Capture of Multi Intruders by Cooperative Multiple Robots using

Mobile Agents

Tadashi Shoji

, Munehiro Takimoto

and Yasushi Kambayashi

Department of Information Science, 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, Saitama 345-8510, Japan

Keywords:

Mobile Agents, Multi Robots, Intruder Capture Systems.

Abstract:

Detecting and capturing intruders are two of the most important function of multi-robot systems. In this paper,

we propose an effective approach to capture agile intruders with not-so-agile mobile robots. In general, it is

difﬁcult to drive several robots to pursue moving objects while adjusting their behaviors. We make mobile

robots cooperate using two mobile software agents that control the mobile robots. The mobile software agents

migrate from one robot to another robot that is located at much suitable positions for guiding moving robots

to the target location. This control manner contributes to not only reduction of movement cost of the mobile

robots, but also their efﬁcient movement without interference from other robots. We have implemented a

simulator, on which we demonstrated effectiveness of our control manner for intruders-capturing system.

1 INTRODUCTION

The advent of a combination of multiple robots and

multiple software agents is realizing a new stage of

cooperation of mobile robotic agents and humans.

We have engaged studies of mobile software agents,

and implemented several applications such as multi-

robot systems that cooperatively search objects (Na-

gata et al., 2013) and cooperatively compose forma-

tions (Shintani et al., 2011) (Kambayashi et al., 2019).

The persuasion is based on the ﬁnding that those sys-

tems are not only feasible but also enjoy the energy

saving and efﬁciency (Takimoto et al., 2007).

In this paper, we focus our attention on detection

and capture of multiple intruders in a building. De-

tecting and capturing intruders are two of the most

important applications of multi-robot systems. We in-

troduce a novel control system for chasing multiple

intruders into a corner in a building by multiple small

not-so-agile mobile robots. Considering the current

robotics technologies and safety requirements, it is

hard to obtain mobile robots as agile as human beings.

In order to cover the laggardness of mobile robots,

we have employed multiple small robots and mobile

software agents. We can arrange multiple small robot

sparsely distributed in a building, and make them co-

operate using mobile software agents that control mo-

bile robots. Since software agents can instantly move

from a robot to another robot, it can migrate to the

most suitably located mobile robot to capture the in-

truder. In addition to the instant migration of mobile

software agents, they can easily duplicate themselves

and simultaneously control multiple mobile robots to

achieve highly cooperative activities such as captur-

ing intruders. In this paper, we describe the fea-

tures of mobile software agents and explain how they

can make multiple mobile robots cooperate to capture

several intruders.

The rest of this paper is organized as follows. The

second section describes the background. The third

section describes the method for coordinating multi-

ple mobile robots using the mobile software agents.

Our mobile agent system consists of two kinds of mo-

bile software agents, i.e. ant agents and pheromone

agents. We describe each of them and present the

role and algorithm to implement the control system

for multiple robots’ coordination. The fourth section

describes the experiments on the simulator, and we

conclude our discussion in the ﬁfth section.

2 BACKGROUND

We have previously presented a framework for con-

trolling intelligent multiple robots connected by wire-

370

Shoji, T., Takimoto, M. and Kambayashi, Y.

Capture of Multi Intruders by Cooperative Multiple Robots using Mobile Agents.

DOI: 10.5220/0009382503700377

In Proceedings of the 12th International Conference on Agents and Artiﬁcial Intelligence (ICAART 2020) - Volume 1, pages 370-377

ISBN: 978-989-758-395-7; ISSN: 2184-433X

less communication networks (Shintani et al., 2011).

The framework makes the multiple mobile robots

indirectly cooperate one another and accomplishes

highly coordinated actions. The core of the frame-

work consists of two kinds of mobile software agents,

namely pheromone agents and ant agents. The idea of

pheromone agents is inspired by the behaviors of so-

cial insects, i.e. ants. In the algorithm, agents mimic

the behaviors of ants communicating with each other

by an indirect communication mediated by modiﬁca-

tions of the environment called stigmergy. The behav-

ioral scientists Goss et al. found that ants exchanged

information by laying down a trail of a chemical sub-

stance called pheromone that is followed by other ants

(Goss et al., 1989). Dorigo et al. abstracted the behav-

iors of ants and applied the extracted algorithm to de-

rive a quasi-optimal solution for NP-complete prob-

lems (Dorigo et al., 1996). Later, Dorigo et al. suc-

cessfully applied the idea to control and coordinate

swarm robots (Dorigo, 2005). Their SWARM-BOTS

project is today considered as the pinnacles of many

multi-robot projects. There are also studies that aim to

compose a circle formation in a two dimensional or-

thogonal coordinate system using mobile agents and

their sensors (Yang and Wang, 2019), (Song et al.,

2019).

Kambayashi et al. presented capturing single

intruder strategy using vector value between robots

(Kambayashi et al., 2020). We improved this strategy

more effectively and adapted for multiple intruders.

We aim to capture intruders using vector values that

points mobile agents on multiple robots and intruders.

3 SINGLE INTRUDER

STRATEGY

Captor robots are small mobile robots. They do not

attack an intruder violently. They quietly chase and

drive him or her into the designated capture loca-

tion. Kambayashi’s system (Kambayashi et al., 2020)

consists of robots and two kinds of mobile software

agents, namely ant agents and pheromone agents.

All the controls for the mobile robots are achieved

through ant agents. Each mobile robot has Wi-Fi ca-

pability. They are connected through wireless LAN.

Each ant agent can freely move among the herd of

mobile robots. Since the ant agents control the robots,

captor robots without ant agents just sit and sense the

environments quietly, while ant agents are hopping

from a robot to another robot to patrol the sensing area

by checking the quietly sitting robots’ sensors. We

assume the captor robots have enough sensors such

as optical cameras and ultrasonic sensors to detect an

intruder.

Once an ant agent arrives at a captor robot who

senses an intruder, it drives the robot to hunt down

the suspect and simultaneously dispatches pheromone

agents to rally nearby robots to bring the intruders to

the bay. We describe the ant agents and pheromone

agents in the following sections.

3.1 Ant Agent

The ant agent (AA) has the following six capabilities.

1) It drives a robot on which it resides depending on

the speciﬁc state they can take. 2) It can migrate from

a robot to another robot through wireless LAN. 3) It

has the IP addresses of all the participating robots. 4)

It creates pheromone agents that attract other mobile

robots. 5) It can have four states and changes its state

as the situation changes. 6) It can acts depending on

the current state.

Without AAs, mobile robots cannot even move;

they are just sitting quietly. Therefore, at the initial

phase, we make a number of AA migrate into the herd

of the mobile robots. Each AA looks for an idle robot

(a robot without AA) that it can drive. Once an AA

arrives on a robot, it is in the “search” state. After

that, AA changes its state depending on the environ-

mental conditions. The idea of changing state to make

AA perform different kinds of tasks is inspired from

(Takahashi et al., 2014).

Initially, the AA is in the “search” state. In this

state, the AA makes the robot randomly walk to ﬁnd

a suspect. The robot has basic capability for colli-

sion avoidance by ultrasonic sensors and electronic

circuit. Each robot has basic capabilities to survive by

on-board controller without AA’s interference. The

robot’s visual sensor can sense objects in the fan-

shaped area with 45 degrees to the right and left from

its driving direction. When the AA ﬁnds an intruder

in its fan-shaped sensing area through robot’s visual

sensor, it changes its state to “chase.”

In the “chase” state, the AA drives its robot to-

ward the intruder while it gives intimidation by warn-

ing sound and ﬂashing light, and tries to drive the

intruders to the designated capture location. Since

the mobile robots are not as agile as human intruder,

it cannot capture or drive the intruders into the cap-

ture location alone. It needs fellow robots to drive

the intruders in cooperation with them. In order to

obtain the cooperation from other robots that do not

sense the intruders, the chasing robot changes its state

into “attract” temporarily, and creates and dispatches

pheromone agents.

In the “attract” state, AA creates and dispatches

pheromone agents as many as it can ﬁnd other AAs in

Capture of Multi Intruders by Cooperative Multiple Robots using Mobile Agents

371

search state. A pheromone agent (PA) is the other

kind of mobile agents, and its purpose is to share

information about the intruders with other robots in

search state within the wireless communication range.

PA has a vector value that points from the robot that

creates the PA to the intruder, so that the PA can

tell another robot to which direction the robot should

move to ﬁnd the intruder. After dispatching PAs,

the AA resume to the “chase” state, and continues to

chase the intruder.

When an AA receives a PA from a chasing AA,

the AA transits to be in the “guided” state, as shown

in Figure 1. In the guided state, the AA moves fol-

lowing the vector value that is given by the arriving

PA until it ﬁnds the intruder. When AA ﬁnds the in-

truder, it changes its state from “guided” to “chase”

and starts chasing the intruder based on its own sen-

sor information. If AA cannot ﬁnd the intruder after

certain period, it changes its state to “search” again,

and starts random walk. We describe how a PA plays

its role in the next section.

Figure 1: AA’s method.

3.2 Pheromone Agent

The pheromone agent (PA) is a mobile agent that

guides AA to the destination of vector value which

it has. Figure 2 and Figure 3 show how a PA guides

and makes a searching robot chase an intruder in co-

operation with other chasing robot. AA1 ﬁnds a robot

controlled by AA2 and is randomly walking while it

is chasing an intruder. Since each robot has a visual

sensor, AA1 can calculate the intruders coordinate V

Since mobile robots are not agile enough to capture

an intruder alone, it needs the other robots’ assistance.

AA1 ﬁnds a robot, and also ﬁnds AA2 is driving the

robot. AA1 decides to make that robot participate in

the chase. AA1 creates a pheromone agent with the

vector value V

new

that points to the intruder, and dis-

patches the PA to the robot where AA2 is driving. The

vector value V

new

is calculated by the equation(1).

new

= V

− V

(1)

Upon receiving the PA, AA2 changes its state

“search” to “guided” and starts following the guid-

ance of PA. When AA2 ﬁnd an intruder, AA2 changes

its state to “chase” and then “attract”, and dispatches

PA to yet another robot as shown in Figure 4.

Figure 2: Before: PA how to attract AA.

Figure 3: After: PA how to attract AA.

Figure 4: Dispatching PA repeatedly.

4 MULTIPLE INTRUDERS

STRATEGY

Our previous method worked well for a single in-

truder. On the other hand, it did not effectively work

for multiple intruders shown in later comparative ex-

periment. Also, as the number of robots is increased,

we have observed that robots frequently collided with

each other. Since the collision, consumes a lot of steps

for recovering from it, so that the time cost to cap-

ture intruders increases. Thus, in order to achieve ef-

ﬁciently capturing multiple intruders, we had to solve

these problems. We address this problem through in-

troducing sophisticated guidance to colleague AAs

so that they can move to more suitable positions to

capture intruders, We also provide collision control

mechanism based on agent migration.

HAMT 2020 - Special Session on Human-centric Applications of Multi-agent Technologies

372

4.1 Ant Agent

In our previous approach, we often observed that an

intruder stacked in corners that are not the goal po-

sition. That is because AAs chase an intruder in a

straight way, and therefore it cannot escape once it

is driven to the corner. In order to avoid this situa-

tion, we give an AA different behaviors depending on

three areas that are determined with relative positions

of the AA and the goal. We call the areas “Inside”,

“Outside” and “Behind” as shown in Figure 5.

Figure 5: Situation with positions.

If an intruder is in “Inside” area, it means the intruder

is closer to the goal than the robot with AA, the AA

basically performs the same behaviors as our previ-

ous method. Once AA ﬁnds an intruder, it changes

its state from “search” to “chase” and “attract”. In

the “attract” state, we have improved AAs’ behavior.

They create PAs in fan-shaped formation so that they

attract other robot only in front of them. We explain

the details of this improvement in the next section.

If an intruder is in “Outside” area, it means the

intruder is far from the line between the robot with the

AA and the goal, but the intruder’s position is closer

to the goal than the AA’s robot. In this situation, the

AA creates PAs, and dispatches them to other robots

in the other side of the line between the intruder and

the goal as shown in Figures 6 and 7. In this situation,

if AA chases an intruder straightly, the intruder can

escape to a corner of the ﬁeld and stack there. On the

other hand, since other robots with AAs attracted by

our PAs chase from the corner side, we can avoid that

situation where the intruders to stack at any corners.

If an intruder is in “Behind” area, it means the in-

truder is behind of the chasing robot, the driving AA

changes its state to “Jump” to migrate to another robot

behind the intruder as shown in Figure 8. In the ﬁg-

ure, the chasing AA migrates (jumps) to a free robot

R2, which is at a more favorable position than R1.

In other words, in “Jump” state, AA looks for a free

robot, i.e. in this case R2, which is within the range of

PA, and behind the intruder, and migrates to the free

robot through wireless LAN.

At the same time, AA creates a PA, with a vector

Figure 6: Before: AA dispatches for limited robot.

Figure 7: After: AA dispatches for limited robot.

value that shows the location in accordance with fol-

lowing equation (2). In the equation, PA’s vector

is determined as enclosing the intruder with the robot

R2 with the AA and the other robot attracted by PA.

as follows:

R1R2

(2)

Once PA are created in “Jump”, the AA dispatches

PA to robots behind the intruder. At the same time,

it migrates to the robot R2. Once AA jumps to R2,

R2 with AA1 goes toward the target after checking

around R2.

If AA cannot ﬁnd any robot to which it mi-

grates, the AA just creates and dispatches PA to robots

behind the intruder, changes its state from “chase”

to “search” and gives up chasing the intruder. In

“search” state, AA drives a robot randomly. Once PA

migrates to a robot with such AA, the AA changes

its state to “guided”, where AA drives a robot along

guidance shown by a vector value of the PA.

4.2 Pheromone Agent Creation

In the improved method, we also made some adjust-

ments to the creation of Pheromone Agent. In the

previous PA creation manner, AA simply creates PA

behind the intruder. Such a creation manner works

well for a single intruder, but for multiple intruders, it

causes a lot of escapes. In our new creation manner,

the AA creates PAs in fan-shaped formation as en-

closing intruders as shown in Figures 10 and 11. The

fan-shaped creation is achieved as follows: 1) AA ro-

Capture of Multi Intruders by Cooperative Multiple Robots using Mobile Agents

373

Figure 8: Before: Jump behind the intruder.

Figure 9: After: Jump behind the intruder.

tates the robot by facing with an intruder, and 2) the

AA creates two PAs in both sides respectively with a

little incline as shown in Figures 10 and 11.

Figure 10: Before: Fan-shaped formation with PA.

Figure 11: After: Fan-shaped formation with PA.

Notice that each PA is dispatched to robots where

they are at least robot size apart. Once other AAs

drive their robots and reach the designated points

based on guide of the PAs, they start chasing the in-

truder.

4.3 Collision Avoidance

Many robots may be in the same ﬁeld. In such a sit-

uation, many collisions will occur. Such collisions

tend to disturb convergence of the system. In order to

suppress the frequent collisions, we introduce mobile

agent based collision avoidance mechanism. The col-

lision avoidance is achieved by simple migration of a

mobile agent between robots. For example, a robot

moving on a line is blocked by another robot on the

line as shown in Figure 12. In a mobile agent model,

the most important mission is making the robot that it

is driving. If the mission is to continue moving on the

line, the mobile agent on a robot can continue mov-

ing through migrating to the other robot and continue

moving as shown in Figure 13. In our system, an AA

has this migration property, which contributes to its

smooth movement as collision avoidance.

Figure 12: Before Agent Exchange.

Figure 13: After Agent Exchange.

5 EXPERIMENTAL RESULTS

In order to investigate the feasibility of our detec-

tion and capture system for multiple intruders, we

have implemented our system on a simulator and con-

ducted numerical experiments and comparative ex-

periments with our previous method. In the experi-

ments, we assume the following conditions.

• Robots are scattered in a 500 x 500 square ﬁeld in

the simulator.

HAMT 2020 - Special Session on Human-centric Applications of Multi-agent Technologies

374

• The capture location is set at one edge of the ﬁeld.

• The number of robots and AAs is selected from 5,

10, 30, 50, 75, 100 and 200.

• The number of AAs is limited to the number of

robot.

• The maximum number of intruders is 5.

• Robots and intruders are randomly placed in the

ﬁeld.

• The range of communication network for each

robot is 200 units and the range of sensor is the

same.

• Each robot can move 1 unit and each intruder can

move 1.5 units in each step in the simulator.

• In each trails, the maximum number of steps is

limited to 10,000 and the trail is counted to “miss”

trail.

We have conducted 100 trails for each number of

AAs, robots and intruders, and measured the duration

time to drive intruders to the capture location. Figure

14 shows an image of simulation with 200 robots, 75

AAs and 3 intruders.

Figure 14: Experiment Example.

Figures 15 and 16 show the results for relation of

the efﬁciency between the number of robots and the

number of intruders in ﬁxing the number of AAs to

30. The vertical axis represents the number of steps

needed to drive intruders, where each bar represents

the number of intruders driven into the capture loca-

tion. The horizontal axis is the number of participat-

ing robots. Figures 17 and 18 show the results for

the relation between the number of AAs and the num-

ber of intruders in ﬁxing the number of robots to 200,

where horizontal axis represents the number of partic-

ipating AAs. Figures 15 and 17 show the results with

the previous method, and Figures 16 and 18 show the

results with our improved method.

Figure 15: Previous: The number of Steps with 30 AAs.

Figure 16: Improved: Steps with 30 AAs.

Figure 17: Previous: The number of Steps with 200 robots.

Figure 18: Improved: The number of Steps with 200 robots.

Comparing the two methods, the improved method

shows obvious advantage regardless of the number

of robots. The advantage becomes more remarkable

as the number of intruders decreases. Also we can

observe that the advantage become more remarkable

as the number of AAs increases. The result shows

that our improvement enhances the effectiveness of

the mobile agents in our system.

Figures 19 and 20 show the result for the relation

of the efﬁciency between the number of AAs and the

number of robots in ﬁxing the number of intruders

to three. The vertical axis represents the number of

steps needed for driving the intruders into the capture

location, and horizontal axis is represents the num-

Capture of Multi Intruders by Cooperative Multiple Robots using Mobile Agents

375

ber of participating mobile agents and each bar repre-

sents the number of robots. As shown in the ﬁgures,

our system becomes more efﬁcient as the number of

AAs increases in improved method than in the pre-

vious method. As shown in Figure 21, the number

of AA exchanges increases dramatically as the num-

ber of robots increases. This phenomenon represents

exchanging agents contributes to the reduction of the

steps.

Figure 19: Previous: The number of Steps with 3 intruders.

Figure 20: Improved: The number of Steps with 3 intruders.

Figure 21: The number of Exchange with 3 intruders.

Figure 22 shows the efﬁciency of the AA’s “Jump” al-

gorithm. We can observe that the smaller the number

of AA is, the greater the advantage of jump behind of

the intruder is obtained. On the other hand, the ad-

vantage decreases as the number of AA increases.

Finally, we show the number of “miss” trails in

Figures 23 and 24. Our improved method especially

contributes to the achievement of capturing intruders

compared with previous method.

Figure 22: Left:with jump, Right:without jump.

Figure 23: Previous: The number of Miss with 3 intruders.

Figure 24: Improved: The number of Miss with 3 intruders.

6 CONCLUSIONS

We have proposed a method for chasing multiple in-

truders that is an improved version of our previous

method that chases a single intruder. We implemented

the algorithm for capturing intruders. Since human

intruders faster than robots, we have made the cap-

tor robots cooperate by using mobile agents. With the

full use of agent mobility in the cooperation, we have

successfully made mobile robots chasing intruders in

the behind them. This is achieved through changing

chasing robots by migration of driving mobile agents.

The chasing manner contributes to efﬁcient capturing.

Also, our collision management based on migration

contributes to reduce failure of capturing intruders.

In order to show the feasibility of our idea, we

have implemented a simulator of our mobile robot

system and demonstrate the effectiveness. We are go-

ing to implement this intruder capturing system with

an actual mobile robots.

HAMT 2020 - Special Session on Human-centric Applications of Multi-agent Technologies

376

ACKNOWLEDGEMENTS

This work is partially supported by Japan Society for

Promotion of Science (JSPS), with the basic research

program (C) (No. 17K01304), Grant-in-Aid for Sci-

entiﬁc Research (KAKENHI) and Suzuki Founda-

tion.

REFERENCES

Dorigo, M. (2005). SWARM-BOT: an experiment in swarm

robotics. In 2005 IEEE Swarm Intelligence Sympo-

sium, SIS 2005, Pasadena, California, USA, June 8-

10, 2005, pages 192–200.

Dorigo, M., Maniezzo, V., and Colorni, A. (1996). Ant sys-

tem: optimization by a colony of cooperating agents.

IEEE Transactions on Systems, Man, and Cybernet-

ics, Part B (Cybernetics), 26(1):29–41.

Goss, S., Aron, S., Deneubourg, J. L., and Pasteels, J. M.

(1989). Self-organized shortcuts in the argentine ant.

Naturwissenschaften, 76(12):579–581.

Kambayashi, Y., Sekido, T., and Takimoto, M. (2020). Cap-

ture of intruders by cooperative multiple robots using

mobile agents. In the 3rd International Conference on

Intelligent Human Systems Integration. Springer. to

appears.

Kambayashi, Y., Yajima, H., Shyoji, T., Oikawa, R., and

Takimoto, M. (2019). Formation control of swarm

robots using mobile agents. Vietnam J. Computer Sci-

ence, 6(2):193–222.

Nagata, T., Takimoto, M., and Kambayashi, Y. (2013).

Cooperatively searching objects based on mobile

agents. Trans. Computational Collective Intelligence,

11:119–136.

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

(2011). A serialization algorithm for mobile robots us-

ing mobile agents with distributed ant colony cluster-

ing. In Knowledge-Based and Intelligent Information

and Engineering Systems - 15th International Confer-

ence, KES 2011, Kaiserslautern, Germany, September

12-14, 2011, Proceedings, Part I, pages 260–270.

Song, C., Liu, L., and Xu, S. (2019). Circle formation con-

trol of mobile agents with limited interaction range.

IEEE Trans. Automat. Contr., 64(5):2115–2121.

Takahashi, R., Takimoto, M., and Kambayashi, Y. (2014).

Cooperatively transporting unknown objects using

mobile agents. In ICAART 2014 - Proceedings of the

6th International Conference on Agents and Artiﬁcial

Intelligence, Volume 2, ESEO, Angers, Loire Valley,

France, 6-8 March, 2014, pages 60–68.

Takimoto, M., Mizuno, M., Kurio, M., and Kambayashi, Y.

(2007). Saving energy consumption of multi-robots

using higher-order mobile agents. In Agent and Multi-

Agent Systems: Technologies and Applications, First

KES International Symposium, KES-AMSTA 2007,

Wroclaw, Poland, May 31- June 1, 2007, Proceedings,

pages 549–558.

Yang, H. and Wang, Y. (2019). Formation optimization

and control for maneuvering target tracking by mobile

sensing agents. IEEE Access, 7:32305–32314.

Capture of Multi Intruders by Cooperative Multiple Robots using Mobile Agents

377