Cooperatively Transporting Unknown Objects using Mobile Agents

Ryo Takahashi

, Munehiro Takimoto

and Yasushi Kambayashi

Department of Information Sciences, Tokyo University of Science, 2641 Yamazaki, Noda 278-8510, Japan

Department of Computer and Information Engineering, Nippon Institute of Technology,

4-1 Gakuendai, Miyashiro-machi, Minamisaitama-gun 345-8501, Japan

Keywords:

Mobile Agent, Multiple Robots, Ant Colony Optimization, Swarm Intelligence.

Abstract:

This paper presents an algorithm for cooperatively transporting objects by multiple robots without any initial

knowledge. The robots are connected by communication networks, and the controlling algorithm is based on

the pheromone communication of social insects such as ants. Unlike traditional pheromone based cooperative

transportation, we have implemented the pheromone as mobile software agents that control the mobile robots

corresponding to the ants. The pheromone agent has the vector value pointing to its birth location inside,

which is used to guide a robot to the birth location. Since the pheromone agent can diffuse with migrations

between robots as well as a physical pheromone, it can attract other robots scattering in a work ﬁeld to the birth

location. Once the robot ﬁnds an object, it brieﬂy pushes the object, measuring the degree of the inclination

of the object. The robot generates a pheromone agent with the vector value to pusing point suitable for

suppressing the inclination of the object. The process of the pushes and generations of pheromone agents

enables the efﬁcient transportation of the object. We have implemented a simulator based on our algorithm,

and conducted experiments to demonstrate the feasibility of our approach.

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). Multi-agent

systems introduced modularity, reconﬁgurability and

extensibility to control systems, which had been tra-

ditionally monolithic. It has made easier the develop-

ment of control systems on distributed environments

such as multi-robot systems.

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. In a mobile agent system,

each agent can actively migrate from one site to an-

other site. Since a mobile agent can bring the nec-

essary functionalities with it and perform its tasks au-

tonomously, it can reduce the necessity for interaction

with other sites. Mobile agent systems are especially

useful in an intermittently connected ad hoc network

environment. In the minimal case, a mobile agent re-

quires that the connection is established only when it

performs 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 in Figure 1 (Kam-

bayashi and Takimoto, 2005). The property of inter-

robot movements of the mobile agents contributes to

the ﬂexible and efﬁcient use of the robot resources.

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 con-

sumption (Takimoto et al., 2007). Here, notice that

any agents on a robot can be killed as soon as they

ﬁnish their tasks. If the agent has a policy of choos-

ing idle robots rather than busy ones in addition to the

power-saving effect, it would result in more efﬁcient

use of robot resources.

We have proposed our model in the previous pa-

per (Takimoto et al., 2007) and have also shown the

effectiveness of saving power consumption and the ef-

ﬁciency of our system for searching targets (Nagata

et al., 2009; Abe et al., 2011) and transporting them

to a designated collection area (Shibuya et al., 2013).

In this paper, we focus our attention on transporta-

tion of a large object that a single robot alone cannot

Takahashi R., Takimoto M. and Kambayashi Y..

Cooperatively Transporting Unknown Objects using Mobile Agents.

DOI: 10.5220/0004816400600068

In Proceedings of the 6th International Conference on Agents and Artiﬁcial Intelligence (ICAART-2014), pages 60-68

ISBN: 978-989-758-016-1

 2014 SCITEPRESS (Science and Technology Publications, Lda.)

Figure 1: A team of mobile robots are working under control of mobile agents.

move. In order to deal with such a large object, sev-

eral robots have to cooperate to achieve the objective

tasks, which seems to require too artiﬁcial behaviors

for each robot. For the cooperatively solving of such

problem, swarm-based approaches, which are based

on the social insect metaphor, have been proposed.

The swarm-based system, which consists of several

robots with simple behaviors, can achieve complex

tasks, just like ants that behave based on simple rules

cooperatively transport a large prey. The swarm of

simple robots makes a system more ﬂexible and fault-

tolerant, and contributes to suppressing costs of build-

ing a large complex system.

We have implemented the ants as actual mo-

bile software agents that control the mobile robots.

The ant agent migrates among robots to look for an

available one. Once the ant agent ﬁnds the avail-

able robot, the ant agent physically drives the robot.

We also implemented pheromone as mobile software

agents, which attract many robots to the large object

to convey through diffusing with the migrations. The

pheromone agent has a vector value pointing to the

birth point. The vector value is modiﬁed in order to al-

ways point to the birth point, each that the pheromone

agent migrates to another robot. Once the pheromone

agent migrates to the robot where an ant agent resides,

it guides the ant agent to lead the robot to its birth

point. Thus, the pheromone agents enable scattering

free robots to efﬁciently attend to the transportation of

the object. In our new approach, we take advantage

of the pheromone agents not only to collect robots,

but also to suppress the rotation of an object during

the transportation. We assume that each robot has a

simple sensor for checking the degree of the inclina-

tion of the object that it is pushing. When each robot

slightly pushes the object, it generates pheromones at

the suitable location to suppress the increase of the in-

clination of the object. As a result, the object can be

linearly transported to a target area, even if the object

has the form such as a stick. Notice that such an ob-

ject could not be transported without suppressing the

rolling.

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 our transport

model based on pheromone agents. In the fourth sec-

tion, we the numerical experiments using a simula-

tor based on our algorithm. Finally, we conclude our

discussions in the ﬁfth section and present future re-

search directions.

2 BACKGROUND

Making multiple robots cooperatively carry and push

common objects has been intensively studied and

yet to established the standard way. Many research

projects have dealt this topic but few of them have

demonstrated on physical multi-robot systems. One

of the most demonstrated tasks involving coopera-

tive transport is the pushing objects by multiple robot

teams (Rus et al., 1995; Stilwell and Bay, 1993).

This task is inherently easy to accomplish when com-

paring to carrying tasks, because a carrying task in-

volves 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

CooperativelyTransportingUnknownObjectsusingMobileAgents

problem is deﬁned in (Gerkey and Mataric, 2002) and

consists of cooperatively moving a box, which is rel-

atively large when compared to the size of the multi-

robots, from an initial position to a destination using

robots that can only perform pushing movements.

On the other hand, algorithms that are inspired

by behaviors of social insects such as ants to com-

municate to each other by an indirect communication

called stigmergy are becoming popular. (Stilwell and

Bay, 1993; Dorigo and Gambardella, 1996; Dorigo

et al., 2006). Upon observing real ants’ behaviors,

Dorigo et al. found that ants exchanged information

by laying down a trail of a chemical substance (called

pheromone) that is followed by other ants. They

adopted this ant strategy, known as ant colony opti-

mization (ACO), to solve various optimization prob-

lems such as the traveling salesman problem (TSP)

(Dorigo and Gambardella, 1996). Deneubourg has

originally formulated the biology inspired behavioral

algorithm that simulates the ant corps gathering and

brood sorting behaviors (Deneubourg et al., 1991).

Wang and Zhang proposed an ant inspired approach

along this line of research that sorts objects with mul-

tiple robots (Wand and Zhang, 2004). Lumer has

improved Deneubourg’s model and proposed a new

simulation model that is called Ant Colony Cluster-

ing (Lumer and Faiesta, 1994). His method could

cluster similar objects into a few groups. Mizutani

et al. has proposed to implement pheromones as mo-

bile agents in ACO (Mizutani et al., 2010). In their

system, ants are also mobile agents, that repeatedly

migrate to robots to searching free robots correspond-

ing to objects. Once the ant agent ﬁnd a free robot, it

drives the robot to a cluster of robots. The pheromone

agent is generated by the ant agent when its driving

robot reaches a cluster, and repeatedly migrates to

robots to guide ant agents to drive robots to the cluster.

The mobile agent based algorithm has been improved

to serialize collected robots (Shintani et al., 2011b;

Shintani et al., 2011a)

The applications of ants’ behaviors to coopera-

tive transportation has been proposed by Kube and

Bonabeau (Kube and Bonabeau, 2000). It transports

an object through the interplay of forces. The robots

can recognize an object through the light emitted

from the object, which works as the pheromone, but

its effectiveness is restrictive. Fujisawa et al. have

proposed the efﬁcient cooperative transportation us-

ing ethanol as physical pheromone (Fujisawa et al.,

2010).

3 TRANSPORT MODEL

We assume that each robot just affects the transported

object through pushing it, and has to simultaneously

push it together while other robots to move it. Also,

since we assume that our multiple robots system is

used for several tasks (Nagata et al., 2009), the robots

available for the transportation are just ones not en-

gaging other task.

In order to achieve the transportation over the

robots shared with several tasks, we introduce two

kinds of agents, which are called Ant Agents (AA)

and Pheromone Agents (PA). The AA has a role for

searching a robot with no other task, and driving a

robot to the object to transport. The PA has a role

for efﬁciently attracting the free robots to the object

by guiding AAs on robots instead of directly driving

them. In this section, we describe the robot’s func-

tionalities required in our system, and the details of

two agents.

3.1 Robots

We assume mobile robots such as PIONEER 3-DX,

which have two servo-motors with tires, one cam-

era and sixteen sonic sensors (Nagata et al., 2009).

The power is supplied by rechargeable battery. The

PIONEER 3-DX has a servo-motor and sensor con-

troller board that sends/receives data to/from a host

computer on it through a USB cable. The camera is

directly operated by the host. Each robot holds a lap-

top computer as a host, which is also used as a server

for the migration of our control agents through the

wireless LAN.

Also, in our transportation system, we assume that

each robot has additional features as follows:

• it can push an object,

• it has the sensor for checking the degree of the

inclination of the pushing object,

• it can always know the direction to which the ob-

ject should be transported, and

• it knows the IP addresses of other robots by iden-

tifying them through the camera.

All functionalities for controlling the robot are in-

troduced into each robot through migrations of mobile

agents. Once an agent migrates to a host, the agent

can communicate with other agents on the same host,

so that the user can construct a larger system by mi-

grations to the host.

ICAART2014-InternationalConferenceonAgentsandArtificialIntelligence

Figure 2: State transition diagram of Ant agent.

3.2 Ant Agent

AA has the list of IP addresses of all the robots in

order to traverse them one by one and to ﬁnd out a

free robot. If it has exhausted all robots, it goes back

to the home host for administration of the robots, and

updates its IP address list in the following cases: 1)

some new robots have been added, and 2) some robots

have been broken, where notice that the addition or

the removal of a robot is recorded on the home host.

However, these are so rare that the home host is made

to go through in most cases.

Once an AA reaches a free robot, the AA controls

the robot along the state transition as shown in Figure

First, the AA drives the robot to the object to trans-

port through two kinds of manners. Initially, the AA

makes the robot randomly walk to search the object as

shown by state Search. Once PA migrates to the same

robot, the AA drives the robot to the object along the

guidance of the PA as shown by state Guided. Af-

ter that, once the AA ﬁnds the object, it prepares for

transporting the object as shown by state Prepare.

Actual transportation is performed by repetition of

brieﬂy pushing the object as shown by state Push

and attracting other robots to the suitable location for

balancing the object as shown by state Attract. The

transportation task of each robot is limited in the spe-

ciﬁc time, and therefore, after the time-out, the robot

is released from the task as shown by state Leave. The

details of each state is as follows:

State Search. In this state, AA drives the robot at

random as shown in Figure 3. If the robot contacts

with other robots, it temporarily stop there, and then

take an action for avoidance. Once the robot bumps

onto the object to transport, the state of the AA is

changed to the state Prepare as shown by T3 in Figure

2. On the other hand, if a PA migrates to the robot, the

Figure 3: State Search of

AA.

Figure 4: State Guided of

AA.

Figure 5: State Prepare of AA.

state is changed to the state Guided as shown by T1.

State Guided. In this state, AA drives the robot

along the guidance of the PA that has migrated to the

same robot as shown in Figure 4. The PA has the vec-

tor value pointing to the birth point of the PA, which

is adjusted to always point to the birth point even if

the robot moves or rotates. Since the PA can observe

the rotations of wheels, the vector value is modiﬁed to

neutralize the change caused by the rotation or the ap-

proach to the birth point of the robot. AA is guided for

driving the robot in the direction of the vector value.

Once the robot reaches the destination of the vector,

the AA checks whether there are some objects around

the robot or not. If some objects are found, the AA

changes its state to the state Prepare for transporting

the closest object as shown by T4; otherwise, the AA

abandons the PA guiding it, and then changes its state

to the state Search as shown by T2. At this time, the

PA is killed.

State Prepare. Once an AA ﬁnds an object, the AA

makes the robot approach to the object till contacting

with it. After that, the AA measures the normal vec-

tor V

against the surface of the object to know the

CooperativelyTransportingUnknownObjectsusingMobileAgents

Figure 6: State Attract of AA.

degree of the inclination, as shown in Figure 5. Fur-

thermore, since the AA always holds vector V

to its

nest to which the object is transported, it can calcu-

late θ that is the angle between V

and V

. The agents

including AA have the vector values in polar coordi-

nate form. Therefore, using the angle θ

of V

and θ

of V

, θ is straightforwardly calculated as follows:

θ = θ

− θ

(1)

If the θ is more than −

and less than

, the AA

judges that robot has a suitable direction for efﬁcient

transportation, and changes its state to Push in order

to start the transportation of the object as shown by

T 6. Otherwise, the AA changes its state to Leave in

order to abandon the transportation, as shown by T 5.

State Push. In this state, the AA just brieﬂy pushes

the object, and then changes its state to Attract in

order to check the change of θ as shown by T 8,

and to attract another robot to the suitable location

for balancing the object. Also, the AA releases the

robot from the transportation task after speciﬁc time.

The time-out functionality enables the ﬁnite robot re-

sources to be reused for pushing another location, an-

other object or another task, and it is achieved by

changing its state to Leave, as shown by T 9.

State Attract. Figure 6 shows the relation of a robot

and an object after the robot pushed the object in the

short period in the state Push, where V

may have

change. Therefore AA calculates θ

to check the an-

gle θ

of new V

as follows:

= θ

− θ

If θ

is negative, it means that the object have rotated

to the right. In this case, pushing the object at the

neighbor location of the robot in the right hand side

may suppress the rotation. Conversely, in the case

where θ

is positive, pushing the object in the left

hand side of the robot may suppress the rotation to

the left. We call the location to push in order to bal-

ance the object additional push point. Once the AA

knows the additional push point, it generates PA with

the vector value pointing to the push point in order

to attract another robot. After that, the PA repeatedly

migrates some robots to ﬁnd a free robot to attract.

Notice that the AA generates no PA when θ

is cor-

rectly 0.

The AA generates some PAs in the predeﬁned

time, and then changes its state back to the Push, as

shown by T 7. The several transitions between the

Push and the Attract achieves the transportation of the

object in balance.

State Leave. In this state, the AA makes the robot

turn to the back, and then, move it in the speciﬁc

distance in order to release it from the transportation

task.

3.3 Pheromone Agent (PA)

The PA has behaviors imitating the physical

pheromone of ants, of which the main role is to attract

the ants to preys or their nest. In a PA, the attraction is

implemented as a guidance functionality based on the

vector value to the birth point. As well as diffusion of

physical pheromones, the PA searches the free robot

to guide in two steps: generation and propagation of

PA. First, in the generation step, the vector value to

the birth point is set to the PA, and then, in the prop-

agation step, it starts migrating to other robots. Fi-

nally, it is killed by the abandonment of AA. Also,

when PA migrates to the robot on which another PA

resides, they are composed in the composition step. In

the remainder of the section, we describe the details

of these steps.

Generation. The PA is generated by AA, of which

the vector value is initialized to point to the neighbor

location of the robot on which the AA resides. For

example, when AA founds that the object rotates to

the right in the state Attract, it generates a PA with

the vector value as shown by V

in Figure 7, where V

is the normal vector against the object. Conversely,

if the object rotates to the left, the vector is initial-

ized as shown by V

. The initial vector has the length

corresponding with the size of a robot and 90 degrees

against V

Propagation. The purpose of the PA is to ﬁnd a free

robot, and then to lead it to the destination of the vec-

tor value. Therefore, as soon as it is generated or

arrives at another robot, it checks around the current

robot, so that if some robots are found, it generates its

clone and makes the clone migrate to the closest one

of them. The migration to the closest robot enables

ICAART2014-InternationalConferenceonAgentsandArtificialIntelligence

Figure 9: The snapshot image of the simulator.

Figure 7: Generation of PA.

Figure 8: Propagation of PA.

the moving distance of the robot guided by the PA to

be shorter. Notice here that in a migration process,

the destination of the vector value of new PA after the

migration has to be same as one of the old PA before

the migration, which is achieved by synthesizing the

old vector value V

old

and the vector value V

from the

destination to source of the migration as follows:

new

= V

old

+ V p

Figure 8 shows the ﬁrst migration of PA after its gen-

eration. In this case, V

new

is set to V

+ V

As mentioned above, a PA generates its clone,

which migrates to another robot. The number of the

generation of a clone is limited to ﬁve. A clone is

generated one at a time. Also, each PA is ﬁnally aban-

doned by the AA in its state Leave, and therefore, the

PA does not continue giving the old information to the

system.

Composition. Once the PA migrates to the robot on

which another PA resides, they have to be composed.

In the case of physical pheromone, The effectiveness

of the attraction is strengthened by the composition,

whereas in PAs, either of them is alternatively se-

lected. All the vector values of them are checked, so

that the PA with shortest vector is selected.

This composition strategy enables decreasing the

total distance that all robots move, so that it con-

tributes to suppressing energy consumption of the en-

tire robot system. In addition, the strategy also con-

tributes to keeping the degree of the current disper-

sion, because it is proﬁtable for robots to uniformly

scatter considering the case where several objects are

simultaneously transported.

4 EXPERIMENTAL RESULTS

In order to demonstrate the effectiveness of our trans-

portation method,we have implemented a simulator.

Figure 9 shows the output image of the simulator,

through which we can observe the behaviors of agents

and robots. Numerical statistic data are recorded apart

from it.

Each small circle shows a robot, the moving di-

rection of which is shown by a triangle inside it. Each

robot is assumed to move from a grid to another grid

CooperativelyTransportingUnknownObjectsusingMobileAgents

Figure 10: Transportation time.

Figure 11: Transportation distance.

in 500 × 500 grids square ﬁeld. Also, the ﬁeld is sur-

rounded by a wall, and therefore, robots cannot go

out the ﬁeld. A big red circle with a robot at its center

shows the view range of the camera equipped on the

robot, and it is also the range where AA or PA can

migrate. In the ﬁgure, the object to transport is shown

by an equilateral triangle with sides 240 grids long.

In order to conﬁrm the accuracy and efﬁciency

of our method, we have conducted eight experiments

through applying two kinds of strategies, which are

ones with PA and with no PA, to four kinds of objects

with the shapes of triangle, square, rectangle, and cir-

cle. For each experiment, we have measured the total

time and the total distance for transporting an object ,

and the total angle where the object rotates.

Figure 10 through 12 show the results of the trans-

Figure 12: Rotation angle.

portation time, the transportation distance and the to-

tal angle of the rotation of the object, respectively. In

all ﬁgures, the horizontal axes indicate the number of

AA. The vertical axis of Figure 10 indicates the num-

ber of the execution steps of the simulator as the trans-

portation time. One of Figure 11 shows the number

of moving grids as the transportation distance. One

of Figure 12 is the total angle in the transportation.

In both ﬁgures, the thin line shows the results without

PA, and the dark lines shows the result with PA.

Figure 10 shows the transportation time decreases

as AAs increases. That just means that the more

robots attend the transportation task, the more ef-

ﬁciently the task is performed, though the slop of

the decrease becomes close to a ﬂat around 70 AAs.

Comparing the results with no PA and one with PA,

the transportation with PA is found to be much more

efﬁcient. Besides, as observed from the fact that the

efﬁciency of 30 AAs with PA is same as one of 70

AAs with no PA, the result shows that our method has

effectiveness for suppressing the required number of

robots.

The result of the transportation distance shows

that the distance of the strategy with PA is less than

with no PA for all object shapes, and furthermore, the

distance of the no PA strategy increases as AAs in-

crease except for just the square, while the PA strat-

egy decreases for all the shapes, as shown in Figure

11.

That means PA contributes to suppressing energy

consumption in the transportation task.

The result of Figure 12 shows that the rotation

angle decreases as AA increases for both strategies.

That is because the object with the symmetry shape

naturally balances when a lot of robots uniformly con-

ICAART2014-InternationalConferenceonAgentsandArtificialIntelligence

tacts with one side. Even if such the fact is consid-

ered, the effectiveness suppressing the rotation in pro-

portion to the number of AAs is remarkable. How-

ever, the degradation around 15 AAs is one of issues

to solve. The degradation is marked for the rect-

angle, which is likely to increase for objects with a

shape such as a stick. It is derived from the gener-

ation manner of PA. Since AA generates PA with the

vector value to the neighbor location of it, the impact

of each attraction is not so strong in the case of fewer

AAs. In order to solve the issue, it would be effec-

tive to determine the destination of PA’s vector value

in proportion to the degree of the inclination occurred

in pushing the object.

5 CONCLUSIONS

We have proposed an effective transportation method

for multiple robots using mobile agents that imitate

social insects. The transportation method enables

robots scattering in a work ﬁled to cooperate for push-

ing an object in balance, which suppresses the rota-

tion of the object, so that it contributes to efﬁcient

transportation and suppressing energy consumption.

In order to show the effectiveness of our transporta-

tion method, we have implemented a simulator, on

which we have conducted some experiments. As a

result, in most cases, our method shows remarkable

effectiveness for the transportation task.

On the other hand, in some case where the num-

ber of Ant agents is small, the effectiveness of our

method is restrictive. In order to solve this problem,

the destination for Pheromone agents to guide would

need to be determined depending on the degree of the

rotation.

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 Villazon, 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.

Deneubourg, J., Goss, S., Franks, N. R., Sendova-Franks,

A. B., Detrain, C., and Chreien, L. (1991). The dy-

namics of collective sorting: Robot-like ant and ant-

like robot. In Proceedings of the First Conference

on Simulation of Adaptive Behavior: From Animals

to Animats, pages 356–363. MIT Press.

Dorigo, M., Birattari, M., and T. St

utzle (2006). Ant colony

optimization–artiﬁcial ants as a computational intel-

ligence technique. IEEE Computational Intelligence

Magazine, 1(4):28–39.

Dorigo, M. and Gambardella, L. M. (1996). Ant colony sys-

tem: a cooperative learning approach to the traveling

salesman. IEEE Transaction on Evolutionary Compu-

tation, 1(1):53–66.

Fujisawa, R., Imamura, H., and Matsuno, F. (2010).

Cooperative transportation by swarm robots us-

ing pheromone communication. In Distributed

Autonomous Robotic Systems - The 10th Inter-

national Symposium, DARS 2010, volume 83 of

Springer Tracts in Advanced Robotics, pages 559–

570. Springer.

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.

Kube, C. R. and Bonabeau, E. (2000). Cooperative trans-

port by ants and robots. Robotics and Autonomous

Systems, 30(1-2):85–101.

Lumer, E. D. and Faiesta, B. (1994). Diversity and adap-

tation in populations of clustering ants, from animals

to animats 3. In Proceedings of the 3rd International

Conference on the Simulation of Adaptive Behavior,

pages 501–508. MIT Press.

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.

Mizutani, M., Takimoto, M., and Kambayashi, Y. (2010).

Ant colony clustering using mobile agents as ants and

pheromone. In Proceedings of the Second Interna-

tional Conference on Applications of Intelligent Sys-

tems, pages 435–444. Lecture Notes in Computer Sci-

ence 5990, Springer-Verlag.

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.

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.

Shibuya, R., Takimoto, M., and Kambayashi, Y. (2013).

Suppressing energy consumption of transportation

robots using mobile agents. In Proceedings of the

5th International Conference on Agents and Arti-

ﬁcial Intelligence (ICAART 2013), pages 219–224.

SciTePress.

CooperativelyTransportingUnknownObjectsusingMobileAgents

Shintani, M., Lee, S., Takimoto, M., and Kambayashi, Y.

(2011a). A serialization algorithm for mobile robots

using mobile agents with distributed ant colony clus-

tering. In Knowledge-Based and Intelligent Infor-

mation and Engineering Systems - 15th International

Conference, KES 2011, Proceedings, Part I, volume

6881 of Lecture Notes in Computer Science, pages

260–270. Springer.

Shintani, M., Lee, S., Takimoto, M., and Kambayashi, Y.

(2011b). Synthesizing pheromone agents for serializa-

tion in the distributed ant colony clustering. In ECTA

and FCTA 2011 - Proceedings of the International

Conference on Evolutionary Computation Theory and

Applications and the Proceedings of the International

Conference on Fuzzy Computation Theory and Appli-

cations [parts of the International Joint Conference on

Computational Intelligence IJCCI 2011], pages 220–

226. SciTePress.

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.

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.

Wand, T. and Zhang, H. (2004). Collective sorting with

multi-robot. In Proceedings of the First IEEE In-

ternational Conference on Robotics and Biomimetics,

pages 716–720.

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.

ICAART2014-InternationalConferenceonAgentsandArtificialIntelligence