Suppressing Energy Consumption of Transportation Robots using
Mobile Agents
Ryosuke Shibuya
1
, Munehiro Takimoto
1
and Yasushi Kambayashi
2
1
Department of Information Sciences, Tokyo University of Science, Chiba, Japan
2
Department of Computer and Information Engineering, Nippon Institute of Technology, Saitama, Japan
Keywords:
Mobile Agent, Multi-robots, Energy Consumption, Intelligent Robot Control.
Abstract:
This paper presents an application for controlling multiple robots connected by communication networks.
Instead of making multiple robots pursue several tasks simultaneously, the framework makes mobile software
agents migrate from one robot to another to perform the tasks. Since mobile software agents can migrate to
arbitrary robots by wireless communication networks, they can find the most suitably equipped and/or the most
suitably located robots to perform their task. We previously implemented an application of searching targets in
our framework, and showed that it could suppress energy consumption of the entire system. In this paper, we
propose a new mobile agent system that transports the found targets to a collection area. The approach is based
on passing the targets among robots through throwing, and therefore, suppresses energy consumption as the
same manner for searching targets. In order to show the effectiveness of our approach, we have implemented
two strategies on a simulator and conducted numerical experiments. As a result, we show that our approach
has more advantages than previous ones, and there is a remarkable difference between the two strategies in
terms of energy saving.
1 INTRODUCTION
In the last decade, robot systems have made rapid
progress not only in their behaviors but also in the
way they are controlled. In particular, a control sys-
tem based on multiple software agents can control
robots efficiently (Takimoto et al., 2007).
On the other hand, excessive interactions among
agents in the multi-agent system may cause problems
in the multiple robot environments. In order to miti-
gate the problems of excessive communication, mo-
bile agent methodologies have been developed for
distributed environments(Kambayashi and Takimoto,
2005). Mobile agent systems are especially useful in
an intermittently connected ad hoc network environ-
ment. In the minimal case, a mobile agent requires
that the connection is established only when it per-
forms migration (Binder et al., 2001).
The model of our system is a set of cooperative
multiple mobile agents executing tasks by controlling
a pool of multiple robots as shown by (Kambayashi
and Takimoto, 2005). A mobile agent can migrate to
the robot that is most conveniently located to a given
task, e.g. closest robot to a physical object such as a
soccer ball. Since the agent migration is much easier
than the robot motion, the agent migration contributes
to saving power consumption (Takimoto et al., 2007).
We have proposed our model in the previous pa-
per (Takimoto et al., 2007) and have also shown the
effectiveness of saving powerconsumptionand the ef-
ficiency of our system for searching targets (Nagata
et al., 2009; Abe et al., 2011). In this paper, we focus
our attention on not only searching targets process but
also transporting the found targets to a designated col-
lection area. The sequence of behaviors for searching
targets and transporting them is one of the most im-
portant tasks for autonomous robots. For example,
the sampling missions on Moon and Mars are well-
known. Also, it is important for transporting target
things from the stricken area where the environment
is too dangerous for human to work. Our approach
enables robots saving energy consumption by pass-
ing targets among robots through throwing them. A
software mobile agent controls a robot and makes it
throw an object, and then the agent migrates to an-
other robot that receives the target. Thus instead of
mobile robots, the software mobile agent has the re-
sponsibly to transport the target object to the desti-
nation, and it achieves suppressing the number of ro-
tations of robot-wheels. Furthermore, our transporta-
219
Shibuya R., Takimoto M. and Kambayashi Y..
Suppressing Energy Consumption of Transportation Robots using Mobile Agents.
DOI: 10.5220/0004240902190224
In Proceedings of the 5th International Conference on Agents and Artificial Intelligence (ICAART-2013), pages 219-224
ISBN: 978-989-8565-38-9
Copyright
c
2013 SCITEPRESS (Science and Technology Publications, Lda.)
tion technique has a good property of holding the way
in which robots are scattered, where it is more prof-
itable for robots to be scattered in a work field for
passing targets through throwing.
The structure of the balance of this paper is as fol-
lows. In the second section we describe the back-
ground. The third section describes an example of
intelligent robot systems in which robots search mul-
tiple target objects and transport them cooperatively.
In our robot control system, the mobility that the mo-
bile agent system provides is the key feature that sup-
presses energy consumption; therefore the details of
mobile agent based strategies are shown here. In the
fourth section, we demonstrate how the properties of
mobile agents can contribute to saving energy con-
sumption through numerical experiments based on a
simulator. Finally, we conclude in the fifth section.
2 BACKGROUND
Multi-agent robotic systems are recently becoming
popular (Yasuda and Ohkura, 2011). In traditional
multi-agent systems, robots communicate with each
other to achieve cooperative behaviors. There are
three major advantages of multi-robot systems over
single robot systems (Stone and Veloso, 2000; Yasuda
and Ohkura, 2005). The first is parallelism; a task can
be achieved by autonomous and asynchronous robots
in a system. The second is robustness; it is real-
ized through redundancy. The system can have more
robots than required for a certain task. The third is
scalability; a robot can be added to or removed from
the system easily.
Making multiple robots cooperatively carry and
push common objects has been intensively studied
and yet to established the standard way. Many re-
search projects have dealt with this topic but few of
them have demonstrated on physical multi-robot sys-
tems. One of the most demonstrated tasks involving
cooperative transport is the pushing objects by multi-
ple robot teams (Rus et al., 1995; Stilwell and Bay,
1993). This task is inherently easy to accomplish
when comparing to carrying tasks, because a carry-
ing task involves multiple robots’ gripping a common
object and navigating to a destination in a coordinated
fashion (Khatib et al., 1996; Wang et al., 2000).
One of the most famous transportation problems
is the box-pushing problem (Mataric et al., 1995).
The problem is defined in (Gerkey and Mataric, 2002)
and consists of cooperatively moving a box, which
is relatively large when compared to the size of the
multi-robots, from an initial position to a destination
using robots that can only perform pushing move-
ments. Then the fundamental research problem is to
design the appropriate control mechanism in order to
achieve the desired global behavior by the multiple
robots’ cooperation. This problem is known as NP-
hard (Reif, 1979), and many research efforts are fo-
cused into planning. Our research objective is saving
energy by much rougher behaviors of multiple robots
but more intelligent behaviors through multiple mo-
bile software agents.
3 ROBOT CONTROLLER
AGENTS
The target recovery process consists of two tasks; one
is the task for searching targets, and the other is for
transporting them to collection areas. They are per-
formed sequentially and exclusively. In our system,
these tasks are allocated to different software mo-
bile agents. They are Searcher agent and Transporter
agent, respectively. In this section, we describe details
of each mobile agent, and show how they contribute
to suppressing energy consumption in a multi-robots
system.
3.1 Mobile Agents for Searching
Considering how to make multi-robots search a tar-
get. The simplest solution would be to make all robots
move in order to search for the target. It means, how-
ever, only one robot that ultimately finds the target
do the job, and the other robots waste their energy in
vain. In addition, even if one finds the target, the oth-
ers may not be able to know the fact because of being
out of the range where communications through wire-
less LAN are available. As the result, they may move
around until they are exhausted to move.
We have proposed the mobile agent approach for
searching targets and showed its effectiveness in the
previous paper (Abe et al., 2011). The Searcher agent
visits each robot through migration. When the agent
reaches a robot, it checks the camera on the robot in
order to find whether it is close to the target or not,
where it would have to spin within 360 degrees to
check all around it if the robot has no special cam-
era such as an omnidirectional sensor. It repeats this
process until it reaches the robot closest to the target,
and finally, it drives that robot to capture the target.
The searching strategy suppresses rotating wheels,
and therefore, contributes to saving energy consump-
tion. We have also adopted the same approach to
the target recovery process. The only difference is
that the recovery agent has a vector value that points
to the collection area as an interior datum, which is
ICAART2013-InternationalConferenceonAgentsandArtificialIntelligence
220
subsequently used to guide the agent for transporting
the target. We assume that the Searcher agent know
where the collection area is. The simplest implemen-
tation would be to make Searcher agents born at their
own collection areas. Each Searcher agent has a vec-
tor to keep track of its collection area. In the search-
ing process, each Searcher agent’s vector value is dy-
namically modified so as to pointing to the destina-
tion. When the robot, which is driven by the Searcher
agent, physically moves, the vector value is adjusted
according to the physical motion. Also whenever the
Searcher agent migrates to another robot, the vector
value has to be modified with respect to the new po-
sition. The modification is achieved by synthesizing
the vector value at the previous robot and the vector
value from the source to destination in the migration.
To achieve this synthesis, it is required that each robot
can locate each other.
3.2 Mobile Agents for Transporting
Once the Searcher agent finds the target object, the
agent has to transport the object to the collection area.
In this subsection we describe a new transporting al-
gorithm that saves energy consumption using a soft-
ware mobile agent. The basic idea is to make the robot
throw the target in the direction of the destination in-
stead of carrying it, where the direction of the desti-
nation is passed from Searcher agent as vector value.
In the simplest case, the collection area would corre-
spond to the location where the Searcher agent was
generated as mentioned above. Assuming such vector
value, all that robots should do is only to check the
directional criteria shared with the other robots. Such
criteria could be given by a simple device such as a
compass. The algorithm for the Transporter agent is
as follows: 1) Transporter makes the robot pick up the
target, and then, 2) throwsthe target to the destination.
3) If there is not any other robot in the direction of the
destination, Transporter agent makes the robot carry
it by itself.
In the second step, there are two strategies of
ThrowStraight and DirectPass as a manner where sev-
eral robots transport the target to the collection area
through cooperatively passing it.
3.2.1 ThrowStraight Strategy
The simplest strategy for passing a target is that the
agent makes the robot throw the target to the des-
tination as further as possible, and then, makes an-
other robot that is closest to the thrown target pick it
up. Once such a closest-to-the-target robot is found,
ThrowStraight agent migrates to that robot to receive
the target. This behavior makes the newly arrived
robot approach to the target in order to pick up and
throw the target again. The ThrowStraight agent can
transport the target to the destination by repeating
these steps.
On the other hand, if ThrowStraight agent cannot
find any robot close to the target, the robot has to carry
the target by itself. ThrowStraight agent throws the
target to the destination regardless whether the closest
robot is found or not. After that, if such a closest-to-
the-target robot cannot be found, all what the agent
should do is to make the robot approach to the tar-
get along thrown path. Notice here that the behaviors
of throwing and approaching correspond to carrying
the target with it to the location where the target is
thrown.
robot 1
robot 2
mobile agent
(a) Throwing a target to the des-
tination.
α
(b) Calculating the difference be-
tween the directions of another
robot and the destination.
β
(c) Calculating the difference be-
tween the directions of the previ-
ous robot and the target at the cur-
rent robot after a migration.
(d) Approaching to the target.
θ
(e) Facing with the destination.
Figure 1: The migration steps of ThrowStraight.
These steps of the ThrowStraight strategy seem to
work well if the direction to the destination, i.e. a col-
lection area, is always known. Initially, the direction
information is given to the ThrowStraight agent by
Searcher agent. Notice here that though the Searcher
agent has vector value to the destination, the Throw-
Straight agent holds only the angle datum of the vec-
tor value. After that, the angle information is modi-
fied according to physical motions of the robot, and
recalculated through every migration. The new angle
value after the migration can be gotten by calculat-
SuppressingEnergyConsumptionofTransportationRobotsusingMobileAgents
221
ing each angle of a triangle consisting of the current
robot, the next robot, and the target in order. Fig. 1
demonstrates the steps of the process, where a star,
multi-circles, a circle with a face, and circles with a
slit respectively represent a target, a destination, an
agent, and robots. The slits of circles show the di-
rection with which each robot faces. The steps are as
follows:
1. The agent makes the robot throw the target to the
destination as shown by Fig.1(a),
2. The agent calculates difference α between the di-
rections of the other robot and the destination as
shown by Fig.1(b).
3. The agent migrates to the next robot (robot 2).
4. The agent calculates difference β between the di-
rections of the previous robot (robot 1) and the
target as shown by Fig.1(c).
5. The agent makes the robot approach to the target
and picks it up as shown by Fig.1(d).
6. The agent makes the robot face to the destination
as shown by Fig.1(e), where the angle to be ori-
ented, θ is given by the calculation 180 (α+ β).
Notice that the thrown point in the first step has to
be within the view range of the robot. Otherwise, the
target cannot be traced in the following process.
Summarizing a sequence of behaviors of Throw-
Straight agent, an agent performs the following steps
1-5 until finding the specific destination, and then the
agent makes the robot throw a target to the destination
once it finds the destination: 1) The ThrowStraight
agent drives the robot forward until finding out other
robots within the angle of 120 degrees in the view
range, 2) the agent makes the robot throw the target
to the destination, 3) the agent migrates to the closest
robot to the target, 4) the agent drives the robot, to
which the agent has migrated, toward the target and
makes the robot pick it up, and 5) the agent calculates
the modified direction and makes the robot orient to-
ward the destination.
3.2.2 DirectPass Strategy
More smart strategy, which we call DirectPass strat-
egy, is to directly pass a target to another robot closer
to a destination. In more detail, the thrower passes
the target to the receiver only if the receiver is within
120 degrees range of the direction from thrower to
the destination. In this strategy, the step for searching
the next robot in ThrowStraight strategy is not nec-
essary because the current robot throw the target to
the next robot, instead of throwing to the direction of
the destination. What DirectPass agent does is simply
migrating to the receiver.
1
v
2
v
(a)
21
v
(b)
Figure 2: Passing process of DirectPass agent.
Thus, DirectPass strategy seems to be simpler and
more direct than ThrowStraight strategy. In order to
keep the direction to the destination, however, the
agent has to have complex data such as a vector value,
though the ThrowStrainght agent needs to have only
the angle as an interior datum. The vector value has to
be not only modified according to the physical motion
of a robot, but also recalculated through migrations.
Fig. 2 shows the process of calculating new vector
value after a migration. The steps are as follows:
1. The agent calculates vector value v
2
from the
current robot to the next robot as shown by
Fig.2(a), where the initial vector value is passed
by Searcher agent.
2. The agent modifies the interior datum v
1
with new
vector value given by calculating v
1
v
2
.
3. The agent migrates to the next robot as shown in
Fig.2(b).
Summarizing a sequence of behaviors of Direct-
Pass agent, an agent performs the following steps 1-
5 until finding the specific destination, and then the
agent makes the robot throw a target to the destina-
tion once it finds the destination: 1) The DirectPass
agent drives the robot forward until finding out other
robots within angle of 120 degrees in the view range.
2) the agent makes the robot throw the target to the
receiver, 3) the agent calculates new vector value as
interior datum, 4) the agent migrates to the receiver
robot, and 5) the agent makes the robot pick up the
target and orient toward the destination.
4 EXPERIMENTAL RESULTS
In order to demonstrate the effectiveness of our sys-
tem, we have conducted numerical experiments on the
example of target transportation that we have just dis-
cussed in the previous section. We have implemented
the mobile agent system simulator as shown in Fig. 3.
The simulator is based on hierarchical mobile agent
(Kambayashi and Takimoto, 2005; Satoh, 2000), of
which migration manner is based on the strong migra-
tion model (Cugola et al., 1997). We have evaluated
our two strategies i.e. ThrowStraight and DirectPass,
ICAART2013-InternationalConferenceonAgentsandArtificialIntelligence
222
0.4
0.6
0.8
1
The ratio of
moving distance / linear distance
R=50
R=100
R=150
R=200
0
0.2
5
10
15
20
25
moving distance / linear distance
The number of robots
R=200
R=250
(a) Results for ThrowStraightAgent
0.4
0.6
0.8
1
The ratio of
moving distance / linear distance
R=50
R=100
R=150
R=200
0
0.2
5
10
15
20
25
moving distance / linear distance
The number of robots
R=200
R=250
(b) Results for DirectPassAgent
Figure 4: The influences of the number of robots to moving distance.
(a) A snapshot of a running
simulator with a single target.
(b) A snapshot of a running
simulator with multi robots and
multi targets.
Figure 3: A snapshot of a running simulator.
in terms of moving distance and how robots are lo-
cated after the target is collected on the simulator. We
assume that energy consumption of a robot is propor-
tional to the moving distance because most energy is
consumed for rotating wheels.
4.1 Energy Consumption
First, we have conducted several experiments on the
sequential process of searching and transporting of a
target. We have repeated a number of the same pro-
cesses with varying the number of targets and the ra-
dius of the view range of each robot. Figs.4(a) and (b)
show ratio of the total moving distance in the trans-
portation process to the direct distance from the start
point of the transportation to the destination. The five
line charts represent the results for different radiuses
of the view ranges: 50, 100, 150, 200 and 250, respec-
tively. Each chart consists of results for 5, 10, 15, 20
and 25 robots. As shown by the shapes of the charts,
the more the number of the robot and/or the radius
of the view range increased, the more the total mov-
ing distance decreased in both cases of ThrowStraight
and DirectPass strategies.
4.2 Arrangements of Robots
In both of our transporting approaches, the more
widely robots are scattered, the more advantages they
seem to have. In order to confirm this observation,
we have evaluated the area size covered by the view
ranges of all robots during the experiments as a crite-
rion of the way in which robots are scattered. In the
experiments, we assumed several targets, and sequen-
tial executions of searches and transportations such as
SA
1
(a search agent) TA
1
(a transport agent) SA
2
TA
2
SA
3
SA
4
TA
3
TA
4
· · · , TA
n
is
created after SA
n
.
Table 1 shows the ratio of the area size covered
by the view ranges in DirectPass strategy to that in
ThrowStraightAgent, for 5, 15 and 25 robots, for 10,
30 and 50 targets and for 50, 150 and 250 view dis-
tance respectively. Though DirectPass strategy are
more widely scattered than ThrowStraightAgent, the
difference between them is negligible. In other words,
the result shows that our approaches have little differ-
ence in influence on other tasks of the multi-robots.
5 CONCLUSIONS
We have implemented a multi-robot system that trans-
ports a target to a destination while suppressing en-
ergy consumption, by adding transport facility to the
previous multi-robot system for searching (Abe et al.,
2011). We have proposed two strategies Throw-
Straight and DirectPass as transport facility. Both
of them can suppress more energy consumption than
carrying directly to destination, but DirectPass has re-
markable advantages. DirectPass, however, requires
more hardware equipment than ThrowStraight. Di-
rectPass has to get both the directional angle and the
distance toward other robots through robot’s sensor
SuppressingEnergyConsumptionofTransportationRobotsusingMobileAgents
223
Table 1: Ratio sum of each robot’s view area: DirectPass / ThrowStraight.
target = 10 target = 30 target = 50
View distance The number of robots
5 15 25 5 15 25 5 15 25
R = 50 1.055 0.997 1.007 1.005 1.005 1.018 1.023 1.045 1.040
R = 150 1.056 1.041 1.016 1.064 1.094 1.049 1.094 1.114 1.094
R = 250 1.030 1.034 1.014 1.099 1.092 1.048 1.100 1.142 1.100
devices. ThrowStraight needs only the directional an-
gle. Thus, when implementing the multi-robots sys-
tem based on our approach, it would be necessary
to decide which strategy should be adopted based on
quality of the sensors equipped with the robots.
REFERENCES
Abe, T., Takimoto, M., and Kambayashi, Y. (2011). Search-
ing targets using mobile agents in a large scale multi-
robot environment. In KES-AMSTA, volume 6682 of
LNAI, pages 211–220.
Binder, W., Hulaas, J. G., and Villaz, A. (2001). Portable
resource control in the j-seal2 mobile agent system.
In Proceedings of the fifth international conference on
Autonomous agents, AGENTS ’01, pages 222–223.
ACM.
Cugola, G., Ghezzi, C., Picco, G. P., and Vigna, G. (1997).
Analyzing mobile code languages. In Selected Pre-
sentations and Invited Papers Second International
Workshop on Mobile Object Systems - Towards the
Programmable Internet, MOS ’96, pages 93–110.
Springer-Verlag.
Gerkey, B. P. and Mataric, M. J. (2002). Pusher-watcher:
An approach to fault-tolerant tightly-coupled robot
coordination. In Proceedings of the IEEE Interna-
tional Conference on Robotics and Automation 1,
pages 464–469.
Kambayashi, Y. and Takimoto, M. (2005). Higher-order
mobile agents for controlling intelligent robots. Inter-
national Journal of Intelligent Information Technolo-
gies (IJIIT), 1(2):28–42.
Khatib, O., Yokoi, K., Chang, K., Ruspini, D., Holmberg,
R., and Casal, A. (1996). Vehicle/arm coordination
and mobile manipulator decentralized cooperation. In
Proceedings of the IEEE/RSJ International Confer-
ence on Intelligent Robots and Systems, pages 546–
553.
Mataric, M. J., Nilsson, M., and Simsarian, K. T. (1995).
Cooperative multi-robot box-pushing. In Proceedings
of the IEEE/RSJ International Conference on Intelli-
gent Robots and Systems 3, pages 556–561.
Nagata, T., Takimoto, M., and Kambayashi, Y. (2009). Sup-
pressing the total costs of executing tasks using mobile
agents. In Proceedings of Hawaii International Con-
ference on System Sciences 42 CD-ROM.
Reif, J. H. (1979). Complexity of the mover’s problem
and generalizations. In Proceedings of the 20th An-
nual Symposium on Foundations of Computer Sci-
ence, pages 421–427. IEEE Computer Society.
Rus, D., Donald, B., and Jennings, J. (1995). Moving furni-
ture with teams of autonomous robots. Proceedings of
the IEEE/RSJ International Conference on Intelligent
Robots and Systems, pages 235–242.
Satoh, I. (2000). Mobilespaces: A framework for building
adaptive distributed applications using a hierarchical
mobile agent system. In Proceedings of the The 20th
International Conference on Distributed Computing
Systems ( ICDCS 2000), ICDCS ’00, pages 161–168.
IEEE Computer Society.
Stilwell, D. J. and Bay, J. S. (1993). Toward the develop-
ment of a material transport system using swarms of
ant-like robots. In Proceedings of the IEEE Interna-
tional Conference on Robotics and Automation, pages
766–771.
Stone, P. and Veloso, M. (2000). Multiagent systems: A sur-
vey from a machine learning perspective. Autonomous
Robots, 8(3):345–383.
Takimoto, M., Mizuno, M., Kurio, M., and Kambayashi, Y.
(2007). Saving energy consumption of multi-robots
using higher-order mobile agents. In KES-AMSTA,
volume 4496 of LNAI, pages 549–558.
Wang, Z. D., Kimura, Y., Takahashi, T., and Nakano,
E. (2000). A control method of a multiple non-
holonomic robot system for cooperative object trans-
portation. In Proceedings of the 5th International
Symposium on Distributed Autonomous Robotic Sys-
tems on Distributed Autonomous Robotic Systems 4,
pages 447–456.
Yasuda, T. and Ohkura, K. (2005). Autonomous role assign-
ment in a homogeneous multi-robot systems. Journal
of Robotics and Mechatronics, 17(5):596–604.
Yasuda, T. and Ohkura, K., editors (2011). Multi-Robot
Systems, Trends and Development, InTech.
ICAART2013-InternationalConferenceonAgentsandArtificialIntelligence
224