See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/325797128

An Algorithm Based on the Bacterial Swarm and Its Application in

Autonomous Navigation Problems

Chapter in Lecture Notes in Computer Science · June 2018

DOI: 10.1007/978-3-319-93815-8_30

CITATION READS

1 86

3 authors, including:

Fredy Martínez M. Arbulu

District University of Bogotá Corporación Universitaria del Huila
120 PUBLICATIONS 394 CITATIONS 56 PUBLICATIONS 253 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by M. Arbulu on 02 August 2018.

The user has requested enhancement of the downloaded file.

 An Algorithm Based on the Bacterial
Swarm and Its Application in
Autonomous Navigation Problems

Fredy Martı́nez1(B) , Angelica Rendón1 , and Mario Arbulú2

1
District University Francisco José de Caldas, Bogotá D.C, Colombia
[email protected], [email protected]
2
Corporación Unificada Nacional de Educación Superior, Bogotá D.C, Colombia
mario [email protected]
http://www.udistrital.edu.co

Abstract. Path planning is a very important problem in robotics, espe-

cially in the development of Automatic Guided Vehicles (AGVs). These
problems are usually formulated as search problems, so many search
algorithms with a high level of intelligence are evaluated to solve them.
We propose a navigation algorithm based on bacterial swarming from
a simplified model of bacterium that promises simple designs both at
the system level and at the agent level. The most important feature of
the algorithm is the inclusion of bacterial Quorum Sensing (QS), which
reduces the convergence time, which is the major disadvantage of the
scheme. The results in both simulation and real prototypes show not only
stability but higher performance in convergence speed, showing that the
strategy is feasible and valid for decentralized autonomous navigation.

Keywords: Artificial bacterium · Autonomous · Path planning

Quorum Sensing

1 Introduction

One of the objectives of autonomous robotics is the development of AGVs [8].

These vehicles allow the autonomous transport of cargo along more or less pre-
defined routes between a point of origin and a target point. It is desirable in
these tasks that the vehicle be autonomous, find its destination regardless of
its current position, and solve problems caused by obstacles on the path. In
addition, depending on the environment and type of cargo, it is ideal to have
swarms of these vehicles. This type of activity is of great importance in indus-
trial automation schemes, since it allows the design of intelligent warehouses,
the implementation of flexible manufacturing systems, and considerably reduce
production times and costs [2,3].
Path planning is one of the key problems to be solved in the operation of
these autonomous vehicles [6,10]. This problem focuses specifically on finding a
c Springer International Publishing AG, part of Springer Nature 2018
Y. Tan et al. (Eds.): ICSI 2018, LNCS 10941, pp. 304–313, 2018.
https://doi.org/10.1007/978-3-319-93815-8_30
 An Algorithm Based on the Bacterial Swarm 305

safe route from the current position of the robot vehicle to the final target posi-
tion, intelligently dodging obstacles in the environment, and according to some
motion policy that contemplates performance conditions and safe movement.
Path planning problems can be divided into two categories according to the
characteristics of the environment: static environments (do not change during
robot navigation) and dynamic environments (change during navigation task).
The majority of AGV’s researches contemplate static and generally observ-
able environments, which allows the mapping of the environment and the design
of the route in an off-line way, to then instruct the movement to the robots [5].
However, in real industrial environments the environments are usually dynamic,
with continuous movements of people and equipment. There are different docu-
mented cases in which biological systems efficiently solve this type of problems,
one of these schemes is observed in the interaction between bacteria [9,11].
Self-assembly processes are responsible for the generation of order in nature
[4]. They involve components at different scales, such as molecules, cells, organ-
isms, communities, ecosystems and weather systems. These interactions between
elements of relatively simple structure can be analyzed, modeled and replicated
in the design of artificial systems. One of these systems is the bacterial com-
munities. As biological organisms bacteria have a very simple structure, their
behavior can be described from local interactions between organisms and with
the environment, and as a system they have proven to be very successful in very
adverse conditions [1].
In the biological model, the cell-cell communication is performed through the
exchange of chemical molecules called auto inducers. This mechanism, called
Quorum Sensing (QS), allows bacteria to monitor their environment for the
presence of other bacteria, and thus, to respond to fluctuations in the number
and/or species present. That means, coordinate collective behavior in bacteria
under environment conditions [7].
During operation of this mechanism, the concentration of the signal molecule
reflects the number of bacterial cells in a particular area. The detection of a
threshold concentration, a molecular signal, indicates that the population has
reached the quorum, i.e., is ready to perform a specific collective behavior. This
means that QS is a mechanism used by bacteria to activate phenotypic changes
in the population, i.e., to coordinate gene-expression.
This research proposes a different design scheme inspired by this mecha-
nism of gene expression control, mechanism which is dependent on cell density,
and that allows to coordinate the navigation of a swarm of robots by means of
indications in the environment (landmarks), and local communication between
robots. The tasks to be carried out by robots are modeled as behaviours that are
expressed by the organism according to population density and a control policy.
That is, robots contain code equivalent to gene expression, which is responsible
for social behaviors of independent cells using extracellular signals.
The paper is organized as follows. In Sect. 2 presents a description of the
problem. Section 3 describes the proposed strategy and a general design outline
306 F. Martı́nez et al.

for basic navigation tasks. Section 4 introduces some results obtained with the
proposed strategy. Finally, conclusion and discussion are presented in Sect. 5.

2 Problem Statement
The multi-agent model adopted for artificial systems design is composed of a set
of artificial bacteria or agents. This set of agents, and their interactions, reflect
the dynamics that will solve the problems. All agents are identical in design.
Nevertheless, to solve tasks, each of them undergoes certain behavior inside the
system along the time.
A bacterium is a pair

V = (f, p) (1)

where f is a nonnegative integer (f ∈ Z+ ) that indicates the amount of neigh-

boring bacteria in contact, and p is a point in q-dimensional space (p ∈ Rq ).
The population density is evaluated using the distance between bacteria. Let:

dij = d (Vi , Vj ) (2)

as the distance between bacteria Vi and Vj , which is calculated by any appro-

priate norm. A bacterial population is defined as a nonempty set of bacteria.

W0 = {V1 , V2 , V3 , · · · , Vm } (3)

with non-zero distance among bacteria:

dij = 0, ∀i, j , i = j (4)

A group of bacteria is characterized by the parameter T , the quorum thresh-

old, it is the parameter defining whether or not it has reached the quorum.
The principle of the algorithm based on QS is the grouping of agents. Agents
are grouped into different areas of the environment according to local readings
and their genome that follows some search criteria. Since the proposed model
considers only local interactions, and that the goal of the design focuses on the
coordination of the movement of the agents, the QS mechanism should facili-
tate the movement of agents in the environment towards areas with sufficient
population, reflecting the decisions of many agents.
The system is composed of the n agents (n artificial bacteria) as a continuous
system, whose dynamics can be described by differential equations. The agents
in this system, all identical in design, experience at any instant a behavior l of
a set la of possible behaviors of the agents. Furthermore, each agent is always
performing a behavior at any given time. The variables in these differential
equations indicate the size of the bacterial population that belongs to each of
these behaviors.
In particular, considering the circumstances under which bacteria can be
grouped, and that the intensity of this grouping is who allows to find a solution,
 An Algorithm Based on the Bacterial Swarm 307

Fig. 1. System and agents with three different behaviors. (a) System composed of a
group of agents (bacteria) in a three-stage process (a = 3), (b) system agents running
behavior l1 , (c) system agents running behavior l2 , and (d) system agents running
behavior l3 .

we propose a general three-stage model (a = 3). The additional intermediate

stage should facilitate the population changes (Fig. 1).
Groups of agents at a given time experiencing the same behavior (any of the
a behaviors). Therefore, the system can be analyzed as consisting of groups of
agents characterized by a common behavior of a set of behaviors la .
The model for this system is hybrid, the different agent behaviors are trig-
gered by certain events (event-based). The set of agents, each with a continuous
dynamic X a , switches between different behaviors during the development of
the task. The group of agents simultaneously experiencing the same behavior,
it also has a continuous collective behavior, so the system also has a continuous
dynamic X p . When changing behavior one or more system agents (with dis-
crete control mode determined by La ), changes the system as well (with discrete
control mode determined by Lp ).
This means that the behavior space of the system is divided into a set Lp of
la regions. The state of the system Hp is therefore defined by the values of the
continuous variables of the system X p and the discrete control mode determined
by Lp (Fig. 2).

Fig. 2. System and its fractions characterized by their behavior.

308 F. Martı́nez et al.

Each of these population fractions is characterized by a set of differential

equations with state variables (vectors) denoted by:

xi  i = 1, 2, 3, · · · , b (5)

A set of dimension b (the number of grouping areas defined by the agents in

the system) that allows to describe the dynamics of the a different behaviors.
The hybrid system then can be described as:

Hp = {X p , Lp }
X p ⊂ Rb (6)
l a ⊂ Lp

The algorithm does not intend to explicitly define the position of the robots.
On the contrary, it looks to abstract the information from the environment to
reduce the complexity in the design of the path. These features ensure navigation
in environments with obstacles, even when they are in motion. It also eliminates
the problem of impossible movements for the robot, since the dynamics of the
robot is not part of the model. This allows the normal differences between robots,
and even the use of radically different robots. Also, the problem of controller
design is independent of the number of robots, which promises scalability to
large groups and system functionality against the failure of a few of them.

3 Methodology
The research proposes the system as a three-stage process of individual behav-
iors. These three basic behaviors are: (1) Reproduction, (2) Exploration, and (3)
Virulence.
The first behavior, Reproduction, allows to activate the robot in the envi-
ronment. The last behavior, Virulence, must be the robot’s final behavior when
it finds the target area. The second behavior, Exploration, should help to find
potential grouping areas. The principle of the algorithm based on QS is the
grouping of agents. Agents are grouped into different areas of the environment
according to local readings and their genome (internal code) that follows some
search criteria. The areas with the highest number of agents must show viru-
lence and will correspond navigation task solution. These population sizes are
then the most important variables in the model.
The population size of each of these fractions is characterized by a continuous
variable in the following way:

– R Agents whose behavior is reproduction.

– Xi Agents whose behavior is exploration in the area i of a total of b areas
in the environment.
– Zi Agents whose behavior is virulence in the area i of a total of b areas in
the environment.
 An Algorithm Based on the Bacterial Swarm 309

When agents begin to explore and activate their virulence, they begin to
group in different areas of the environment (a total of b areas). This increases
the number of system variables, where i indicates the grouping area of the agents.
When the agents are reproducing there is no grouping.
The reproduction is the first behavior for all bacteria. After an artificial agent
is enabled on the system (R), it begins to take local information but without
activating its virulence (Xi ), to finally activate its virulence when the quorum
is fulfilled (Zi ). The activation or virulence of agents that are exploring the
environment starts when Zi = T , the quorum threshold. Both explorers and
virulent can switch to new areas of the environment if the local information
indicates that they are in areas of higher performance (according to the task).
This is the most important part of the process, because it gives the opportunity
to each agent individually to find the best solution.
This parallel navigation structure allows that the explorers consider different
areas of the environment before gathering around some of them. Thus, when a
quorum is reached, the agents begin to become virulent according to the best
option for each of them. The flow chart of Fig. 3 shows the rules by which agents
interact with each other and with the environment to meet the goal.

Fig. 3. Flowchart for the proposed dynamic of bacterial QS.

310 F. Martı́nez et al.

4 Results
The strategy was evaluated in different simulations and real platforms. In a
first simulated case we designed a simple navigation task. Let us think in an
environment with four high performance areas (for example, meeting areas with
some interesting variable: temperature, noise, light, humidity, ...) denoted by
the points p1 , p2 , p3 and p4 . Among these points, we will assume that p1 , p2
and p4 have a similar performance (local maxima, the variable is higher than
the average values of the environment, but not the maximum), while point p3
has a superior performance (global maximum, the variable has the maximum
value in the environment). Under these conditions, we expect that the agents
are distributed in the environment and navigate initially grouped around these
four points. But then, they must migrate to the point of higher performance p3 .
We assume for simplicity that for the initial selection of the four areas do
not exist favoritism criteria, i.e.:

μ1 = μ2 = μ3 = μ4 = μ
(7)
λ1 = λ2 = λ3 = λ4 = λ

The migration process is based solely on the performance of the areas

detected by the agents (ρ and k calculation) and the QS. Accordingly, the system
of equations can be written as ecu. (8).

Ṙ = −4μR
Ẋ1 = μR + λZ1 Θ (R) Θ (T − Z1 ) − ρ12 X1 + ρ21 X2 − k1 X1
Ẋ2 = μR + λZ2 Θ (R) Θ (T − Z2 ) − ρ23 X2 + ρ32 X3 + ρ12 X1 − ρ21 X2 − k2 X2
Ẋ3 = μR + λZ3 Θ (R) Θ (T − Z3 ) − ρ34 X3 + ρ43 X4 + ρ23 X2 − ρ32 X3 − k3 X3
Ẋ4 = μR + λZ4 Θ (R) Θ (T − Z4 ) + ρ34 X3 − ρ43 X4 − k4 X4 (8)
˙
Z1 = k1 X1 − ρ12 Z1 + ρ21 Z2 − λZ1 θ (R) Θ (T − Z1 )
Z˙2 = k2 X2 − ρ23 Z2 + ρ32 Z3 + ρ12 Z1 − ρ21 Z2 − λZ2 θ (R) Θ (T − Z2 )
Z˙3 = k3 X3 − ρ34 Z3 + ρ43 Z4 + ρ23 Z2 − ρ32 Z3 − λZ3 Θ (R) θ (T − Z3 )
Z˙4 = k4 X4 + ρ34 Z3 − ρ43 Z4 − λZ4 Θ (R) Θ (T − Z4 )

Each agent must determine the value of the coefficients from local interactions
between agents and with the environment. For this test will be assumed the
following values:

μ = 0.01 ρ34 = 0.001

λ = 0.03 ρ43 = 0.01
ρ12 = 0.001 k1 = 0.01
(9)
ρ21 = 0.001 k2 = 0.01
ρ23 = 0.01 k3 = 0.02
ρ32 = 0.001 k4 = 0.01

These values agree with the characteristics described for this example (higher
value for ρ23 , ρ43 and k3 ). Figures 4, 5 and 6 show the results. We simulate the
system with an initial population of 300 agents in reproduction and a quorum
 An Algorithm Based on the Bacterial Swarm 311

threshold of T = 10. Figure 4 shows how this population is rapidly reduced by

activating the other two behaviors (almost 10% over the first 50 min). In Fig. 5
is shown as explorers agents increase in the four areas. While growth is similar
in all four cases, we can appreciate more agents in area 3 (blue curve). Finally,
in Fig. 6 is shown as increasing the virulent agents in the four regions. Initially,
the agents increase in all four cases. However, in areas 1, 2 and 4 the number
of agents decreases after about 200 to 300 min, while in area 3 (blue curve) the
number of agents always increases. These curves show the migration of agents
to the area 3.

Number of agents in reproduction.

300

250

200
Number of agents

150

100

50

0
0 50 100 150 200 250 300 350 400
t [min]

Fig. 4. Macro model simulation for a simple grouping task (T = 10) - Reproduction.

Number of explorers agents. − X1. Number of explorers agents. − X2.

60 50

50
40
Number of agents

Number of agents

40
30
30
20
20

10
10

0 0
0 100 200 300 400 0 100 200 300 400
t [min] t [min]

Number of explorers agents. − X3. Number of explorers agents. − X4.

60 50

50
40
Number of agents

Number of agents

40
30
30
20
20

10
10

0 0
0 100 200 300 400 0 100 200 300 400
t [min] t [min]

Fig. 5. Macro model simulation for a simple grouping task (T = 10) - Exploration.
(Color figure online)

With a quorum threshold of T = 10 the four areas achieve to activate the

QS at some point (area three at 25 min, area one at 43 min, area two at 68 min
312 F. Martı́nez et al.

Number of virulent agents − Z1. Number of virulent agents − Z2.

70 25

60
20

Number of agents

Number of agents
50

40 15

30 10
20
5
10

0 0
0 100 200 300 400 0 100 200 300 400
t [min] t [min]

Number of virulent agents − Z3. Number of virulent agents − Z4.

200 25

20
150
Number of agents

Number of agents
15
100
10

50
5

0 0
0 100 200 300 400 0 100 200 300 400
t [min] t [min]

Fig. 6. Macro model simulation for a simple grouping task (T = 10) - Virulence. (Color
figure online)

and area four at 73 min). Upon reaching a population of 10 agents, in four cases
there was an increase in the rate of population growth. With a quorum threshold
of T = 60, only area three activates the QS at 137 min. In the latter case, the
population of virulent agents in two and four areas never exceeds 10 agents, and
in the area one reaches up to 18 agents.
In experimental tests we evaluate the behaviors on a 45 cm × 61 cm robot
equipped with a set of nine infrared sensors (each with a range of 0.2 to 0.8 m).
From the signals of the infra-red sensors the robot determines the proximity to
other robots or obstacles in the environment. Once this information is identified,
the robot establishes its behavior, following the navigation policies of the model.

5 Conclusions
In this paper we propose a navigation scheme for robot swarms that mimics the
behavior of bacterial QS. The navigation scheme is supported in robot grouping,
which can be oriented to specific navigation tasks through landmarks in the envi-
ronment that the robot can identify locally. Grouping, and therefore navigation
is accelerated according to the number of agents in an area. In the cases ana-
lyzed, it is clear how the size of the quorum threshold affects QS. The variation
in the value of T directly affects the characteristics of the population growth in
each of the areas. Small values of T allow virulent agents to reach the QS and
increase their bacterial population in local minima, but they do not exceed the
local maximum. In this case, the grouping is also faster. With large values of T ,
the QS tends to be expressed only in the global maximum, but the process takes
a little longer. In both cases, the QS helps to find the global maximum (more
virulent bacteria gather at the area of higher performance).
 An Algorithm Based on the Bacterial Swarm 313

Acknowledgments. This work was supported by the District University Francisco

José de Caldas and Corporación Unificada Nacional de Educación Superior. The
authors thank the research group ARMOS for the evaluation carried out on proto-
types of ideas and strategies.

References
1. Adamou, A., Gueroui, A., Labraoui, N., Omer, B., Titouna, C., Damakoa, I.: Adap-
tive scheme for collaborative mobile sensing in wireless sensor networks: bacterial
foraging optimization approach. In: IEEE 27th Annual Inter. Symposium on Per-
sonal, Indoor, and Mobile Radio Communications (PIMRC 2016), pp. 1–6 (2016)
2. Hazza, M., Nabila, A., Yulian, E., Hatem, A.: Empirical study on AGV guiding in
indoor manufacturing system using color sensor. In: 5th International Symposium
on Computational and Business Intelligence (ISCBI 2017), pp. 125–128 (2017)
3. Iwamura, K., Chen, J., Tanimizu, Y., Sugimura, N.: A study on transportation
processes of autonomous distributed AGV based on social force model. In: Inter-
national Symposium on Flexible Automation (ISFA), pp. 206–209 (2016)
4. Jones, G., King, P., Morgan, H., Planque, M., Zauner, K.: Autonomous droplet
architectures. Artif. Life 21(2), 195–204 (2015)
5. Li, G., Yang, Y., Zhao, F., Zhou, Y.: An improved differential evolution based
artificial fish swarm algorithm and its application to AGV path planning problems.
In: 36th Chinese Control Conference (CCC 2017), pp. 2556–2561 (2017)
6. Martı́nez, F., Acero, D.: Robótica Autónoma: Acercamientos a algunos proble-
mas centrales. CIDC, Distrital University Francisco José de Caldas (2015). ISBN
9789588897561
7. Martı́nez, F.H., Delgado, J.A.: Hardware emulation of bacterial quorum sensing.
In: Huang, D.-S., Zhao, Z., Bevilacqua, V., Figueroa, J.C. (eds.) ICIC 2010. LNCS,
vol. 6215, pp. 329–336. Springer, Heidelberg (2010). https://doi.org/10.1007/978-
3-642-14922-1 41
8. Oleari, F., Magnani, M., Ronzoni, D., Sabattini, L., Cardarelli, E., Digani, V., Sec-
chi, C., Fantuzzi, C.: Improving AGV systems: integration of advanced sensing and
control technologies. In: IEEE International Conference on Intelligent Computer
Communication and Processing (ICCP 2015), pp. 257–262 (2015)
9. Padilha, F., Subtil, M., Fabro, J., Schneider, A.: Trajectory planning in dynamic
environments for an industrial AGV, integrated with fuzzy obstacle avoidance. In:
12th Latin American Robotics Symposium (LARS) and 3rd Brazilian Symposium
on Robotics (LARS-SBR 2015), pp. 347–352 (2015)
10. Pia, M., Marcello, A., Pedroncelli, G., Ukovich, W.: A software tool for the decen-
tralized control of agv systems. In: International Conference on Control, Decision
and Information Technologies (CoDIT 2016), pp. 478–483 (20016)
11. Rendón, A.: Evaluación de estrategia de navegación autónoma basada en compor-
tamiento reactivo para plataformas robóticas móviles. Tekhnê 12(2), 75–82 (2015)

View publication stats