You are on page 1of 8

2020 the 3rd International Conference on Intelligent Robotics and Control Engineering

Prevention of Ant Mills in Pheromone-Based Search Algorithm for Robot Swarms

Ahmad Reza Cheraghi Jochen Peters


Technology of Social Networks Technology of Social Networks
Heinrich Heine University Heinrich Heine University
Düsseldorf, Germany Düsseldorf, Germany
e-mail: ahmad.cheraghi@hhu.de e-mail:jochen.peters@hhu.de

Kalman Graffi
Honda Research Institute Europe GmbH
Offenbach am Main, Germany
e-mail: kalman.graffi@honda-ri.de

Abstract—Computer scientists use animal-based phenomenon Therefore, with higher number of ants the efficiency of the
as source of inspiration to develop swarm algorithm and food search and delivery increases. Computer scientists use
simulate their behaviors. One example is ants foraging. For this kind of phenomenon, termed "pheromone-based
finding food the ants are moving randomly in any direction. searching algorithm", to solve e.g. the famous travelling
After they found food, they spread pheromones to make it easy salesman problem [2] or for secure routing solutions and
for other ants to find it. The more ants, the better the misbehavior detection mechanisms in wireless mash
pheromone trail. This pheromone-based search algorithm networks [3]. However, an undesirable situation, that is well
should increase the efficiency of food delivery. However, this known by scientists, is created by the pheromone-based
phenomenon includes a problem termed ant mill, which is well
searching termed ant mill (see Fig. 1). Ant mill occurs when
known. An ant mill is happening when many ants are
spreading their pheromones in a closed area. This causes an
many pheromone spreading ants are in a closed area and
overlapping of the pheromones and creates a circle trail. their pheromones trails overlaps with each other so that the
Consequently, the ants move in a never-ending loop without trail becomes a closed circle. Thus, the ants get into this
ever reaching the food source. The result is starvation and circle and they run continuously in a loop without ever
death. reaching the food source. Consequently, the ants starve to
In this paper, we take a closer look at three types of search death because they are moving continuously in this
algorithms. Especially pheromone-based searching overlapped pheromone circle trail. This undesired situation is
algorithm, which creates ant mills and provide a solution to described for the first time in 1944 the American Museum of
overcome them. Our evaluation shows that a pheromone-based Natural History (New York City). They describe ant mill
search algorithm with ant mill prevention brings a significantly very detailed in [4]. The first question of the creation of ant-
better food delivery. will is discussed in Delsuc [5]. They ask the question, why
ant mills (part of the ―army ant syndrome‖) are not
Keywords-multi-agent systems; simulation; robots; swarms; eliminated by the evolutionary history.
ants; pheromones; search algorithms

I. INTRODUCTION
Swarm is the collective behavior of many agents that
increases the efficiency. An example of swarm behavior are
ant colonies. Ants use pheromones to increase the efficiency
of foraging. For example, ants are moving randomly to find
food. After an ant finds a food source, it collects one piece of
the food, and goes back to its nest. While on its way back, it
spreads pheromones to communicate the location of the food
source with the other ones. The result is a pheromone trail
that leads to the food source. Thus, if other food searching
ants get into this pheromone trail, they stop the random
searching and follow this trail to get to the food source.
Therefore, the pheromone trail is an indirect communication
method between the ants. This type of communication is
termed stigmergy [1]. Without this pheromone trail the
running behavior of the ants would be always random.
Consequently, the efficiency of the foraging should increase. Figure 1. Ant mill.

978-1-7281-8972-7/20/$31.00 ©2020 IEEE 23

Authorized licensed use limited to: Carleton University. Downloaded on September 22,2020 at 10:42:19 UTC from IEEE Xplore. Restrictions apply.
In this paper, we propose a pheromone-based search with food source has a given nutritive value. The nutritive value
ant mill prevention algorithm for robot swarms. With this tells how many pieces a food source has that can be taken by
algorithm we solve the ant-mill problem. We compare this the ants. Therefore, a food source is reduced each time an ant
with two search algorithms: random walk and a pheromone- visits it.
based search algorithm. We use these algorithms to simulate
agents, called ants, that are randomly walking to find food.
We compare them by average age and food delivery of the
ants to find out which search algorithm is more efficient. For
our knowledge there is no pheromone-based search
algorithm that treats or tries to solve the ant mill problem.
In the following Section we first describe the related
work. Among them are also works that deal with the tracking
accuracy of ants or the probability of finding food. Next, we
propose the problem statement. In Section IV we present the
search algorithms: random walk, pheromone-based search
algorithm, and the pheromone-based search algorithm with
ant mill prevention. In Section V we present first the
simulation environment Swarm-Sim and all the necessary
simulator set-ups. Next, the metrics that are necessary for the
evaluation. Last but not least, in Subsection V-D we describe
the results of the simulation. The last Section belongs for the
conclusion and future works. We also consider new
questions regarding the simplifications of the ant algorithm
as well as areas that we have not yet considered with our
simulation. Figure 2. Ants in area with a food source and a nest%.

II. RELATED WORK


In the following Section we look at existing work of
different ant algorithms and their use cases. In addition, we
selected works that deal with the finding of food and the
formation of ant mills.
The first ant algorithm is documented at ―Distributed op-
timization by ant colonies‖ [6] in 1991 and they try to find a
heuristic to solve the travelling salesman problem (TSP).
Another work by Couzin [7] focuses on lane formation. Both
works set simulated ants on existing trails and analyze their
movement. Couzin tries to understand in which situation ant
mills are built in respect to the strength of following a trail.
Additionally, Couzin compares the movement of simulated
ants with real ones taken by video camera. In [8] the author
handles the probability of finding a single food location. He
adds aspects like ant speed depending on the age of an ant. Figure 3. Triangular grid and its coordinates.
The result is that ants with a higher age have a bigger range In this paper, we present a pheromone-based search
for finding food. The author in [9] uses desert ants. The algorithm for an efficient food delivery with the focus on
sandy surface of the desert removes pheromone marks very average age of each ant. In this chapter we describe in detail
fast and the desert ants use their eyes and brain primarily to all the above mentioned points for the problem statements.
re-find a location. They can remember a small number of
routes next to a location. A. World
None of the above mention works tries to solve the ant The world in this paper is a triangular grid. This grid con-
mill problem on its own. In contrast, in our work we added a sists of a defined amount of locations. Food and pheromones
kind of traffic control to the ant algorithm, so that the ants can only position on those locations. The food is placed
can still orient themselves on pheromone trails and to randomly on these locations. This triangular grid is placed on
overcome arising ant mills. a normal (x; y) coordinate system. If we talk about distance,
III. PROBLEM STATEMENT we use the term hop, that is the distance between each
location point on the triangular grid. Additionally, on a
Given is a world with several ants, food sources, and a triangular grid an ant can only move in six directions. For
nest as shown in Fig. 2. All the ants start from their nest and example, Fig. 3 shows a triangular grid and its coordinates. If
walk randomly to find food. The aim is for the ants to find
food sources and deliver pieces of it to their nest. Thus, each

24

Authorized licensed use limited to: Carleton University. Downloaded on September 22,2020 at 10:42:19 UTC from IEEE Xplore. Restrictions apply.
the ant starts at (0; 0), it can make a step to ( 1:0; 0:0), (1:0; A. Random Walk Search Algorithm (RW)
0:0), ( 0:5; 1:0), ( 0:5; 1:0), (0:5; 1:0) or (0:5; 1:0). In the random walk search algorithm, the ants start from
B. Ants their nest and move randomly in any direction until they find
food. When they find food, they will go back the same path
Our virtual ants are searching for food to survive and to from where they came from. Thus, it must be mentioned, that
live. They are starting from their nest and walk randomly while the ants are searching for food, they remember after
until they find food. Each ant has an age and a hunger level. each movement the direction from where they came from.
Both values start with 0 and increases after each simulation The ants use this memory to find their way back to their nest.
round. The ant age tells us how long an ant is living and is
useful for the evaluation. However, the hunger level B. Pheromone-Based Search Algorithm (PSA)
influences the ants’ mortality. Cause, each ant has a hunger The pheromone-based search algorithm (PSA) is a
limit. If this is reached without finding any food, the ant dies. modifi-cation of the random walk search. Basically, the ants
However, if it finds food, it is fed (the hunger level stays 0) are going out from their nest, walking randomly to find food.
until it reaches its nest and goes back to search food. Another When an ant finds food, it will take one piece of the food,
ability of the ants is that they remember all the directions and go back to its nest. On the way back, it is spreading
where they are going to find their way back to their nest. pheromones to help other ants to find the path to the food
Depending which kind of searching algorithm they are using, source, too. The PSA comes with different modes that we
they are equipped with pheromones. describe as follow. Search-Mode: When the ants are leaving
C. Food the nest, they are in search mode to find food. In the
searching-mode ants walk randomly until they find a food
Ants are searching for food to survive. However, the ant source or pheromone path. If an ant that is in ―Search-Mode‖
cannot take all the food source on its own. Thus, the food encounters a single pheromone trail it follows the trail until it
sources consist of many small pieces that can be carried by reaches the food source. If more than one pheromone trail is
an ant. The total sum of pieces we termed as nutritive value. encountered, the ant does a scoring for choosing the
Thus, if an ant reaches the food, the nutritive value is pheromone trail to follow. Nest-Mode: If an ant encounters a
decremented by one piece and if the nutritive value becomes food source, it takes one piece of the food, runs with it
zero, the food source is vanished. directly to its nest, and places a pheromone trail. This trail
D. Pheromones depends on the nutritional value of the food source.
Pheromones are used by the ants to communicate indi- Pheromone-Mode: If an search-mode ant reaches a
rectly the direction to the food source with other ants. These pheromone, it will changes to pheromone-mode. With the
pheromones are generated by ants when they find food and pheromone-mode the ant is following the pheromone trail
run back to their nest. However, the nature is not keeping the until it reaches food. Then it will change to nest-mode.
pheromones forever. Therefore, each spread pheromone Based on those mode the pheromone-based search
starts with a potency value. The potency value decreases algorithm works. However, the use of the pheromones brings
with a given evaporation rate. If the potency of the the problem of the ant mill. With the increase of ants, food
pheromone reached to zero, the pheromone does not exist sources, and the number of pheromones trails this algorithm
anymore. As a result, the ants cannot use it anymore. will create many small ant mills that we aim to resolve in this
paper and which we present next.
E. Nest
C. Pheromone-Based Search Algorithm with Ant Mill
The nest is starting and end point for the ants. It is the Preven-tion (PSAMP)
place from where the ants start to search for food. After they
found food, they will take a piece of it and go back to their Ideally, a pheromone trail should be the way between
nest. After they reached the nest and delivered the food, they food and nest because an ant places this trail exactly like that.
start again to search for food. However, due to the world size if many pheromone trails are
Now that the problem statement with its given points are spread parallel nearby the food source and there is an
define. Next, we describe three methods for solving this increase of ants, the trails overlap and cause a circle trail
problem. termed ant mill. This result of ants ―misbehavior‖ that they
run in a circle without reaching ever the food. The
IV. METHODOLOGY Pheromone-Based Search Algorithm with Ant Mill
Prevention (PSAMP) is a modification of the PSA that works
In this Section we describe the search algorithm for find-
as follows. An ant cannot recognize an ant mill, but it can see
ing food and additionally the direction decision scoring. We
how many adjacent ants are in its surroundings. As a result,
introduce first a simple random walk search algorithm. Next,
an ant should avoid ant groups to create or prevent ant mills.
we present the pheromone-based search algorithm and last
We have therefore triggered an escape behavior. If an ant
the pheromone-based ant mill prevention search algorithm.
recognize that it has more than three ants in its surroundings,
At the end we show how the ants direction decision scoring.
it will move to another direction to escape the ant mill or to
The ants use this scoring to decide in which direction they
be in a group. This behavior is straight forward, the ant
should go, if they are facing many adjacent pheromones or
randomly selects a direction and stops after a number escape
food sources.

25

Authorized licensed use limited to: Carleton University. Downloaded on September 22,2020 at 10:42:19 UTC from IEEE Xplore. Restrictions apply.
steps. The PSAMP pseudo code is shown in Algorithm 1 in exactly in front of it. After moving to this location, it also
detail. increases then the potency of the pheromone. In this example
the pheromone potency is not in percentage and the ant
Algorithm 1 Pseudo code of PSAMP increases the pheromone potency by 0.1. However, the
1: procedure PSAMP scoring is different when an ant has adjacent food sources,
2: Scan the environment too.
3: if I found food and going back to nest then In Fig. 5(a) we see an ant surrounded by one pheromone
4: if I am at the nest then
location and three food sources with different nutritive
5: Put food on the nest and go searching for food
6: else if I am on a pheromone position then
values. The ant scans all the six adjacent locations and check
7: Fill up the potency of the pheromone what is located at each adjacent location. However, in our
8: else if I am on a non pheromone location then example there are many food locations. So the proceeding of
9: Create pheromone the weighting is different as we can see in Fig. 5(b). The ant
10: end if weights all the food locations with the factor 10 and the rest
11: Move forward to reach the nest the way we described earlier. As a result, the ant neglects the
12: else if Searching for food then pheromone trail and select randomly one of the food
13: if I have still escape steps left then locations. Next, the ant goes to the random chosen food
14: Choose a random direction
15: Go to this direction
source (Fig. 5(c)) and takes one piece from the food source
16: else if I am on food then (Fig. 5(d)). The ant now switches to nest mode, orientates
17: Take food itself towards the nest, takes a step forward to a free location,
18: Go to the nest and spreads its pheromone (Fig. 5(e)). While the ant is going
19: else if Are more than three ants around me then back to its nest the hunger level of the ant is set to zero until
20: Reactive your escape steps it goes back to search mode.
21: Choose randomly a free location
22: Go to this direction V. EVALUATION
23: else if Are only pheromones around me? then
24: Make the direction scoring In this Section we first present the simulator Swarm-Sim,
25: Go to the pheromone with the biggest scoring which we use for evaluation. We provide a description of the
26: Fill up the pheromone parts of our simulation and how it works. Next, the initial set
27: else if Are food sources around me then ups of the simulator are presented and metrics are described
28: Choose randomly one of the food sources as well as parameters that we vary for the evaluation. Last,
29: Go to this direction we present the results and explain them.
30: else
31: Choose randomly a direction A. Simulator: Swarm-Sim
32: Go to this direction
33: end if
For simulation we use the round-based simulator Swarm-
34: end if Sim. Swarm-Sim is a 2D/3D swarm simulator [10]
35: end procedure developed at the Heinrich Heine University in
Düsseldorf/Germany. Swarm-Sim is written in Python. It
provides a simulation core to develop and evaluate in a
D. Direction Scoring simple way algorithm for swarms. The simulator is open
After the ants found food, they go back to their nest and source and can be downloaded from [11].
on their way, they are spreading pheromones. However, if a B. Simulation Setup
searching ant has many adjacent pheromones trail, how does
it decide which one to choose? The answer is that the We simulate the three search algorithms 10 times. Each
direction decision happens with a direction scoring, that we simulation starts with a different seed number. The seed
present in this chapter. In Figure 4(a) we see an ant that is number causes that the food source location and nutritive
neighboring two fields of pheromones. The ant now scores value changes. Hence, the ants are moving randomly, the
them to decide which one to choose. This is done with a direction decision of them changes, too. For the evaluation, a
weight factor. These are factors defined by us, which reflect Swarm-Sim world with 30 ants, 16 food source location, and
the characteristics of the sensory system and the ant’s a nest for the ants are given. The ants hunger limit is set to
willingness to make special decisions. For example, in Fig. 50 rounds. So, if an ant reaches this limitation without
4(b) we see the directions with a weight factor. The direction finding a food source, it dies, reborn, and starts to search
in which the ant is looking has the weight factor 3:0, to the again. 16 food sources are randomly located. Each food
left or to the right has a factor of 2:0, going back to the source becomes a nutritive value between 1-20 pieces, that
previous location 0:0, and all others 1:0. are randomly chosen by the simulator. The pheromones
Now for deciding the direction the pheromone potency potency is at the beginning 100% and the standard
must be multiplied the direction weight factor. This we can evaporation rate is set to 4%. The simulator runs for
see in Fig. 4(c)). For each location, the pheromones potency maximum 1000 rounds. That means, each search algorithm
is multiplied by its own weight factor. The location with the will be called for 1000 times, before the simulator terminates.
highest scoring is the one the ant decides to go. In Figure 4(d) The summary of the setup is as follow:
we see that the ant decides to go to the location that is  Maximal round numbers: 1000 rounds

26

Authorized licensed use limited to: Carleton University. Downloaded on September 22,2020 at 10:42:19 UTC from IEEE Xplore. Restrictions apply.
 Seed values: 1-10 average age we want to have a metric that is
 Food sources: 16 food locations influenced by the food delivery, irrespective of the
 Nutritive value: 1-20 pieces 30 Ants quantity and accessibility of the food sources and
 Ants Hunger Limit: 50 rounds additionally to the number of simulated ants. A high
 Pheromones potency: 100% average age of ants indicates how well the ant
 Pheromones evaporation rate: 4% colony found and delivered food.
 Deaths: With this metric we show how many ants
C. Metrics died through the simulation. More dead ants mean
For a simulation it is important to define the metrics. that the efficiency of the search algorithm is low. In
With the metrics we can see how efficient and successful the combination with the next shown metric (Death
algorithms are. We use the following metrics for the distance steps) it allows us to find more information
evaluation. about the ant mills.
 Rounds: We uses Rounds to tell how many times  Death Distance Steps: The death distance steps
the search algorithm has been called. For each means how far an ant covers before it dies. That
calling a movement for each ant is happening. means the simulator stores each time the location of
Hence, Swarm-Sim is a round-based simulator the the ant. After an ant’s death, the simulator estimates
time unit is round. the distance between the ant’s death location and
 Food Delivery: The total amount in percentage of the stored location. For example, after that an ant
food pieces that the ants have taken from the food died at round r, the simulator calculates the distance
source and delivered to their nest. While this is between the death location and the location stored at
happening the ants hunger level is set to zero and r-10 rounds. The death distance steps are an
hunger level counting stops. The counting starts indicator of mass mortality in a small area (ant mill)
again after the ant delivered the food. A high food or on a trail where there is no more food at the end.
delivery brings, consequently, a higher age. The In Subsection V-D we go into more detail.
food delivery is an indication of the success of the  Evaporation Rate: Additionally, in our
search algorithm. How evenly the task of collecting measurements we varied the evaporation rate of the
food is distributed in the ant colony can be pheromones. The evaporation rate is used to reduce
determined by this value, too. the pheromone potency after each round. With the
 Average Age: The average age is the sum of the different evaporation rates, we like to see how the
ages all the current living ants over the amount of average age of each search algorithm changes and
current living ant. The unit is rounds. As described to see if they become the same if the evaporation
earlier the ants start with the age 0 and it has rate is 100%, which means no pheromones.
nothing to do with their death. They only die of
starvation and not of age. By considering the

Figure 4. Pheromone scoring.

Figure 5. Food scoring.

27

Authorized licensed use limited to: Carleton University. Downloaded on September 22,2020 at 10:42:19 UTC from IEEE Xplore. Restrictions apply.
constant. Even if there are still food sources left, the food
D. Results delivery is not increasing anymore. Thus, it indicates that
In this Section we first describe the evaluation results of there is no food delivery anymore. Here we can see the effect
the three search algorithms, compare them and clarify their of the ant mill to PSA. This is because around the nest only
connections with a view for solving the ant mill problem. For pheromones trails to ant mills have accumulated. As a result,
all the results the ants age and hunger level are set to 0 and the ants are going into the ant mills, being trapped, and not
after each round they increase by 1. When a food source is delivering food. For PSAMP and RW, however, food
found the hunger level counter stops. The hunger limit is set delivery is continuously increasing. But the PSAMP results
to 50 rounds. If not mentioned the evaporation rate of the with a higher food delivery which means that PSAMP is the
pheromones is 4%. The simulation terminates after round most efficient algorithm
1000. 3) Death Distance Steps: Next, we take a look at
1) Age Comparison: The average age tells us if the Figures 8, 9, and 10 to get more background information
ants are good in searching and food delivery. Figure 6 about the dying ants. These graphs show the results with
illustrates for each search algorithm the ants average age three different evaporation rates. In these graphs the values
over the entire simulation time. At the beginning, the average for the random walk search algorithm are identical because it
age for all the three algorithms are equally rising. After works without pheromones and therefore it cannot be
round 50 there is a sudden fall of the average age. This is influenced by the evaporation rate.
because the hunger limit is reached, and they died because of As already shown in Section V-C, the graphs show which
starvation. However, from here this round onward the distance an ant covered shortly before it died. Since ant mills
comparison of the three algorithms can be seen, because the and trails to finished food sources are characterized by the
living ants have found food. It is interesting to know the fact that they cover a particularly short or long distance, they
reason why from this round 50 onward the difference occurs. can be seen by the peaks in the graph, because in those cases
This we explain for each algorithm. First, for RW the the ants do not find any food (especial in ant mills). In these
average age increases from 24 to 60 rounds between time 50 three plots we have those peaks in the PSA and in our
and 100. From here onward and till the end of the simulation PSAMP. However, in RW they do not occur because the ant
the average age fluctuates between 40 and 60 rounds. mill cannot occur with the random-walk search algorithm.
However, the fluctuation is not even because of the Thus, this curve serves as a good comparison of how such a
randomness nature of the RW. graph should look without ant mills. What we can also see in
Next, for PSA we see a rapid increase of the average age these three plots is that the number of dead ants (especially
till 120 rounds. More ants found food and the ages are higher the height of the peaks) decreases with higher evaporation.
than the starting value. However, suddenly the average age Let’s look at Figure 9. These high death rate peaks
decreases to almost 30 rounds. The reason for this characterize the ant mills and high frequented pheromone
occurrence is that the appearance of ant mills. Most of the trails that are leading to vanished food sources. The peaks on
food sources have been found and thus more pheromones are a distance 1 and 2 are ant mills. These ants covered a
spread. As a result, ant mills are developed, causes that the distance of only 1 or 2 for the last 10 steps before dying. The
ants are starving, and the average age decreases suddenly. other high peaks at distance 7 and 9 are coursed of trails to
This happens until round 350. From here the average age vanished food sources. Before the ants’ fateful death, they
fluctuate between 20 and 30 rounds through the entire run straight forward for the last 10 steps. None of the
simulation. The reason is that the nest is full of pheromones algorithm have a high effect on these ant behaviors. Thus,
trail that brings the ants to ant mills. Thus, the death rate and the ant mill is not the only problem. Pheromone trails to
the new reborn rate of the ants become constant and thus for vanished food causes death, too. Thus, a solution is that trails
a constant fluctuation of the average age. to gone food sources should evaporate after 25 rounds (4%
Finally, the average age of the ants that are using PSAMP evaporation) thus no problem as dying ants in the first 25
seems at the first glance to be better compared to the others. rounds to not leave pheromones.
The ants age increases, like PSA, rapidly at the beginning. 4) The Effect of Different Evaporation Rate: Finally,
However, it is not decreasing suddenly. The average age let us look at Figure 11 to examine the aspect of different
starts to vary between 70 and 90 rounds. Thus, this shows the evaporation rate. In this graph, the total average age of the
effect of the ant mill prevention if we compare PSAMP with ant colony is shown with different evaporation. The value for
PSA. Nevertheless, after round 650 the average age RW is not changing, hence it is independent of the
decreases and becomes close to RW. The reason is that all evaporation rate. However, the values for PSAMP and PSA
the food sources have been found and the pheromones trail changes, because the effect of the pheromone trail
towards them are established. As a result, the ants are evaporation has an effect of the ant’s movement and age.
preventing the ant mills by escaping, choosing randomly new Starting with an evaporation rate of 0% PSAMP starts with a
directions rather to deliver food. Thus, they average age of higher average rate compared to the other ones. This is the
PSAMP becomes equally to RW. result of the ant mill prevention.
2) Efficiency Comparison: The food delivery tells us
how efficient the three search algorithms are. Fig. 7 shows
the food delivery in percentage. At the first glance, we see
after round 250 that the food delivery for PSA becomes

28

Authorized licensed use limited to: Carleton University. Downloaded on September 22,2020 at 10:42:19 UTC from IEEE Xplore. Restrictions apply.
Figure 6. Average age with an evaporation of 4%. Figure 7. Carried food with an evaporation of 4%.

Figure 8. Evaporation rate of 2%.


Figure 9. Evaporation rate of 4%.

Figure 11. Average ant age in relation to evaporation.


Figure 10. Evaporation rate of 6%.

However, the average age PSAMP starts to decrease the same effect as PSAMP happens i.e. the higher the
higher the evaporation rate becomes. This continuous until it evaporation rate becomes, PSA turns to be like RW.
becomes almost the same as RW. This is because of the
PSAMP escape function that tells the ants to move in a V. CONCLUSION & FUTURE WORK
random direction for certain steps. Hence, the pheromone In this paper, we investigate the occurrence and resolving
trails evaporate faster the more ants with PSAMP are moving of ant mill issues in ant-based algorithms. We simulate ants
in a random walk. Additionally, if they see other ants’ groups that are searching for food and deliver food pieces to their
they will try to escape, too. That is the reason why at the end nest. We compare three types of search algorithms: Random
PSAMP behaves like RW. In comparison, the PSA average Walk (RM), Pheromone-Based (PSA), and Pheromone-
age becomes with an evaporation rate of 40% the highest. Based with ant mill Prevention (PSAMP). First, we show
This is because PSA is not using the prevention ant mill and that a modification of the random walk search with
thus the ants are using any pheromone trail, independent if it pheromones turns to be inefficient. The reason is that a
is an ant mill or not. Additionally, the ant mills break, pheromones-based search algorithm causes ant mills which
because of the fast evaporation rate. As a result, the PSA ants makes the modification inefficient. As a result, we develop a
are using any pheromone trail and becomes faster to food pheromone-based search algorithm with ant mill prevention
delivery compared to PSAMP. Nevertheless, for PSA the (PSAMP). Our PSAMP is a solution to prevent ant mills and

29

Authorized licensed use limited to: Carleton University. Downloaded on September 22,2020 at 10:42:19 UTC from IEEE Xplore. Restrictions apply.
makes collecting food in a swarm more successful. We show [3] P. S. Mogre, K. Graffi, M. Hollick, and R. Steinmetz, ―Antsec,
that PSAMP reduces the rate of death ants, increases the ants watchant, and antrep: Innovative security mechanisms for wireless
mesh networks,‖ in 32nd IEEE Conference on Local Computer
average age and food delivery efficiency. However, by Networks (LCN 2007). IEEE, 2007, pp. 539–547.
increasing the evaporation rate, PSAMP becomes more [4] T. C. Schneirla, A unique case of circular milling in ants, considered
inefficient compare to PSA. Additionally, with evaporation in relation to trail following and the general problem of orientation.
rate close to 100% both PSA and PSAMP acts like RW, American Museum of Natural History, 1944. [Online]. Available:
because the pheromones are disappearing much faster. http://hdl:handle:net/2246/3733
However, our algorithm does not generate a certain [5] F. Delsuc, ―Army ants trapped by their evolutionary history,‖ PLOS
number of ant mills by direct influence and cannot detect ant Biology, vol. 1, no. 2, 2003.
mill on its own. For example, the PSA is not able to detect a [6] A. Colorni, M. Dorigo, V. Maniezzo et al., ―Distributed optimization
by ant colonies,‖ vol. 142, 1991.
trail to the food but reacts only to a few pheromone strengths
in the immediate vicinity. The PSAMP does not recognize [7] I. D. Couzin and N. R. Franks, ―Self-organized lane formation and
optimized traffic flow in army ants,‖ Proceedings of the Royal
ant mills but reacts only to the ant density. The reason why a Society of London B: Biological Sciences, vol. 270, no. 1511, 2003.
pheromone-based search algorithm can solve a task, such as [Online]. Available:
finding tracks or avoiding possible ant mills, lies in the http://rspb:royalsocietypublishing:org/content/270/1511/139
swarm behavior. Collectively a task is solved in the swarm [8] L. Li, H. Peng, J. Kurths, Y. Yang, and H. J. Schellnhuber, ―Chaos–
and not each individual separately. For example, with a order transition in foraging behavior of ants,‖ Proceedings of the
strong evaporation of pheromones, a single ant may no National Academy of Sciences, vol. 111, no. 23, 2014.
longer find its self-laid trail to the food source. Only other [9] S. Sommer, C. von Beeren, and R. Wehner, ―Multiroute memories in
desert ants,‖ Proceedings of the National Academy of Sciences, vol.
ants in contact with this trail reinforce it for such a long time 105, no. 1, 2008.
that a complete connection between nest and food can be
[10] A. R. Cheraghi, K. Actun, S. Shahzad, and K. Graffi, ―Swarm-sim: A
maintained. The above-mentioned points are some future 2d & 3d simulation core for swarm agents,‖ in 3rd Int. Conf. of
works. Additionally, other types of search and stigmergy Intelligent Robotic and Control Engineering (IRCE 2020), 2020.
algorithm from animals’ nature, e.g. of bees (see [12], [13]) [11] A. R. Cheraghi and U. Duesseldorf. (2020) Swarm-sim. [Online].
can be developed and compared with PSA and PSAMP. Available: https://gitlab:cs:uni-duesseldorf:de/cheraghi/swarm-sim
Nevertheless, we consider only normal ants, nature has many [12] D. T. Pham, A. Ghanbarzadeh, E. Koç, S. Otri, S. Rahim, and M.
other types of ants. For example, the army ants. Their nest in Zaidi, ―-the bees algorithm—a novel tool for complex optimisation
nature is moved by the army ant and produces a new breed. problems,‖ 2006.
In addition, it would be interested to know how to manage [13] Z. S. Liang, T. Nguyen, H. R. Mattila, S. L. Rodriguez-Zas, T. D.
Seeley, and G. E. Robinson, ―Molecular determinants of scouting
and prevent misbehavior of ants in the swarm through behavior in honey bees,‖ Science, vol. 335, no. 6073, 2012.
different ―behavior anomalies of single individuals‖? Our [14] K. Graffi, P. S. Mogre, M. Hollick, and R. Steinmetz, ―Detection of
findings on secure ant algorithms in [3], [14], [15] can be colluding misbehaving nodes in mobile ad hoc and wireless mesh net-
further used to analyze the problem of the emergence of the works,‖ in IEEE GLOBECOM 2007 - IEEE Global
ant mills. Interesting is to add a direct communication with Telecommunications Conference, 2007, pp. 5097–5101.
the pheromones and to see the new results [16]. Another [15] P. S. Mogre, K. Graffi, M. Hollick, and R. Steinmetz, ―A security
framework for wireless mesh networks,‖ Wireless Communications
example, ants’ movement patterns could be like a breath or and Mobile Computing, vol. 11, no. 3, pp. 371–391, 2011.
depth-first search [17]. Further, how about if they could
[16] A. R. Cheraghi, J. Zenz, and K. Graffi, ―Opportunistic network
smell the food without knowing from where the smell is behavior in a swarm: Passing messages to destination,‖ in 2020 5th
coming from like the phototactic movement in [18]. Last but Asia-Pacific Conference on Intelligent Robot Systems (ACIRS).
not least, the ants move like a flock [19] and coat the food IEEE, 2020.
source from each side [20] to transport it together away. [17] A. R. Cheraghi, A. Abdelgalil, and K. Graffi, ―Universal 2-
Additionally, the ant queens searches for food, scans the dimensional terrain marking for autonomous robot swarms,‖ in 2020
5th Asia-Pacific Conference on Intelligent Robot Systems (ACIRS).
food source, and brings all the other ants to coat the food IEEE, 2020.
source [21].
[18] A. R. Cheraghi, F. S. Vila, and K. Graffi, ―Phototactic movement of
battery-powered and self-charging robot swarms,‖ in 2020 5th Asia-
ACKNOWLEDGMENT Pacific Conference on Intelligent Robot Systems (ACIRS). IEEE,
We would like to thank the Arab-German Young 2020.
Academy of Sciences and Humanities for funding this [19] A. R. Cheraghi, A. B. Janete, and K. Graffi, ―Robot swarm flocking
research as well as our collaboration partner Dr. Ahmed on a 2d triangular graph,‖ in 2020 5th Asia-Pacific Conference on
Intelligent Robot Systems (ACIRS). IEEE, 2020.
Khalil for the valuable discussions on future nanorobot
[20] A. R. Cheraghi, G. Wunderlich, and K. Graffi, ―General coating of
technologies. arbi-trary objects using robot swarms,‖ in 2020 5th Asia-Pacific
Conference on Intelligent Robot Systems (ACIRS). IEEE, 2020.
REFERENCES
[21] A. R. Cheraghi and K. Graffi, ―A leader based coating algorithm for
[1] R. Beckers, O. Holland, and J.-L. Deneubourg, ―From local actions to simple and cave shaped objects with robot swarms,‖ in 2020 5th
global tasks: Stigmergy and collective robotics,‖ in Artificial life IV, Asia-Pacific Conference on Intelligent Robot Systems (ACIRS).
vol. 181, 1994, p. 189. IEEE, 2020.
[2] M. Dorigo and L. M. Gambardella, ―Ant colonies for the travelling
salesman problem,‖ biosystems, vol. 43, no. 2, 1997.

30

Authorized licensed use limited to: Carleton University. Downloaded on September 22,2020 at 10:42:19 UTC from IEEE Xplore. Restrictions apply.

You might also like