The International

Journal of Flexible Manufacturing Systems, Vol.4,

pp.5-16.

Branislav, H., Mrafko, L., Bielko, V., 2002. A

compraison of the AGV control approaches unsing

Petri Nets. International conference on Systems,

Man and Cybernetics, Tunisia.

Breton, L., Maza, S., Castagna, P., 2006. A multi-agent

based conflict-free routing approach of bi-

directional automated guided vehicles. American

Control Conference, USA.

Castagna, P., Mebarki, N., Maza, S., 2004. Evaluation

du nombre d’AGVs sur un système de production

de type Job shop. 5ème Conférence Francophone

de Modélisation et SIMulation (MOSIM’04),

pp.619-626, France.

Collinot, A., Droguoul, A., 1996. La méthode de

conception multi-agent cassiopée: Application à la

robotique collective. Technical Report LAFORIA,

LIP6.

Egbelu, P.J., Tanchoco, J.M.A., 1984. Characterization

of automatic guided vehicle Dispatching rules.

International Journal of Production Research,

Vol.22, No.3, pp.359-374.

Egbelu, P.J., 1987. The use of non-simulation

approaches in estimating vehicle requirements in an

automated guided vehicle based transport systems.

Material flow, No.4, pp.209-221.

Gaskins, R.J., Tanchoco, J.M.A., 1987. Flow path

design for automated guided vehicle systems.

International Journal of Production Research,

Vol.27, No.6, pp.915-926.

Krishnamurthy, N.N., Batta, R., Karwan, H., 1993.

Developing conflict-free routes for automated

guided vehicles. Operations Research, 41(6),

pp.1077-1090.

Kim, C.W., Tanchoco, J.M.A., 1991. Conflict-free

shortest-time bi-directional AGV routing.

International Journal of Production Research,

Vol.29, No.12, pp.2377-2391.

Lawley, M., Reveliotis, S., Ferreira, P., 1999. Design

guidelines for deadlock handling strategies in

flexible manufacturing systems. International

Journal of Manufacturing Systems, Vol.9, pp.5-30.

Maza, S., Castagna, P., 2001. Conflict-free AGV

routing in bi-directional networks. 8th IEEE

International conference on Emerging Technologies

and Factory Automation, pp.761-764, France.

Maza, S., Castagna, P., 2005a. A performance-based

structural control policy for conflict-free routing of

bi-directional Automated Guided Vehicles.

Computers in industry, Elsevier publication,

Vol.56, n°7, pp.719-733.

Maza, S., Castagna, P., 2005b. Sequence-based

hierarchical conflict-free routing strategy of bi-

directional automated guided vehicles. 16th IFAC

World Congress, Praha.

Oboth, C., Batta, R., Karwan, M., 1999. Dynamic

conflict-free routing of automated guided vehicles.

International Journal of Production Research,

Vol.37, No.9, pp.2003-2030.

Pia Fanti, M., Maione, B., Mascolo, S., Turchiano, B.,

1997. Event-based feedback control for deadlock

avoidance in flexible production systems. IEEE

Transactions on Robotics and Automation, Vol.13,

No.3, pp.347-363.

Reveliotis, S.A., Conflict resolution in AGV systems.

IIE Transactions, Vol.32, No.7, pp.647-659.

Reveliotis, S.A., Ferreira, P., 1996. Deadlocks

avoidance policies for automated manufacturing

cells. IEEE Transactions on Robotics and

Automation, Vol.12, No.3, pp.845-857.

Silberschatz, A., Peterson, J.L., 1991. Operating system

concepts, Addison Wesley.

Taghaboni, F., Tanchoco, J. M. A., 1995. Comparison of

Dynamic Routing

Techniques for Automated Guided Vehicle Systems.

International Journal of Production Research, Vol.10,

No.33, 2653-2669.

Tanchoco, J.M.A., Egbelu, P.J., 1986. Potentials for bi-

directional guide-path for automated guided vehicle

based systems. International Journal of Production

Research, Vol.24, No.5, pp.1075-1097.

ON THE USE OF AUTOMATED GUIDED VEHICLES IN FLEXIBLE MANUFACTURING SYSTEMS

483