Multi-agent Control Approach for Autonomous Mobile Manipulators
Determination of the Best Footsteps Combination
Messous Mohamed Ayoub, Hentout Abdelfetah and Bouzouia Brahim
Centre for Development of Advanced Technologies (CDTA),
Division of Computer-Integrated Manufacturing and Robotics (DPR),
BP 17, Baba Hassen, Algiers 16303, Algeria
Keywords: Multi-agent Control, Mobile Manipulators, Best Footsteps Combination, RobuTER/ULM, Simulation,
JADE.
Abstract: This paper presents our ongoing efforts toward the development of a distributed multi-agent framework for
autonomous control of mobile manipulators. The proposed scheme assigns a reactive agent (Joint agent) to
control each articulation of the manipulator, a hybrid agent (Mobile base agent) for the control of the mobile
base, and a Supervisory agent to coordinate and synchronize the work of all the precedent agents. Each
Control agent implements a Simulation-verification technique, in order to optimize, locally and
independently from the other agents, the value of a predefined Objective function (f
Obj
).
The paper illustrates, also, the methodology we have followed to determine the best combination among
possible footsteps for the Control agents of the system. This combination will be the input, among others, of
the procedure seeking the optimal solution bringing the position of the end-effector of the mobile
manipulator as close as possible to the imposed Target position. For this aim, different simulation scenarios
are described and carried out, with and without considering breakdowns of some articulations of the
manipulator or the mobile base. For the evaluation of the obtained solutions and the selection of the best
footsteps combination, we have considered two criteria (i) f
Obj
values and (ii) the number of iterations.
1 INTRODUCTION
An intelligent mobile manipulator is a kind of
intelligent system, which can autonomously perform
scheduled tasks in complex, unknown and changing
environments with sensing, perceptive, knowledge
acquisition, learning and inference capabilities,
decision making and acting ability (Nebot et al.,
2004), by using only its limited physical and
computational resources with a reduced human
intervention (Medeiros, 1998).
The intelligence of a mobile manipulator is
constructed as an integrated system of many special
software subsystems with different functions such as
manipulation, locomotion, vision, planning, etc. To
realize the global task by the robot, the different
subsystems need to cooperate with each other and
compete for limited resources. Thus, it is very
important to build a high performance autonomous
robot control system to make these subsystems
harmoniously work together to achieve this goals.
The control systems are mainly divided into two
different approaches (i) in the single-agent system,
the only agent has to perform all the actions
(sensing, planning, control, etc.) (ii) in contrast, the
multi-agent system (MAS) decomposes the large
system into many small and distinct agents (Duhaut,
1999) (Erden et al.,2004) (Delarue et al., 2007).
MAS offer simple solutions and benefit of all the
advantages of distributed problem solving. This
perspective made it possible to consider the system
as composed of simple modules, which gave an
easier way to design the whole system. In addition,
the need for massy mathematical models of the robot
(Inverse kinematics model and differential-equation-
solvers) is overcome (Duhaut, 1999). Therefore,
there is a considerable decrease in design effort and
computation time compared to single-agent system.
Finally, with such a usage of MAS, the control is
more flexible to be applied to any robot (mobile,
manipulator or mobile manipulators). Toward
modularity, complexity and advanced adaptive
behaviors of intelligent robot system (Nebot et al.,
2004), we have adopted the multi-agent paradigm.
340
Mohamed Ayoub M., Abdelfetah H. and Brahim B..
Multi-agent Control Approach for Autonomous Mobile Manipulators - Determination of the Best Footsteps Combination.
DOI: 10.5220/0005064403400347
In Proceedings of the 11th International Conference on Informatics in Control, Automation and Robotics (ICINCO-2014), pages 340-347
ISBN: 978-989-758-039-0
Copyright
c
2014 SCITEPRESS (Science and Technology Publications, Lda.)
Our ongoing efforts concern the development of a
generic distributed multi-agent framework for
autonomous control of mobile manipulators. The
proposed scheme assigns a reactive agent (Joint
agent) to control each articulation of the
manipulator, a hybrid agent (Mobile base agent) to
control the mobile base, and a Supervisory agent to
coordinate and synchronize the work of all the
precedent agents. Each Control agent (Joint agent,
Mobile base agent) makes a virtual movement, in all
the possible directions with different footsteps (Joint
footstep, Base Translation footstep, Base Rotation
footstep), locally and independently from the other
agents. After that, the agent computes an Objective
function (f
Obj
) between the current position of the
end-effector and the imposed Target position. The
retained movement, for each agent, is that
optimizing f
Obj
.
The purpose of the current paper is to determine
the best combination of footsteps, for all the Control
agents, in order to reach the optimal solution
bringing the end-effector position as close as
possible to the Target position. For this aim,
different simulation scenarios are described, with
and without considering breakdowns of some
articulations of the manipulator or the mobile base.
The rest of the paper is organized as follows.
Section two describes the proposed multi-agent
scheme for autonomous control of mobile
manipulators. The interaction between the agents of
the system is detailed, also, in this section. Section
three shows the implementation of the proposed
approach on RobuTER/ULM mobile manipulator. It
determines, moreover, the best combination of
footsteps via the obtained simulation results. Section
four concludes the paper and draws-up future works.
2 CONTROL APPROACH
Mobile manipulators are composed of two
heterogeneous very unlike sub-systems (a mobile
base and a manipulator). Consequently, different
controlling entities are solicited to ensure a modular,
yet robust control scheme. The interactions among
these entities assure the required cognitive behaviors
for the robot to accomplish different tasks. Through
a class diagram, Figure 1 shows a generalized view
for the proposed scheme, in which two kinds of
agents can be distinguished:
2.1 System Agents
This kind of agents is intended for the treatment of
data issued from the different sensors equipping the
robot. It can assure the main functionalities related
to or used in the control process such as (i) vision
module, (ii) robot localization module, (iii) target
localization module, etc. However, this list is not
exhaustive; other modules would be thought and
implemented progressively.
2.2 Robot Agents
We focus mainly, in our study, on the second type of
agents, which are dedicated to the control process
itself. We do have though two sub-classes:
(i) Control agents: they are involved in the
computation of the commands to be sent to the
robot (Joints agents and Mobile base agent).
(ii) Supervisory agent: it is responsible for the
synchronization and the coordination between
the Control agents, and for the selection of the
most fitted choices.
Figure 1: Class diagram of the proposed system.
Each Control agent receives, from the
Supervisory agent, the initial situations of the robot
(Configuration
Init
(q
1
, …, q
dof
)
Init
and Base
Init
(x
B
, y
B
,
B
)
Init
) and the imposed coordinates of the Target(x
T
,
y
T
, z
T
) to be reached. The Control agent drives its
corresponding mechanical subset independently
from the other agents, trying to bring the end-
effector as close as possible to Target. Throughout
this process, the current position of the end-effector
Effector(x
E
, y
E
, z
E
) is computed using the Direct
Kinematic Model (DKM) of the robot. This
operation is rather straightforward and doesn’t
require any complex matrix inversion calculus in
contrast of classical robotic approaches (Hentout et
al., 2013). The computation of the DKM is given by
(1) where:
Base(x
B
, y
B
,
B
): it represents the current
situation of the mobile base.
Multi-agentControlApproachforAutonomousMobileManipulators-DeterminationoftheBestFootstepsCombination
341
Configuration(q
1
, …, q
dof
): it is the current
configuration of the manipulator joints (dof:
degrees of freedom).
Effector=DKM(Base, Configuration)
(1)
In the following, we describe the elementary
movements for each Control agent, along with a
short description of the corresponding behaviors.
More details are given in (Hentout et al., 2014).
2.2.1 Joint Agents
Each articulation is controlled by a Joint agent,
which allows two possible elementary movements.
The following process, illustrated in Figure 2, is
implemented to figure out the most fitted choice:
A Joint agent makes a virtual rotation
(MoveUp) in the positive direction
(Configuration
Up
) with a Joint footstep.
The agent computes the objective function value
f
Obj
(Distance
Up
) as shown in (2):
Distance
Up
=F_Objective(Base, Configuration
Up
,
Target)
(2)
The agent repeats these two previous actions
while changing the direction of the rotation
(MoveDown, Configuration
Down
, Distance
Down
).
After comparing the two values (Distance
Up
,
Distance
Down
) with the previous value of f
Obj
, the
Joint agent chooses the action to be made. For some
cases, the best choice would be to stay still because
neither of the two virtual movements would improve
f
Obj
. Subsequently, the selected movement will be
sent, as a proposal (Distance_Joint,
New_Configuration), to the Supervisory agent.
Figure 2: Elementary movements for each Joint agent.
2.2.2 Mobile Base Agent
For the Mobile base agent, we have defined four
elementary movements as presented in Figure 3:
The Mobile base agent makes a virtual forward
movement (MoveForward) with a Base
Translation footstep (Base
FW
).
The agent computes the new objective function
value (Distance
FW
).
The mobile base agent repeats the previous
actions for the other elementary (MoveForward,
TurnRight, TurnLeft) movements (Table 1).
Table 1: Elementary movements for the Mobile base
agent.
Elementary
movements
New situation of
the mobile base
Footsteps f
Obj
Values
MoveForward Base
FW
Translation
footstep
Distance
FW
MoveBackward Base
B
W
Distance
BW
TurnRight Base
TR
Rotation
footstep
Distance
TR
TurnLeft Base
TL
Distance
TL
The Mobile base agent will choose, afterward, its
local best choice, which will be the one optimizing
f
Obj
value amongst the four elementary movements
and the actual position. The selected choice will be,
finally, sent as a proposal (Distance_Base,
New_Base), to the Supervisory agent.
Figure 3: Elementary movements of the Mobile base. agent.
Step 1: MoveUp
Target
E
nd-effector
Distance
Up
Joint2Agent
Joint3Agent
Joint1Agent
Manipulator
Step 2: MoveDown
Target
E
nd-effector
Distance
Down
Joint2Agent
Joint3Agent
Joint1Agent
Manipulator
Effector
Base
FW
Base
BW
Distance
Fw
Distance
BW
Distance
TR
Distance
TL
Step 1: MoveForward
Step 2: MoveBackward
Step 4: TurnLeft
Step 3: TrunRight
Base
TR
Base
TL
Target
Targe
Target
Target
ICINCO2014-11thInternationalConferenceonInformaticsinControl,AutomationandRobotics
342
2.2.3 Supervisory Agent
This hybrid agent is responsible for the coordination
and synchronization between the Control agents.
After receiving the Target position from the human
operator, the Supervisory agent verifies its
reachability (z
T
[z
Min
, z
Max
] where z
Min
and z
Max
are
respectively the minimum and maximum reachable
height in the workspace of the robot). If not
reachable, the agent displays a Target unreachable
error message and terminates the process. Otherwise,
it computes the initial situation of the end-effector
(Effector
Init
) and the initial value of the Objective
function (fObj_
Init
). After that, the Supervisory agent
sends this information along with the initial situation
of the mobile base (Base
Init
) to the Control agents,
and waits, then, for their proposals.
Once all the proposals are received, the Supervisory
agent chooses the best one (best f
Obj
value). The
holder of this proposal will be expecting a Contract
message to confirm the selection and to execute,
thus, the proposed movement.
When f
Obj
reaches a predefined optimum value
(f
Obj
<
), the Supervisory agent terminates the process
and sends a Target reached successfully message.
Otherwise, the Supervisory agent reiterates the
precedent steps and continues the process until
reaching the goal (f
Obj
is optimal).
2.3 Message Exchange Scheme
The interaction between the agents is implemented
via a messages exchange protocol based upon the
famous Contract-net protocol (Davis and Smith,
1983). The sequence diagram of Figure 4 gives a
global vision of the whole interactions among the
Control agents and the Supervisory agent. The
different messages are defined as follows:
INFORM: it is sent by the Supervisory agent to
all the active Control agents at the beginning of
each iteration. This message contains the current
situation of the mobile base (Base), the current
configuration of the manipulator
(Configuration) and the current value of the
objective function (f
Obj
).
CFP (Call For Proposal): this message, sent
after the precedent message by the Supervisory
agent, contains the position of the Target. If we
are dealing with a stationary Target, it is sent
just once at the beginning of the process.
Figure 4: Sequence diagram for the whole system.
Multi-agentControlApproachforAutonomousMobileManipulators-DeterminationoftheBestFootstepsCombination
343
PROPOSE: following the reception of a CFP,
this message, sent by each Control agent to the
Supervisory agent, comprises the best local
proposition of the Control agent
(Distance_Joint, New_Configuration /
Distance_Base, New_Base).
ACCEPT_PROPOSAL/REJECT_PROPOSAL:
after receiving all the propositions from the
Control agents (PROPOSE), the Supervisory
agent selects the best choice and sends an
ACCEPT_PROPOSAL message to the agent
holding this proposition. All the other Control
agents will receive a REJECT_PROPOSAL
message.
ACK: following the reception and the execution
of the best selected movement by the chosen
Control agent, an acknowledgment message is
sent to the Supervisory agent.
3 EXPERIMENTAL WORK
We have adapted the proposed control system to the
characteristics of our experimental robotic platform.
Figure 5 presents a global view for the control
structure of RobuTER/ULM mobile manipulator.
The architecture involves a set of eight agents
(Hentout et al., 2014):
Six reactive Joint agents are assigned to control
the six-dof ULM manipulator.
One hybrid Mobile base agent to control the
mobile base RobuTER of the robot.
One hybrid Supervisory agent to coordinate and
synchronize the precedent agents.
Figure 5: Control scheme of RobuTER/ULM.
In this work, the local objective of each agent is to
reduce the quadratic distance between the current
situation of the end-effector Effector(x
E
, y
E
, z
E
), and
the imposed position of the Target(x
T
, y
T
, z
T
). This
distance is computed as follows:
(3)
JADE (Java Agent DEvelopment Framework)
has been used as an implementation tool of the
proposed approach. The main reason was the fact
that JADE is one of the best modern agent open
source platforms (Iñigo-Blasco et al., 2012). This
framework provides basic middleware-layer
functionalities which simplify the implementation of
distributed applications using a software agent
abstraction (Bellifemine et al., 2008) (Floroian and
Moldoveanu, 2010).
3.1 Simulation Scenarios
Different simulation scenarios have been considered.
They are intended to tune the performances of our
approach facing different Target positions with
different initial situations for the robot. To achieve
this goal, we have implemented the graphical
interface, shown in Figure 6, to be able to determine
the finest combination of footsteps for the Control
agents, which is the main contribution of this paper.
The considered tasks consist of bringing the end-
effector of RobuTER/ULM to the final operational
position Target. The parameters that define each task
are (i) the initial situation of the mobile base Base
Init
(ii) the initial configuration of the manipulator
Configuration
Init
and (iii) the dictated Target. For
validation purposes, we have chosen the five tasks
presented in Table 2. The distances are given in
millimeters (mm) and the angles in degrees (°).
Table 2: Details of the considered validation tasks.
Task Configuration
Ini
t
Base
Ini
t
Target (mm)
f
Ob
j
Ini
t
(mm)
1
(0, 0, 0, 0, 0, 0) (0, 0, 0) (-330, -630, 1080) 1126.9129
2
(0, 0, 0, 0, 0, 0) (0, 0, 0) (-4260, 0, 665)
4
698.9355
3
(0, 60, 0, 0, 32, 0) (0, 0, 0) (-2408, -108, 1472)
3
114.8048
4
(0, 87, 0, 0, 5, 0) (0, 0, 0) (-2400, -63, 1325)
2
946.8779
5
(0, 87, 0, 0, 5, 0) (0, 0, 0) (-2400, -67, 1320)
2
946.8226
The following sub-sections illustrate the followed
methodology in order to determine the best
combination among possible footsteps, for the
Control agents of the system. This combination will
be the input, among others, of the procedure seeking
the optimal solution bringing the position of the end-
effector of the robot as close as possible to the
imposed Target position. In this paper, we have
ICINCO2014-11thInternationalConferenceonInformaticsinControl,AutomationandRobotics
344
considered a static selection of footsteps. So, for each
Joint agent, we have used an elementary rotation
footstep (Joint footstep), whereas we identified two
elementary footsteps for the Mobile base agent (i)
Base Translation footstep and (ii) Base Rotation
footstep.
Figure 6: Screenshot of the developed simulation system.
3.2 Best Combination of Footsteps
A combination of footsteps is composed of the
previous three elementary footsteps, which will affect
greatly the quality of returned results. Therefore, to
find the best combination, the previous tasks have
been tested using various combinations.
After many tests, we have fixed the interval [1,
10] for the values of the footsteps. This will leave us
with thousands of possibilities. Because just one
combination is admissible, we need to make sure to
select the best one. Therefore, and to overlay the
maximum of values, we have selected the
enumerable set {1, 5, 10} for all the footsteps. This
produced 27 (3x3x3) combinations, having the
following structure {Joint, Base Translation, Base
Rotation} footstep, and varying from {1, 1, 1}, {1, 1,
5}, {1, 1, 10} to {10, 10, 10}.
A criterion needed to be defined for the
evaluation of each footsteps combination. Thus, we
considered two criteria (i) the final value of f
Obj
,
along with (ii) the number of iterations. These two
criteria will be evaluated for each of the five previous
tasks. Figure 7 presents a summary histogram for all
the obtained simulation results of each task with the
27 different footstep combinations.
The prior tasks are evaluated without considering
breakdowns (first case). For enhancing further the
selection of the best combination of footsteps, two
other cases are considered (i) breakdown of joints 3
and 4 of the manipulator (second case), and (ii)
failure of the mobile base (third case).
For the following study, we have overlooked the
results obtained from the last case. This is justified
because all the scenarios gave almost identical
solutions. Therefore, considering these results in the
selection of the best footstep combination is
insubstantial. The obtained results for the second
case are given in Figure 8.
Depending on f
Obj
, we have ranked the five best
results for the first two cases (without breakdown and
with joints breakdown) for all the tasks. The
collected data allowed us to draw different tables to
determine the best footsteps combination. The best
combination is the one generating the maximum of
best solutions for all the considered tasks.
Figure 7: Obtained results without breakdown.
Multi-agentControlApproachforAutonomousMobileManipulators-DeterminationoftheBestFootstepsCombination
345
3.2.1 Joint Footstep
According to Table 3, the best Joint footstep=1° ({1,
x, x}).
Table 3: Selection of the best Joint footstep.
Best solution: Without breakdown + Breakdown axes 3 and 4
Footsteps
Task
1
Task
2
Task
3
Task
4
Task
5
Sub-
total
Total
1, x, x
1, 1, x 6 2 2 2 1 13
34 1, 5, x 2 2 2 2 3 11
1, 10, x 2 2 2 2 2 10
5, x, x
5, 1, x 2 1 2 0 1 6
17 5, 5, x 1 2 2 1 2 8
5, 10, x 1 0 1 1 0 3
10, x, x
10, 1, x 0 0 1 1 0 2
04 10, 5, x 0 0 0 1 0 1
10, 10, x 0 1 0 0 0 1
3.2.2 Base Translation Footstep
Table 4 gives the best footstep for the translation
movement of the mobile base (Base Translation
footstep). In this case, we selected two different
footsteps. The first Base Translation footstep=1mm
({x, 1, x}) and the second Base Translation
footstep=5mm ({x, 5, x}).
Table 4: Selection of the best Base Translation footstep.
Best solution: Without breakdown + Breakdown axes 3 and 4
Footsteps
Task
1
Task
2
Task
3
Task
4
Task
5
Sub-
total
Total
x, 1, x
1, 1, x 6 2 2 2 2 14
21 5, 1, x 2 1 2 0 1 5
10, 1, x 0 0 1 1 0 2
x, 5, x
1, 5, x 2 2 2 2 2 10
20 5, 5, x 1 2 2 1 2 8
10, 5, x 0 0 0 1 1 2
x, 10, x
1, 10, x 2 2 2 2 2 10
5, 10, x 1 0 1 1 0 3
14
10, 10, x 0 1 0 0 0 1
3.2.3 Base Rotation Footstep
Table 5 summarizes the best footsteps for the rotation
movement of the mobile base (Base Rotation
footstep). The trivial choice is Base Rotation
footstep=1° ({x, x, 1}).
Table 5: Selection of the best Base Rotation footstep.
Best solution: Without breakdown + Breakdown axes 3 and 4
Footsteps
Task
1
Task
2
Task
3
Task
4
Task
5
Sub-
total
Total
x, x, 1
x, 1, 1 3 3 5 3 3 17
47 x, 5, 1 1 4 4 4 5 18
x, 10, 1 1 3 3 3 2 12
x, x, 5
x, 1, 5 3 0 0 0 0 3
5 x, 5, 5 1 0 0 0 0 1
x, 10, 5 1 0 0 0 0 1
x, x, 10
x, 1, 10 2 0 0 0 0 2
4 x, 5, 10 1 0 0 0 0 1
x, 10, 10 1 0 0 0 0 1
3.2.4 Recapitulation
Four possibilities have been selected by using the
first evaluation criterion (f
Obj
) {1, x, x}, {x, 1, x} and
{x, 5, x}, and {x, x, 1}. The crossing between these
latters generates two different combinations:
{1, 1, 1}.
{1, 5, 1}.
For the selection of one final combination, it is
necessary to adopt another criterion, which consists
of comparing their respective iterations numbers.
From Table 6, it can be noted that the search of the
best solutions by using the first combination, {1, 1,
1}, requires a very high number of iterations and,
consequently, an important execution time compared
with the other combination {1, 5, 1}. This latter is
selected as the best combination of footsteps for
Figure 8: Obtained results with breakdowns of joints 3 and 4 of the manipulator.
ICINCO2014-11thInternationalConferenceonInformaticsinControl,AutomationandRobotics
346
searching the best solution.
Table 6: Final selection of the best footstep between (1, 1,
1) and (1, 5, 1).
Iterations: Without breakdown + Breakdown axes 3 and 4
Footsteps
Task
1
Task
2
Task
3
Task
4
Task
5
1, 1, 1
Without breakdown
75 3555 1719 1647 1638
With breakdown
110 4039 2004 1932 1933
1, 5, 1
Without breakdown
75 810 395 410 411
With breakdown
85 905 444 422 420
4 CONCLUSION
Throughout this paper, we deal with the development
of a novel generic approach to control autonomous
mobile manipulators. The proposed approach is
centered upon an agent-based framework. For the
implementation, we have used the JADE platform
which is one of the most interesting multi-agent
development frameworks. A graphical interface was
developed in order to perform the various validation
scenarios. Finally, simulation results have been
presented and discussed.
The control approach assigns a hybrid agent to
control the mobile base (Mobile base agent), a
reactive agent to control each articulation of the
manipulator (Joint agent), and a hybrid Supervisory
agent to coordinate and to synchronize between the
previous agents. Each Control agent has its own local
goal to be reached independently from the other
agents. It consists of bringing the end-effector as
close as possible from the imposed Target position.
In its current version, the proposed approach
considers a default static selection of footsteps for
each Control agent. Consequently, finding the most
fitted combination of footsteps was an important
quest. Details of the methodology and the measures
we have followed to search the best combination,
were presented in this paper.
As an improvement of the proposed approach, we
intend to implement a heuristic-based technique for
dynamically tuning the footsteps. A fuzzy-logic
inference system seems to be a realistic choice.
Future perspective consists, also, of implementing
and testing the proposed approach on the physical
experimental mobile manipulator robotic platform
(RobuTER/ULM) while accomplishing real
positioning tasks.
REFERENCES
Bellifemine F. L., Caire G., Poggi A., Rimassa G., (2008),
“JADE: A software framework for developing multi-
agent applications. Lessons learned”, Information and
Software Technology, 50, pp. 10-21.
Davis R., Smith R. G., (1983), “Negotiation as a metaphor
for distributed problem solving”. Artificial
Intelligence, 20(1), pp.63-109.
Delarue S., Hoppenot Ph., Colle E., (2007), “A Multi
Agent Controller for a Mobile Arm Manipulator”. The
International Conference on Informatics in Control,
Automation and Robotics (ICINCO2007), 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.
Floroian D., Moldoveanu F. (2010), “Using Robosmith for
Multi-agent Robotic System”. Bulletin of the
Transilvania University of Brasov, Series I:
Engineering Sciences, 3(52).
Hentout A., Messous M. A., Oukid S., Bouzouia B.,
(2013), “Multi-agent Fuzzy-based Control
Architecture for Autonomous Mobile Manipulators:
Traditional Approaches and Multi-agent Fuzzy-based
Approaches”. The 6
th
International Conference on
Intelligent Robotics and Applications (ICIRA2013),
pp. 679-692, Busan, Korea, 25-28 September.
Hentout A., Messous M. A., Bouzouia B., (2014), “Multi-
agent Control Approach for Autonomous Mobile
Manipulators: Simulation Results on RobuTER/ULM”.
The 19
th
World Congress of the International
Federation of Automatic Control (IFAC WC2014),
Cape Town, South Africa, 24-29 August.
Iñigo-Blasco P., Diaz-del-Rio F., Romero-Ternero C.,
Cagigas-Muñiz D., Vicente-Diaz S., (2012), “Robotics
software frameworks for multi-agent robotic systems
development”. Robotics and Autonomous Systems,
60(1), pp. 803-821.
Medeiros A. A. D., (1998), “A Survey of Control
Architectures for Autonomous Mobile Robots”.
Journal of the Brazilian Computer Society, 4(3),
Campinas.
Nebot P., Saintluc G., Berton B., Cervera E., (2004),
“Agent-based Software Integration for a Mobile
Manipulator”. The International Conference on
Systems, Man and Cybernetics (SMC’04), pp. 6167-
6172, The Hague, The Netherlands, 10-13 October.
Multi-agentControlApproachforAutonomousMobileManipulators-DeterminationoftheBestFootstepsCombination
347
