European Journal of Scientific Research

ISSN 1450-216X Vol.25 No.2 (2009), pp.310-324

© EuroJournals Publishing, Inc. 2009
http://www.eurojournals.com/ejsr.htm

Real-time AGV Action Decision in AD-FMS by

Hypothetical Reasoning

Rizauddin Ramli
Department of Mechanical and Materials Engineering, Faculty of Engineering, Universiti
Kebangsaan Malaysia, 43600 UKM Bangi, Selangor, Malaysia
E-mail: rizauddin@eng.ukm.my
Tel: +60-3-58917022; Fax: +60-3-89259659

Hidehiko Yamamoto
Department of Human and Information Systems Engineering
Faculty of Engineering, Gifu University, 501-1193 Gifu-shi, Japan

Abu Bakar Sulong

Department of Mechanical and Materials Engineering
Faculty of Engineering, Universiti Kebangsaan Malaysia
43600 UKM Bangi, Selangor, Malaysia

Dzuraidah Abdul Wahab

Department of Mechanical and Materials Engineering
Faculty of Engineering, Universiti Kebangsaan Malaysia
43600 UKM Bangi, Selangor, Malaysia

Jaber Abu Qudeiri

Department of Mechanical Engineering, Faculty of Engineering
Philadelphia University, Amman 19392, Jordan

Abstract

In this study we present an approach of hypothetical reasoning for action decision

of Automated Guided Vehicle (AGV) in Autonomous Decentralized in Flexible
Manufacturing Systems (AD-FMS). The AD-FMS is characterized as being online, in real-
time mode and of a short-term nature that responds to frequent changing of the production
order. The decentralized control in AD-FMS enables to solve dynamically some typical
task of production system without using a fixed centralized control system. We adopt a
hypothetical reasoning approach that will decide the conceivable next action from the
competition hypothesis. Simulation results show that the efficiency of AGV in AD-FMS
increased.

Keywords: Autonomous Decentralized, Flexible Manufacturing Systems, Automated

Guided Vehicle, Hypothetical Reasoning
Real-time AGV Action Decision in AD-FMS by Hypothetical Reasoning 311

1. Introduction
Today, in order to survive in the rapid challenging environment of the modern manufacturing era,
manufacturers are forced to adopt new technologies, especially for products that are made in small
batch production. Product and process improvements are widely acclaimed to provide economics gains
and to increase flexibility, resulting in fast response and rapid adaptability to changing market
condition. However, most of the production facilities are made as a complex dynamical environment,
which is always plagued by unexpected situations. Equipments may break down and an unexpected
urgent job may suddenly be released to the production line, depends to the demands of the clients.
Also, the priority of the job may be changed. Consequently, the aspect of flexibility in manufacturing
system becomes an essential point in dealing with the unexpected situations.
Basically, flexibility is an attribute of contemporary manufacturing systems which is
necessitated by the time-based competition underlying current manufacturing strategy. As a result, a
Flexible Manufacturing System (FMS) provides an alternative to improve the situation [1-4]. Later, the
advancement in numerical control (NC) and computer technology has made lightly manned
manufacturing system possible. Conventionally, the FMS are equipped with several CNC machines
tools, automated warehouses and Automatic Guided Vehicle (AGV). An AGV based material handling
system is designed and implemented to gain production the flexibility and efficiency [5-7]. However,
even if the FMS are able to deal with the unexpected situations, most of FMS are still leaning on the
centralized control system. A host computer controls and gives instruction to each agents by a pre-
decided rule or route scheduling, i.e. what is the next action they should do after performing the present
task.
So far, the routing algorithms for AGV are often divided by either a centralized approach or a
decentralized approach. For a centralized approach, the route planning of AGV systems is determined
by centralized decision making, which handles the entire system [8]. The Petri Net approaches [9-11]
are a useful way to analyze the conditions to avoid deadlock in AGV systems. Dispatching algorithms
[12-13] and Genetic Algorithms (GA) [14-17] have also been studied to cope with AGV routing
problems.
In the autonomous decentralized system, the AGV routing is generated by several decision
making subsystems. One of the approaches is zone control [18, where the AGV system can be divided
to several non-overlapping regions, which restricts the available AGV for a time. Nishi et al. [19] have
proposed a distributed routing method for multiple mobile robots using a Lagrangian decomposition
and coordination technique. The original problem is decomposed into an individual routing problem
for each AGV. Most of the conventional research on autonomous decentralized real time scheduling
systems for AGV are based on agent decision selection and object orientation method [20-23]. In the
method, the fastest action that can be finished at the existing time is selected as the action that should
be taken for the agent.
Therefore, in this paper, the concept of Autonomous Decentralized [24-27] in Flexible
Manufacturing Systems (AD-FMS) is introduced. It is to realize that each agents in FMS such as
machine tools and AGVs run independently from each others. The AD-FMS architecture has the
feature that every agent has autonomy to manage itself and coordinates with the other agents.
Furthermore, we adopt a concept of hypothetical reasoning to retrieve whether the action taken by the
agent is the truth or false and use the hypothesis for the next action decision.
Consequently, the proposed coordination between agents is achieved by communication with
other agents through a proposed intelligent knowledge (IK), in which the information of the agent
circulates or transmits to another agent and after receiving the information; the other agents analyze its
content to proceed for the next action decision.
312 Rizauddin Ramli, Hidehiko Yamamoto, Abu Bakar Sulong, Dzuraidah Abdul Wahab and
Jaber Abu Qudeiri

2. Architecture of AD-FMS
2.1. Conventional FMS
Conventionally, FMS is a computer-controlled configuration of semi-independent work stations and
material handling systems designed to efficiently manufacture more than one type of parts at a low to a
medium volumes [28]. There are three essential physical components of FMS:
• Numerical control (NC) machine tools
• Conveyance network or material handling system
• FMS control system
The NC machine tools do not only consists of NC machining centers but it also may comprise
any of the machining units in the FMS such as NC lathe machines, turning machines, etc. On the other
hand, the conveyance network or the material handling system such as AGV functions as a device to
transfer the work piece between the parts warehouse, product warehouse and machining centers.
The FMS control system performs as a centralized host computer that controls the sequence and
coordinates all the task flow for every machine tool, parts handling system and the work pieces. Figure
1 depicts the conventional FMS control system giving all instructions to other sub-level control
systems, i.e. the machine control system, station control system and transportation control system.
Then, these sub-level control systems command the equipment in the FMS to do their task based on the
pre-decided sequence or schedule. The disadvantage of this case is once the centralized host fails to
communicate with the sub-level control system, the operation of FMS is terminated due failure of
coordinating the equipment. In order to overcome this problem, a better control system which does not
depends to a centralized oriented structure is needed, i.e., a decentralized control system where every
element in the FMS are independently and flexible to decide their own decision through
communication, exchanging information and cooperation among them.

Figure 1: Schematic hierarchy of FMS control system

Production Control
System

Part W-house CAD/CAM

station Automatic Pallet Automatic
Stocker W-house

Station Control
FMS Control Conveyance
Control system
System System

AGV

3-D measurement
MC Jig& Tool station
MC Control
System

MC-2
MC-1
Real-time AGV Action Decision in AD-FMS by Hypothetical Reasoning 313

2.2. Concept of AD-FMS

The AD-FMS is based on the Autonomous Decentralized (AD) that was derived from the analogy of
living organisms. Each living thing is composed of cells and each cell is independent in the body.
These cells are totally self sufficient for the information for living and multiplication. In our proposed
AD-FMS, these cells are analogous to autonomous multi-agents systems. Any communications
between these agents are carried between themselves without routing through a centralized system. In
Figure 2, a schematic view of information sharing between the autonomous agents of the AD-FMS
structure is shown. It shows the agents which are AGVs and machine tools communicate and exchange
their information via a wireless communication system. By this way, they are able to estimate the
ability of themselves, which machine tools are busy and leisure; so that the AGV can decide which
machine tools should the parts to input, or which machine tools are finished processing the parts.
Furthermore, through the coordination between AGVs, they can avoid the collision between
themselves or to prevent deadlock that can decrease the efficiency of the FMS.
Consequently, the AD-FMS can be realized with the multi-flow of information from the any
level of computer. For example, once a daily production command is given every control system will
automatically schedule their job scheduling based on it. By this way they transmit the information
through LAN system to the other computer system and to the station controller. As shown in Figure 3,
the station controller is then autonomously give instruction to the agents, which are the AGVs,
machine tools and automatic warehouse about the job scheduling and keep coordinating them online.

Figure 2: Information sharing between autonomous agents

Communicate-Exchange-Cooperate

AGV AGV MC MC

With this architecture, it can be realized that in the AD-FMS, there are no specific centralized
host controller and no relation of master-slave among the multi-agents. However, few criteria’s of AD-
FMS should be satisfied in order to gain a full efficiency of AD-FMS that are:
• On-Line Expansion
As the AD-FMS size increases, a step by step construction is required. Even after completion of
construction, the function in the AD-FMS may be are added or removed.
• On-Line Maintenance
In the AD-FMS, it is possible that frequency of fault occurrence somewhere in the system will
be increased. Due to this, it should be ensured that the maintenance and the test procedures can
be carried out without suspending the operation of the AD-FMS itself, especially in the case of
on-line and real-time systems.
• Fault-Tolerance
The hardware in the AD-FMS must be reliable to compete with faulty. The hardware in the
AD-FMS should be more sufficiently improved in comparison with the software. However,
314 Rizauddin Ramli, Hidehiko Yamamoto, Abu Bakar Sulong, Dzuraidah Abdul Wahab and
Jaber Abu Qudeiri

even if the software includes some bugs, the system is not required to stop its entire operation
to prevent the fault.
• Performance
In AD-FMS, a high performance of control systems is required. In order to attain the high
performance control system, reducing the peak of the computer load and making the load
smooth by improving the software processing are required.
All this criteria should be met in the construction and operation of new AD-FMS. However, it
is difficult to satisfy all these criteria. In this paper, we focused at the forth criteria that is to improve
the software processing by introducing the concept of hypothetical reasoning in order to relieve the
load of software processing.

2.3. Difficulties of Realizing AD-FMS

In AD-FMS, the larger scale of product demands will result to a huge combination of parts varieties,
job scheduling, maintenance and control systems of every agent. Due to this, it is necessary to optimize
the huge combination efficiently so that the productivity in the AD-FMS will be enhanced. For
instance, a manufacturer without AD architecture in their FMS will face difficulties to cope with the
customer demands if one of the AGV caused trouble. Most of the Japanese manufacturers practice the
concept of Just in Time (JIT) that ensures the needed product with the needed volume in the times
could be delivered to the customer without any delay [29]. The failure of it will cause a big problem to
the customer and lost the trust for the manufacturer. That is why a robust AD-FMS architecture is
needed.

Figure 3: Distribution of task by Controllers

Machine
Tool

Computer

Controller Controller Automatic

Warehouse

Controller
Controller

AGV AGV
Controller Controller

Computer

Machine Machine
Tool Tool

However, it is not an easy task to develop a robust AD-FMS even now if we have the ability of
an efficient computer. This is because it is considerably a difficult task to obtain the optimal solution
for all combinations. For instance, if the scheduling job is one of the combinations problems, the
computer can enumerate all the solution by its candidates and search for the best solution from among
Real-time AGV Action Decision in AD-FMS by Hypothetical Reasoning 315

them. However, the combination becomes large with the increase in the types of parts, number of
machines and AGV. When all these large combinations simultaneously correspond to the control
system of the machine tool, the breakdown of the control computer, and the changes in the production
plan, etc, it becomes extremely complex to design such an efficient AD-FMS. In other words, it is one
of the major problems to construct a real-time AD-FMS.
Here, we consider the problems that happen in constructing the AD-FMS as follows:
• The method of deciding an autonomous action:
Unlike the centralized approach, where all the action decisions are made by a FMS
control system, all the agents in the AD-FMS have to decide their own decisions quickly
and then execute the action. Therefore, it is important to develop a highly efficient and
high-speed algorithm.
• The method of achieving cooperated action:
In order to perform an autonomous action, each agent should behave as an individual with
an intelligent system, so that they can communicate and understand the behaviour of the
other agents. However, the control systems of this method are extremely crucial and to
structuralize the algorithm is a troublesome task.
• The communication cost:
To implement a cooperative action between each agent, much information needs to be
transferred between agents which require a sophisticated telecommunication system. It
required a lot of money to realize a telecommunication system in AD-FMS which not
every industry affords to.

3. Model of AD-FMS
3.1. Action Decision in AD-FMS
In this paper, the model of the AD-FMS that we study consists of multi-agents inside a factory that is
shown in Figure 4. These agents are a Parts Warehouse (PAW) that supplies parts for a factory, a
Product Warehouse (PRW) that stores the finished parts from MC, transportation systems for material
handling (AGVs) that carry parts and several MCs arranged at some specific positions. Each AGV
carries only one parts at one time. The movement of the AGV inside the FMS is restricted on the
dashed line grid with uniform speeds. The MCs can machine several types of parts and the machining
time and manufacturing process for each type of MC is decided. The set of the same MC is called a
group MC and is classified by describing subscript, for example, MC1, MC2,…,MCn. Each MC in the
same group is distinguished by attaching a hyphen and figures after the name of the group MC, for
example, MC1-1, MC1-2, …,MCn-m.
316 Rizauddin Ramli, Hidehiko Yamamoto, Abu Bakar Sulong, Dzuraidah Abdul Wahab and
Jaber Abu Qudeiri

Figure 4: Model of AD-FMS

P a rts AGV 1 AGV 2 AGV 5

W a reh o u se
(P A W )

P ro d u cts M C 1 -1 M C 2 -1 M C 5 -1 M C 1 -2 M C 3 -2
W a reh o u se
(P R W ) AGV 3 AGV4

M C 3 -1 M C 4 -1 M C 6 -1 M C 2 -2 M C 1 -3

M C 7 -1 M C 8 -1 M C 7 -2 M C 2 -3 M C 4 -2

M C 3 -3 M C 5 -2 M C 4 -3 M C 8 -2 M C 6 -3

M C 6 -2 M C 5 -3 M C 8 -3 M C 7 -3

The information exchange and cooperation between each agent in this AD-FMS is described as
follows. The PAW sends the information on the names of the parts that are stored in the PAW.
Meanwhile, the AGV transmits the information of the name of the parts that it brings and the
destination of where it is going. The MC gives information of the name of parts that is currently
machined and the time remaining to finish the machining process. The PRW sends information to the
PAW of what product is kept in it and which product has been sent to customers. By this way, the
PAW can understand of the current logistic condition so that it can prepare the next parts that should be
input to the FMS. In other words, the information exchange between agents in the AD-FMS is used by
the needed agent as a source to decide the next action.
In AD-FMS, the full usage of information source among MC, PAW, PRW and AGVs
simultaneously will result to the realization of enhancing production efficiency [30]. In the
conventional FMS, the action planning of the AGV action is done by a pre-decided scheduling system
that does not consider an unexpected problem such as MC troubles or machining delay time. Once this
unexpected trouble occurs, the production plan needs to be re-scheduled. Furthermore, in the case of an
AD-FMS where many MCs and AGVs are mixed together, it is difficult to schedule an effective
instruction for the AGV about where it should go and which parts should be input.
In this paper, the processing procedures that we adopt are described as follows;
• The usage of information from each agent
• The inference of a few steps of AGV action
• The forecasting of the AD-FMS operating condition.
In order to implement these procedures, we propose an algorithm which is able to forecast the
next action decision in the real-time production scheduling of the AGV that is based on hypothetical
reasoning. Hypothetical reasoning is the activity of evaluating the effect of the actions that affect a
given domain that is now an established subfield of knowledge representation [31-35. The algorithm
that we proposed is able to forecast the next action decision in the real-time production scheduling of
Real-time AGV Action Decision in AD-FMS by Hypothetical Reasoning 317

AGV that we called as Future Anticipative Reasoning Algorithm (FARA) which is based on
hypothetical reasoning method. The FARA’s real-time production scheduling is performed by 2 types
of the hypothetical reasoning; the first hypothesis is the Action Decision Hypothetical Reasoning
(ADHR) that decides where the AGV will move to and the second hypothesis is the Parts Input
Hypothetical Reasoning (PIHR) that decides the kinds of parts to be input onto the production floor.

Figure 5: Action Planning of AGV

Current
Position

Next Action M C1-1 M C1-2 M C2-1 PAW PRW M C3-1

M C1-1 M C1-2
The Follow ing
Action

PAW : Part W arehouse

PRW : Product W arehouse

The next action decision for AGV is related to many reasons, such as the existence of many
MCs with the same machining process, the transportation of product to product warehouse, the input of
new parts to the AD-FMS, the existing of other AGV that are doing the same action, etc. Due to these
reasons, the action decision necessitates not only a single action decision but could be a multi action
decision that is based on the action decision selection branch. Attentively, if the AGV selects one
choice from the selection branch and then based on the selected branch, each of the agents inside the
AD-FMS is given moving and working instructions.
Furthermore, when the AGV meets the condition that is required to do the selection again, then
it will re-select one of the choices from the action decision selection branch. In this way, the operating
condition in AD-FMS is realized by the continuous process of selecting the AGV next action decision.
In other words, as shown in Figure 5, the operating condition of AGV is eternally broadening like a
tree structure, where the node is assumed as the next AGV action.
Figure 6 shows the outline of the proposed hypothetical reasoning process. The tree structure
shows the form of retrieval vertically. In every stage of the tree structure, the retrieval will be done
until it finds FALSE results and it will return to one stage back and start the retrieval process again at
another selection branch. The peak of the tree structure is set as the last AGV action with the
hypothesis depth 0 and the hypothesis depth under this stage represents the action that is taken by the
AGV. Similarly, the hypothesis with depth 1 has its own selection branch under it with all of them
have their own selection branch respectively. In Figure 6, the selection branch and the arrows
connecting the selection branch shows the hypothetical reasoning simulation process of FARA. The
hypothetical reasoning simulation runs as if the end of the arrow is TRUTH, then by doing the task
following the selection branch, it will display the result of the simulation.
The algorithm of the hypothetical reasoning simulation is performed through the following
steps:
Step1: The existing AGV hypothesis depth is set as 0.
318 Rizauddin Ramli, Hidehiko Yamamoto, Abu Bakar Sulong, Dzuraidah Abdul Wahab and
Jaber Abu Qudeiri

Step2: For the next hypothesis depth, if the selected branch is FALSE, then the farthest left side branch
is selected and assumed as the TRUTH.
Step3: Run simulation to the selected branch.
Step4: Based on the simulation result, the selection branch is judged whether it is TRUTH or not.
Step5: If the simulation result is FALSE, then go to STEP 6. If it is TRUTH, then go to STEP 8.
Step6: In the depth of the same hypothesis that has been judged to be FALSE, if the left side of the
selection branch that has not yet judged as TRUTH or FALSE is then selected to be assumed as
TRUTH and go to STEP 3.

Figure 6: Hypothetical reasoning processes

Hypothesis Depth 0

Hypothesis Depth 1

Hypothesis Depth 2

FALSE decision branch Hypothesis Depth 3

TRUTH decison hypothetical reasoning
Order of reasoning finished

Step7: Go up to another depth of hypothesis and go to STEP 6.

Step8: If the hypothesis is above a value then go to STEP 9, if not go to STEP 2.
Step9: Selection branch becomes TRUTH, then the hypothetical reasoning is finished.

3.2. Action Decision in AD-FMS

Two types of hypothetical reasoning, the ADHR and PIHR can be used by recalling each others
respectively. For instance, when one AGV is moving from one place to the parts warehouse under the
ADHR, if the hypothesis is judged as TRUTH, then after it arrives at the parts warehouse; it will
become the next hypothesis. Then, when the AGV takes parts from the parts warehouse under the
decision of PIHR, next it will execute the ADHR for the next action. Furthermore, when the
hypothetical reasoning is being performed, only one AGV will carry out the hypothesis. In other
words, when the hypothetical reasoning is performed by one AGV, the other AGV will only start the
hypothesis after the AGV finishes its hypothesis.
In the hypothetical reasoning simulation process, in the case where there is a selection branch
with same level of efficiency, the higher ranking of selection branch will be selected, i.e., the
possibility of the higher ranking of selection branch to be selected arises. For examples, when there are
two selection branches with the same efficiency in existence, the selection branch that is precedent
judged to be TRUTH will be automatically carried out. In this way, we use the characteristic of
hypothetical reasoning to bring the products production rates to be closer to its target by using the
selection order function which depends to the production situation.
Real-time AGV Action Decision in AD-FMS by Hypothetical Reasoning 319

3.3. AGV with Intelligent Knowledge

In this paper, we consider the AGVs as intelligent agents that are able to adopt knowledge, transmit
their information to each other and understand AGVs behaviour. If one AGV can understand the
behaviour of another AGV, it is possible to avoid their collision, and to cooperate in their task together.
Here, we define each AGV as having 6 types of intelligent knowledge, i.e., Routing knowledge, Self
knowledge, Sending knowledge, Others knowledge, Answer knowledge and finally Avoidance
knowledge. These 6 types of knowledge are divided into 2 types of memories: long term memory and
short term memory. Sending, Answer and Avoidance knowledge are kept in the short term memory,
while Routing knowledge, Self and Others knowledge are kept inside the long term memory.
Figure 7 shows the Self knowledge. The first parameter and second parameter indicates the
name and emergency command of the specified AGV respectively. The third, fourth and fifth
parameter indicates the position of the location that the AGV has just passed, the next position that the
AGV will go to and the next position after that. The Others knowledge has the same characteristic with
the Self knowledge.
The Answer knowledge has 4 parameters, i.e., the AGV’s name, emergency command, the
name of AGV opponent and the next position that the AGV will go. All of the information are
transmitted to other AGV, so that they can share and exchange their information to avoid collision
between them. By this way, all the AGV can reduce their collision and as well as can avoid deadlocks
that will decrease the production efficiency.

Figure 7: Self knowledge

1st para 2nd para 3rd para 4th para 5th para
AGV
para: parameter

4. Simulations
The production floor is a 60m × 60m square. (The parts warehouse and the product warehouse are
located outside of the floors). The distance that AGV are able to move is 5 m from the inside floor
wall. It is assumed that the entrance to the parts warehouse and the product warehouse use the shortest
route from the grid. Furthermore, the positions of MCs are located at the edge of the grid of the AGV
route.
In this research, in order to facilitate the dynamic AD-FMS simulation, the following
assumptions are made.
• The maximum number of AGV is 5. The AGV moves on the grid of the floor, and the
travelling speed of AGV is constant, but it depends on the type of carried parts and
products.
• The maximum types of MCs are 8 types and the maximum numbers of same type of MC
are 3 types. The position is assumed as entrance of parts handling position.
• The maximum numbers of parts types are 9 types with each types can only be processed
maximum 8 times.
In this paper, 3 kinds of simulation conditions were performed to ascertain the effectiveness of
our proposed algorithm with AGV-wIK. The position of parts warehouse, product warehouse, and the
MCs on the production floor are configured as Figure 8, Figure 9 and Figure 10, respectively. The
320 Rizauddin Ramli, Hidehiko Yamamoto, Abu Bakar Sulong, Dzuraidah Abdul Wahab and
Jaber Abu Qudeiri

position of MCs (represents by ■) and numbers of AGVs are changed in each different simulation
conditions. Each simulation time is 8 hours and the numbers of AGV used in each simulation condition
are 3:4:5. In order to verify the effectiveness of AGV-wIK, we ran simulations with the same condition
without intelligent knowledge (i.e. Conventional Method) where the moving destination is decided and
fixed.

(a). Simulation 1: AD-FMS with 9 Machine Tools

In Simulation 1, we ran 3 types of simulations (S1-1, S1-2 and S1-3) with different numbers of AGV
are performed with the following conditions: 3 types of products with the rates of each product and
production ratios as P1:P2:P3=5:6:3.

Figure 8: MCs position in Simulation 1

PAW

PRW

(b). Simulation 2: AD-FMS with 18 Machine Tools

In Simulation 2, we ran 3 types of simulations (S2-1, S2-2 and S2-3) with 6 types of products with the
following conditions: The rates of each product and production ratios as P1:P2:P3:P4:P5:
P6=5:6:3:3:2:1.

Figure 9: MCs position in Simulation 2

PAW

PRW
Real-time AGV Action Decision in AD-FMS by Hypothetical Reasoning 321

(c). Simulation 3: AD-FMS with 24 Machine Tools

Figure 10: MCs position in Simulation 3

PAW

PRW

In Simulation 3, we ran 3 types of simulations (S3-1, S3-2 and S3-3) with different numbers of
AGV under the following condition: 9 types of products with the rates of each product and production
ratios as P1: P2: P3: P4: P5: P5: P6: P7: P8: P9 = 5: 6: 3: 3: 2:1: 4: 5: 2.
In Figure 11(a), each Simulation 1, Simulation 2 and Simulation 3 indicates that the efficiency
of AGV becomes better with AGV-wIK than using the Conventional Method. Similar results obtained
where the number of AGV collisions are reduced as shown in Figure 11(b). This proved that our
proposed technique works effectively.

5. Conclusions
In this paper, we proposed a technique of forecasting the next action of AGV that includes the advance
prediction of action in few steps, which will enable to enhance the efficiency condition of the
autonomous decentralized flexible manufacturing systems. The technique, Future Anticipative
Reasoning Algorithm or FARA is used to forecast the next action decision of AGV. By using FARA,
we adopt a hypothetical reasoning technique that will decide the conceivable next action from the
competition hypothesis. Simulation results show that the efficiency of AGV in AD-FMS increased.
The numbers of collisions are also decreased, which means we can obtain a proper navigation for the
AGV. It confirmed that our technique is useful in collision avoidance.

Figure 11: Results in each simulation conditions

70

60
AGV efficiencies(%)

50

40

30
Conventional Method
20 AGV-wIK
10

0
S1-1 S1-2 S1-3 S2-1 S2-2 S2-3 S3-1 S3-2 S3-3
Simulation Conditions
322 Rizauddin Ramli, Hidehiko Yamamoto, Abu Bakar Sulong, Dzuraidah Abdul Wahab and
Jaber Abu Qudeiri

(b) Comparison of number of collisions between AGVs

1400

1200

Number of collisions 1000

800 Conventional M ethod

600 AGV-wIK
400

200

S1-1 S1-2 S1-3 S2-1 S2-2 S2-3 S3-1 S3-2 S3-3

Simulation Conditions

References
[1] Groover, M.P., 1987. Automation, Production System and Computer Integrated Manufacturing,
Prentice Hall: New York
[2] Kaighobadi, M. and Venkatesh, K., 1994. Flexible Manufacturing Systems: an overview,
International Journal of Operations & Production Management 14, pp. 26-49.
[3] Chen, F.F. and Adam, E.E., Jr., 1991. The impact of flexible manufacturing systems on
productivity and quality, IEEE Transactions on Engineering Management 38, pp.33 -45
[4] David, H. Ben-Arieh, Colin, L. Moodie and Chi-Chung, Chu, 1988. Control Methodology for
FMS, IEEE Journal of Robotics and Automation 4, pp. 53-59
[5] Guy, P. E. and Castleberry, A., 1991. The AGV Handbook, Braun-Brumfield, Inc., Ann Arbor,
Michigan.
[6] Doo, Yong Lee and DiCesare, F., 1994. Integrated scheduling of flexible manufacturing
systems employing automated guided vehicles, IEEE Transactions on Industrial Electronics
41, pp. 602-610
[7] Furukawa, M. Watanabe, M. and Kakazu, Y., 1999. AGV autonomous driving based on scene
recognition acquired by simplified SDM, Proceedings of 1999 IEEE International Conference
on Systems, Man, and Cybernetics 6, pp. 649-654
[8] Leduc, R.J., Lawford, M., and Dai, P., 2006. Hierarchical Interface-Based Supervisory Control
of a Flexible Manufacturing System, IEEE Transactions on Control Systems Technology 4,
pp.654-668
[9] Moro, A.R., Yu, H. and Kelleher, G., 2000. Advanced scheduling methodologies for flexible
manufacturing systems using Petri nets and heuristic search, Proceedings. ICRA '00. IEEE
International Conference on Robotics and Automation 3, pp. 2398-2403
[10] Chung, Y.Y, Fu, L.C. and Lin, M.W., 1998. Petri net based modeling and GA based scheduling
for a flexible manufacturing system, Proceedings of the 37th IEEE Conference on Decision and
Control 4, pp. 4346-4347
[11] Yoo, J. W., Sim, E. S., Cao C. and Park, J. W., 2005. “An Algorithm for Deadlock Avoidance
in an AGV System”, International Journal of Advance Manufacturing Technology 26,pp. 659-
668.
[12] Kim, S. H. and Hwang, H., 1999. An adaptive dispatching Algorithm for Automated Guided
Vehicles based on an evolutionary process, International Journal of Production Economics(
60-61), pp. 465-472
Real-time AGV Action Decision in AD-FMS by Hypothetical Reasoning 323

[13] Indrayadi, Y., Valckenaers, H.P. and Van Brussel, H., 2002. Dynamic multi-agent dispatching
control for flexible manufacturing systems, Proceedings of 13th International Workshop on
Database and Expert Systems Applications, pp. 578-582
[14] Endo, S., Konishi, M., Moriwaki, T. and Yoshida, M., 2000. Motion Planning Method for
Mobile Robots in Large Scale Transportation Systems by Genetic Algorithms, Transaction of
Instrumental System Control Information Engineers 13, pp. 115-123.
[15] Kawakami, S. and Kakazu, Y., 1993. A study on the autonomous robot navigation by use of the
classifier systems, Transaction of Japan Society of Mechanical Engineers C Series 59, pp.
2339-2445
[16] Chen, J.-H. and Ho., S.-Y., 2005. A novel approach to production planning of flexible
manufacturing systems using an efficient multi-objective genetic algorithm, International
Journal of Machine Tools and Manufacture 45, pp. 949-957
[17] Hsu, T., Dupas, R. and Goncalves, G., 2002. A genetic algorithm to solving the problem of
flexible manufacturing system cyclic scheduling, 2002 IEEE International Conference on
Systems, Man and Cybernetics 3
[18] Ho, Y.C., 2000. A dynamic zone strategy for vehicle-collision prevention and load balancing in
an AGV system with a single-loop guide path, Computer in Industry 42, pp. 159-176
[19] Nishi, T., Konishi, M. and Hasebe, S., 2005. An Autonomous Decentralized Supply Chain
Planning System for Multi-Stage Production Processes, Journal of Intelligent Manufacturing 3,
259-275
[20] Ouelhadj, D., Hanachi, C. and Bouzouia, B., 2000. Multi-agent architecture for distributed
monitoring in flexible manufacturing systems (FMS), Proceedings of IEEE International
Conference on Robotics and Automation 3, pp. 2416- 2421
[21] Kawano, K., Sameshima, S. and Adachi, Y., 1999. An autonomous decentralized
manufacturing system architecture and its trends for de-facto standards, Proceedings of 1999
IEEE International Conference on Systems, Man, and Cybernetics 6, pp. 422-427.
[22] Kouiss, K., Pierreval, H. and Mebarki, N., 1997. Using multi-agent architecture in FMS for
dynamic scheduling, Journal of Intelligent Manufacturing 8, pp.41-47
[23] Weiming, S., 2002. Distributed Manufacturing Scheduling Using Intelligent Agents, IEEE
Intelligent Systems 17, pp. 88-94
[24] Murphy, R.R. and Arking, R.C., 1988. Autonomous mobile robots in flexible manufacturing
systems, Proceedings of the Fourth Conference on Artificial Intelligence Applications, pp. 412-
414
[25] Sankar, S.S., Rajkumar, R., Ponnambalam, S.G. and Gurumarimuthu, M., 2004. Integrated
scheduling of material handling and manufacturing activities in flexible manufacturing system,
Proceedings of 2004 IEEE International Conference on Systems, Man and Cybernetics, pp.
4254- 4259
[26] Yamamoto, H. and Marui, E., 2005. Intelligent Communication Between Agents of
Autonomous Decentralized FMS, Proceedings of 2005 IEEE International Symposium on
Computational Intelligence in Robotics and Automation, pp. 433-438
[27] Kawamichi, H., Sameshima, S., Kawano, K. and Adachi, Y., 2001. Transmission performance
testing for cooperating processes of an autonomous decentralized manufacturing system,
Proceedings of 8th IEEE International Conference on Emerging Technologies and Factory
Automation 2, pp. 13-22.
[28] Draper, C., 1984. Flexible Manufacturing Systems Handbook, Automation and Management
Systems Division, The Charles Stark Draper Laboratory, Inc., Noyes Publication
[29] Ohno, T., 1998. Toyota Production System: Beyond Large-Scale Production, Productivity Press
[30] Yamamoto, H., 2000. AGV's Actions Decision by Reasoning to Anticipate Future in
Decentralized Autonomous FMS, Proceedings of 2000 Japan-USA Symposium on Flexible
Automation.
324 Rizauddin Ramli, Hidehiko Yamamoto, Abu Bakar Sulong, Dzuraidah Abdul Wahab and
Jaber Abu Qudeiri

[31] Poole, D., 1990. A methodology for using a default and abductive reasoning system,
International Journal of Intelligent Systems 5, pp.521-548.
[32] Provetti, A., 1996. Hypothetical reasoning about actions: from situation calculus to event
calculus, Computational Intelligence 12, pp. 478-498.
[33] Baldoni, M., Giordano, L. and Martelli, A., 1998. A modal extension of logic programming:
Modularity, beliefs and hypothetical reasoning, Journal of Logic and Computation 8, pp.597-
635
[34] Bonner, A.J., 1988. A Logic for Hypothetical Reasoning, Proceedings of Seventh National
Conference on Artificial Intelligence, pp. 480-484
[35] Dubois, D. and Prade, H., 1996. Combining hypothetical reasoning and plausible inference in
possibilistic logic, Journal of Multiple Valued Logic 1, pp. 219-239