Work Efficiency in Mobile Multi-Robot System

Through Energy Management Inspired by Honeybee

Foraging Behavior
Faisul Arif Ahmad, Abd Rahman Ramli, Khairulmizam Samsudin, S. J. Hashim

[email protected], [email protected], [email protected] and [email protected]

Department of Computer and Communication System Engineering, Faculty of Engineering, Universiti Putra Malaysia, 43400
UPM Serdang, Selangor MALAYSIA1

Abstract—Abstract--- Mobile multi-robots system is capable to robot can do but they also capable to operate what single
work more effectively when navigated with cooperative system. mobile robot cannot do.
There were several cooperative systems that applied on the
mobile multi-robot such as centralized system, decentralized Mobile multi-robot can cooperatively complete task faster.
system or distributed system. The systems have their own In order to develop mobile multi-robot system with
advantages for their purposed of the task. Multi-mobile robots cooperatively behaviors, several architectures has been
have been developed to assist humans such as in high-risk developed such as centralized system, decentralized system,
environment and complex environment. In mobile robots, the and distributed system (see Figure 1).
energy utilization is very important to be handle wisely, because
the more it could utilized in the working behavior, the more work
and more faster it can be done. In such theory, the management
in energy utilization is very important in order to work
efficiently, mainly for mobile robots that used in search and
rescue mission because the time and work is very important. In
this work, the algorithm for a common mobile multi-robots, the
most important is to navigate and coordinate themselves in the
environment that shared with other mobile robots in order to
avoid frequently interference and also redundant in work that (a) (b) (c)
could reduce their effectiveness in their work. The low Figure 1: The cooperative systems in mobile multi robot (a) centralized
interference of mobile robots can be used as communication system (b) decentralized system and (c) distributed system
behavior. Inspired by honeybees system which is also known as
part of the distributed system is applied to optimize the foraging Centralized system (Figure 1(a)) is a system that has a central
power station and maximize the work in swarm robotic system. control for mobile multi-robot. This system is similar with
The algorithm also give the mobile robots advantages in term of server-client system which the central control is controlling
cooperation by sharing information and also the dynamic everything by receiving information from clients or agents and
threshold based on the knowledge of the power station. The result then process the information. The central control than produce
of the swarm robotic system showed the increasing of the the decision which given back to mobile robots to execute the
parameters in time consumption, energy utilization, and work command. The system is currently popular in soccer robotic
utilization in the working behavior. application which has player as client robot, and a server which
is a computer that control the robot players.
Keywords—mobile multi robot; inspired algorithm; honeybee;
swarm robot; energy management Decentralized system (Figure 1(b)) is a group of mobile
robots that shared and communicate the information through
I. INTRODUCTION their work such as a leader or follower in the system. The
Nowadays, multi-robot is becoming popular in the information given by a leader is normally followed by other
application of mobile robots. It has been popular in the area mobile robots to complete the task. The system is more likely a
such as mobile sensor node system, soccer robotic, search and group of army that have a leader that given them information
rescues and etc. Researchers looking the mobile multi-robots in for their task and also ask for information when needed.
different ways of application such as in the space exploration, Distributed system (Figure 1(c) in mobile robot is a
search and rescue missions, and also in the complex cooperative behavior system that each of individual mobile
environments for human to conduct. Another matter, using robot has intelligence to react and act by their own. On top of
mobile multi-robots is more competitive compared to use a that, they are cooperating in their task through communication.
single, powerful and complex mobile robot. This is because the The intelligent algorithm can be inspired by biological social
multi-robot is not only capable to operate what single mobile
system such as ant colony [1, 2], swarm of honeybees [3, 4], inertial. The spontaneous robots applied a simple threshold
termites [5], and fishes [6]. model in deciding from working to the recharging. The inertial
robot is using more complex model based on the priority of
Corell et al. [7] developed an algorithm that used mobile behavior to decide the working and recharging behavior. The
multi-robots as module sensors and actuator networks in order result shown that, the inspired honeybees mobile robots is work
to act as sensor in inspection and maintenance of the jet turbine more efficient compared to normal threshold mobile robots.
engine. In the aircraft industry, the inspection of the jet turbines
is a complex task that needs to be checking in periodically, In application of mobile multi-robot system in industries or
which currently is using the manual work. Using the miniature other platform, the issues of energy resource that known as
mobile is one of the advantages [8]. This matter is not only finite energy, which is supplied from batteries is one the
providing the benefit to the aircraft industry but also the others important issues. Several researches have been done to
that using the large machine that need to be inspect in strategies the energy consumption such as [3,13]. This research
periodically. One of the limitations of mobile robots in their will discuss a strategy based on the algorithm inspired of
experiment was energy limitation, because of the small honeybees to optimize the working behavior and minimize the
physical size. foraging behavior based on the knowledge of the power station
location. The scope of work is simulation environment with
Liu et. al. [9] designed algorithms for mobile multi-robots
two main behaviors in two areas which are working and
to optimize the energy consumption through the mechanism of foraging.
automatically adjust the ratio of rester and forager within
cooperative rules which is inspired by honeybee system. The In this paper, Section Background started with the
rules that had been identified in the works were internal cues, discussion about the background of the swarm system inspired
environmental cues, and social cues which were dynamically by honeybees in the mobile multi-robot. Algorithm inspired by
change the time consumed in the resting and foraging behavior. honeybees system is designed and applied on mobile multi-
The objective of this algorithm was to reduce the energy robots to control the work behavior and forage behavior for
consumption in the unwanted movement such as large numbers power station. The algorithm also shared the information of the
of mobile robots used to forage objects that only has little in the power station by local communication. Section III explains the
environment. methodology in applying honeybees algorithm in mobile multi-
robots. While section 4 highlight the results and discussion.
Satizabal et al. [10] designed an algorithm that Finally the summary of the work has been written in section 6.
cooperatively perform the task to finding a location by using
the knowledge that gathered by other mobile robots regarding II. BACKGROUND
the location of the targeted position. Foraging task has been
divided into two behaviors, which were searching and homing In the environment of biological honeybees system, they
parts. The experiment is done in the environment of the had survive by cooperating within their group. In the
foraging based on the typical real-world application which is honeybees system, the works or behaviors have been
garbage collection. The algorithm has been implemented in the categorized by several divisions [14]. Starting from the stage of
simulation and real world. In the simulation, Enki (an open eggs hatching, cleaning hive, tending larvae until to end of its
source 2D robot simulator) was used. While for real-world life, the honeybees compete each other’s for the survival. The
environment, 20 of the mobile robots called marXbot were division of honeybees has been identified by zoologist based on
used. In order to verify the behavior of mobile robots, the time their stage of ages [14, 15] as Figure 2. The first division (which
of robots spent in traveling between two targets were used. The is started from bottom) till to final division (foraging food)
results show that there were weak relationship between robot shows the reasonable work of honeybees. From the division in
performance and the amount of robot performing task. Also honeybees system, it can be concluded that the early stage of
another comparison on GPS-like approach and social which honeybees life, it spent more work in the hive. After honeybees
results showed that the social approach was more robust in the getting more experience with the environment, than honeybee
environment that had longer obstacle in the environment. started to work outside of the hive. From five division of
honeybees system, only one that needs the bees to work outside
Bodi et al. [11] investigated the robustness of the completely, which is 80% of their life is inside the bee hive.
BEECLUST algorithm (an inspired algorithm by honeybees) to
act against disturbances by another groups of mobile robots. Frish [16] had identified the bee communication techniques
The algorithm was applied on the simulation areas with two in his experiment and defined the behavior as bee dance. Two
different tasks of Jasmine III mobile robots. One of the group types of movements that have been defined by Karl are round
aggregates at high illumination (light spot) and the other at dance (to indicate food source less than 50m) and wagging
places of low illumination (shadow spot). The population of dance (food source more than 50m), as shown in Figure 1.
mobile robots also has been test with small group, medium III. METHODOLOGY
group and large group of mobile robots.
A. Platform
Kernbach et al. [12] organized a strategy to manage energy
resource of mobile robots without using global knowledge in An open source software known as Player/Stage was used
order to have mobile robots work efficiency. They had defined to simulate mobile multi-robot. The Player is a software tools
4 main behaviors which were working, searching, waiting and for mobile robot and sensors applications. It provides a network
recharging for operating mobile robots. Two algorithms have interface to different types of robots and sensors. The Player
been developed in the experiments which are spontaneous and platform can modeling a client/server system Stage is multi-
mobile simulator which simulate a populationn of mobile robots B. Mobile Robots
and sensing in two dimensional environmennts. The simulator The miniature mobile robbot that known as autonomous
was run on Ubuntu Precise Pangolin versionn (12.04) platform miniature robots (AMiR) is reconstructed
r in the simulation
with 4GB RAM, 500 gigabyte of hard-diskk. The version of
platform. AMiR (as shown in Figure 4) was built by
Player is 3.0.1 and Stage is 3.2.2.
Khairulmizam et al. [17] purpoosely to develop a swarm robotic
platform. Due to number of AM MiR could not accommodate the
minimum number of mobile robot that needs to be used; the
simulation platform is chosen to overcome the problems.
AMiR in the Player/Stage simuulator is reconstructed based on
real AMiR with six sensors, twwo wheels of motor and smaller
size.

(a) (b)
Figure 4 : Images of mobile robots AMiR
A in (a) real AMiR, and (b) AMiR
in simulation environment

Figure 2: Labor Division of honeyb

bees [14]

20 mobile robots is running in two separate area, which was

designed for a working area anda foraging area as shown as
Figure 5. This environment was designed based on the
inspiration of the honeybee ennvironment which is divided by
bees hive and foraging enviroonment. Bee hive is separated
with foraging area, normally can be accessed by doors. While
in the foraging area, the food sources are separated away and
located in random.

Figure 3: Foraging honeybees forr foods

Figure 5 : Environment of the experiments with working area and

This research will be developing thhe algorithm that foraging area inspired by honeybees system
maximizes the work utilization in multi-m mobile robots. The
C. Honeybees inspired algorithhm
environment of multi-mobile robots that applied with the
algorithm is a closed environment with reepresented by two The decision of mobile roobots in foraging behavior and
main behaviors, which are work and forage behavior. Inspired working behaviors are designeed with adaptive decision. The
by honeybees foraging system, the thresshold of foraging adaptive decision is defined byy using dynamic threshold into
mobile robots is designed to be dynamic thrreshold. Threshold decision of changing from worrking to foraging behavior. It is
is selected by referring to the knowledge off the power station referring to the threshold of thhe remaining energy of mobile
that shared through local communication. The mobile robot robot. The algorithm is illustraated in the Figure 6. It shows two
that having the knowledge of the power statioon is utilized more thresholds, which is one for mobile
m robot without knowledge,
energy in the work area before it started to doo a recharging. thr1 and one for mobile robbot with knowledge, thr2. The
remaining battery energy has been chosen as parameter to
decide the threshold. The knowledge of the location of the measured by dividing the utilization time in work by the
power station is used to identify which threshold should be algorithm and the time utilized by mobile robots that consumed
applied by mobile robot. all from the battery energy Equation 4. The more percentage of
time that has been spent in the work area, will result the more
work can be done by mobile robot.
(3)

% 100 (4)

The work consumption is the parameter of energy

utilization and time spending on working area. It has been
measured during the mobile robots started to work until its
changed to the behavior of foraging power station. The work
has been measured by calculating how much the energy is
consumed over the time as shown in Equation 5 and the unit of
work in SI is Joule (J). The percentage of work is measured by
dividing it with the work that has been done by mobile robots
with full energy of battery until it lost the energy as shown in
the Equation 6.
(5)

% 100 (6)
Figure 6 Honeybee inspired algorithm with dynamic threshold decision

D. Parameters Utilized
Work utilization of mobile robots in the working area has IV. RESULT AND DISCUSSION
been chosen as the parameter to identify the efficiency of the Based on the experiments that operated by mobile robots
working mobile robots. In order to measure the work the in simulation platform, the result of percentage energy
parameter of energy utilization and time consumption in the utilization, time utilization and the work utilization in the
working area is measured. The algorithm is established based working area are shown and discussed. The results show 3
on the results of working behavior by: works iteration started from second iteration for each
• the static and the dynamic threshold decision and, measurement.

• the crowdedness of mobile robot with model The energy utilization of mobile robots is illustrated in
environment. Figure 7. The results shows the energy utilization of 10 mobile
robots with static threshold (ST) decision consumed around
Measuring the energy utilization of mobile robots 70% of the energy in the working behaviors (Figure 7(a)).
during the work behavior by referring the remaining of the There are slight differences in each iteration. This is happened
power mobile robot at the position of home entry and the because in the community of mobile robots, when there were
power remaining of mobile robot before it change to the interfered with each other’s, the energy consumption is
behavior forager . The energy is measured using the increased.
Equation (1). The percentage of energy utilization is measured
by dividing it with the energy of mobile robot that utilized all So the little interference of the mobile robots is increasing
the energy from battery as Equation (2) the work utilization. While the energy utilization in dynamic
threshold (DT) shown the mobile robots had utilized more
(1)
than 80% of their energy in the battery (Figure 7(b)). This
result gives advantage to mobile robots more than 10% of their
% 100 (2)
energy. While for energy utilization in 20 units mobile robot
(Figure 7(c)) compared to 10 unit mobile robots, the result
The time of mobile robot spending on working behavior at shown there are no significant differences.
work area also can be measured as the mobile robots positioned
at the entrance of home area, . While the end time of time Time utilization of mobile robots in working area are
consumption is recorded during the time of mobile robot measured and illustrated in Figure 8. 10 units of mobile robots
decided to change to a forager . The equation to measure time with ST behavior is shown in Figure 8(a), DT in Figure 8(b)
consumption is shown as Equation 3. Time is measure by and 20 unit mobile robots in Figure 8(c) with DT. Working
second, s. The percentage of time in working behavior is
time of 10 mobile robots, with ST shown that, the mobile
robots consumed around 70% of the time. While with DT
mobile robots consumed around 80%. Both of the DT behavior
mobile robots show almost similar in time utilization.

(a) (b)

(a) (b)

(c)
Figure 8 : Time consumption in working behavior with (a) 10 unit robots
in ST, 10 unit robots in DT, and (c) 20 Unit robots in DT.

Average and standard deviation of percentage in work

(c) utilization are illustrated in Figure 10(b). The differences of
Figure 7: Energy utilization in working behavior with (a) 10 unit robot in average work in 10 mobile robots with ST and DT is quite
ST, (b) 10 unit robots in DT and (c) 20 unit robot in DT. large, which is around 10~20 %. While, for 20 mobile robots
Based on the result of the energy utilization and the time with dynamic threshold is slightly lower than 10 mobile robots
utilization, the work of mobile robots with ST and DT can be as shown in Figure 10(b). In the standard deviation, the DT
measured as illustrated in Figure 9. The result of 10 unit behavior for 10 mobile robots environment, the high variation
mobile robots that work with ST, utilized work around 50% as are indicated by the mobile robot R01, R02, R03, R05 and
shown in Figure 9(a). While DT mobile robots with 10 unit R09. These mobile robots had foraged the power station
and 20 unit mobile robots shown the work measured is around without the information during the first working iteration. This
70% of the energy in battery, as illustrated in Figure 9(b) and behavior result the mobile robot to work less because of the
Figure 9(c). increase energy in the foraging behavior. The low variation in
DT behaviors in environment of 10 mobile robots such as R04,
From the results of the measurement works of mobile R08 and R10, also environment with 20 mobile robots, are
robot, each of three examples from ST and DT are taken and mobile robots that had forage with the information from the
illustrated in Figure 10(a). Work with ST shown the value of first working iteration.
work is lower than work with DT behavior. The work
utilization with DT behavior is around 20% higher compared
to ST behavior. This means that mobile robots with DT
behavior can do work more than ST behavior. For the first
work iteration the value is lower because of the remaining
energy of mobile robots started with different value. Some of
them such as R04 and R05 from ST and R01 and R05 for DT
started with zero because the mobile robots started with
foraging for power station.
(a) (b)
 [1] S.-H. Chia, K.-L. Su, J.-H. Guo, and C.-Y. Chung, “Ant colony
system based mobile robot path planning,” in Genetic and Evolutionary
Computing (ICGEC), 2010 Fourth International Conference on, 2010.
[2] M. Dorigo, M. Birattari, and T. Stutzle, “Ant colony optimization,”
Computational Intelligence Magazine, IEEE, vol. 1, no. 4, pp. 28 –39,
November 2006.
[3] W. Liu, A. F. T. Winfield, J. Sa, J. Chen, and L. Dou, “Towards
energy optimization: Element task allocation in a swarm of foraging robots,”
Adaptive Behavior, vol. 15, no. 3, pp. 289–305, September 2007.
[4] A. Faisul Arif, A. Ramli, K. Samsudin, and S. Hashim, “Energy
management in mobile robotics system based on biologically inspired
honeybees behavior,” in Computer Applications and Industrial Electronics
(ICCAIE), 2011 IEEE International Conference on, December 2011, pp. 32 –
35.
[5] K. Petersen, R. Nagpal, and J. Werfel, “Termes: An autonomous
robotic system for three-dimensional collective construction,” in Robotics:
Science and Systems VII, ser. VII, H. F. Durrant-Whyte, N. Roy, and
(c) P. Abbeel, Eds. The MIT Press, 2011, p. 257.

Figure 9: Work utilization in working behavior with (a) 10 unit robot [6] I. D. Neveln, Y. Bai, J. B. Snyder, J. R. Solberg, O. M. Curet,
with ST and (b) 10 unit robot with DT and (c) 20 unit robots with DT. K. M. Lynch, and M. A. MacIver, “Biomimetic and bio-inspired robotics in
electric fish research,” Journal of Experimental Biology, vol. 216(13), pp.
2501–2514., March 2013.

One of the reasons of the decreasing of work in [7] N. Correll, C. Cianci, X. Raemy, and A. Martinoli, “Self-organized
embedded sensor or actuator networks for smart turbines,” in in IROS 2006
environment of 20 mobile robots compared to 10 mobile Workshop: Network Robot System: Toward, 2006.
robots because of the increasing of the interference among
[8] N. Correll and A. Martinoli, “Multirobot inspection of industrial
mobile robots in the platform. The percentage of the machinery,” IEEE Robotics & Automation Magazine, vol. 16,, no. 1, pp. 103–
differences is slightly small when comparing the mobile robots 112, Mac 2009.
with 10 and 20 units. Other related problems are the number of
[9] W. Liu, A. F. T. Winfield, J. Sa, J. Chen, and L. Dou, “Strategies
the power station compared to numbers of mobile robots that for energy optimisation in a swarm of foraging robots,” Swarm Robotics, vol.
increased also could increase the crowdedness during the 4433, pp. 14–26, 2007.
recharging behavior.
[10] H. Satizabal, A. Upegui, A. Perez-Uribe, P. Retornaz, and
F. Mondada, “A social approach for target localization: simulation and
V. SUMMARY implementation in the marxbot robot,” Memetic Computing, vol. 3, pp. 245–
Application of mobile multi-robots in work inspired by 259, 2011. [Online]. Available: http://dx.doi.org/10.1007/s12293-011-0061-z
honeybee behavior shown that, they are able to increase the [11] Robustness of two interacting robot swarms using the BEECLUST
time, energy and work utilization in their working behavior by algorithm, 2009.
improving the algorithm with the dynamic threshold in the [12] S. Kernbach, V. A. Nepomnyashchikh, T. Kancheva, and
decision of working to foraging behavior. This comparison O. Kernbach, “Specialization and generalization of robot behavior in swarm
results show that, the mobile robots with dynamic threshold energy foraging,” Mathematical and Computer Modelling of Dynamical
increased the percentage of work utilization in working area Systems, vol. 18, no. 1, pp. 131–152, February 2012.
compared to the mobile robot with the static threshold. The [13] J. H. Lee, H. T. Kim, and C. W. Ahn, “Foraging swarm robots
increased of the percentage is more than 10% of the energy system adopting honey bee swarm for improving energy efficiency,” in
ICUIMC ’12 Proceedings of the 6th International Conference on Ubiquitous
and the time utilization. While, in the work utilization, it Information Management and Communication, Kuala Lumpur, Malaysia,
increased more than 20% in the dynamic threshold compared February 2012.
to the static threshold algorithm. In future work, it is important
[14] H. Yahya, The Miracle of Honey-bees, 1st ed., T. Mossman, Ed.
to identify the limitation of the number of mobile robots that Global Publishing, Mac 2007.
could be increased to avoid the unnecessary crowdedness in
[15] S. E. Fahrbach and G. E. Robinsson, “Behavioral development in
the environment. Another parameter that could be look at is the honey-bees: Toward the study of learning under natural conditions,” Learn
the capability of mobile robot that could identify the nearest and Memory, no. 2, p. 1990224, 1995.
power station with the knowledge more than one power
[16] K. V. Frisch, “Decoding the language of the bee,” in Nobel
station. Lecture, December 1973. [Online]. Available: http://www.nobelprize.org/-
nobel_prizes/medicine/laureates/1973/frisch-lecture.html
AKNOWLEDGMENTS
[17] F. Arvin, K. Samsudin, and A. R. Ramli, “Development of a
This study was supported in part by Research University Grant miniature robot for swarm robotic application,” International Journal of
(RUGS) no. 05-01-12-1625RU. The authors acknowledge the Computer and Electrical Engineering, vol. 1, no. 4, pp. 1793–8163, 2009.
comment of the reviewers.