Proceedings of the 2010 IEEE

International Conference on Robotics and Biomimetics

December 14-18, 2010, Tianjin, China

Multi-robot System Based on Model of Wolf Hunting Behavior to

Emulate Wolf and Elk Interactions
John D. Madden, Ronald C. Arkin, Fellow, IEEE, and Daniel R. MacNulty

Abstract Wolves are one of the most successful large

predators on earth. Their success is made apparent by their
presence in most northern ecosystems. They owe much of this
success to their generalized hunting behavior which allows
them to quickly and effectively adjust to different species of
prey. The success of this hunting behavior for wolves is the
inspiration for a project to bestow this behavior onto a system
of robots with the hopes that they might utilize the apparent
strengths of the behavior to achieve their own success.

I. INTRODUCTION
As part of a project for the Office of Naval Research,
models of behavior from biology are being used to develop
heterogeneous unmanned network teams (HUNT) of robots.
An earlier study in this project used lekking behavior from
prairie chickens to develop a basis for structuring groups of
robots [1]. The results of the lekking study were used as a
starting point for the current study with wolves. Our group
includes a biologist (D.R. MacNulty), who specializes in

wolf behavior and has conducted extensive studies of wolves

in Yellowstone National Park (YNP). The model of wolf
behavior used for this project was based on observations
from these studies.
This is not the first project to use wolf behavior as a
model for robots. Weitzenfeld et al. created packs of robot
wolves where alpha wolves would lead and beta wolves
would follow [3]. Our project breaks from this work in two
significant ways. First, Weitzenfeld assumed a tight structure
to exist in the coordination of wolf packs; however, direct
observations of wolves hunting elk in YNP indicate no
obvious pattern of coordinated hunting behavior [4]. For this
reason, our wolf model has been given no hard constraints to
keep them together. Second, Weitzenfeld also assumed roles
such as alpha wolf and beta wolves, to control the position
and actions of each individual throughout a hunt. The
observations on which our models are built show that roles
do exist but they may change in an ad hoc manner and are
based on physical abilities and not on a pre-existing
dominance hierarchy. The field observations, from which
this understanding of wolf hunting behavior is based,
indicate that there is a lack of explicit coordination between
the wolves. Their group behavior is evidently not a well
structured set of strategies but rather generalized rules of
thumb that are used to react to the preys escape behavior in
order to minimize the risk of injury to themselves [4].
II. WOLF BEHAVIOR

Figure 1. Elk are the primary food source of wolves in Yellowstone

National Park and wolves hunting elk are the focus of this project. [2]

This work was supported in part by the Office of Naval Research under
MURI Grant # N00014-08-1-0696.
J.D. Madden and R.C. Arkin are with the Mobile Robot Laboratory,
College of Computing Georgia Institute of Technology, 85 5th ST NW,
Atlanta, GA, 30332. email {jmadden,arkin}@gatech.edu
D.R. MacNulty is with the Department of Ecology, Evolution, &
Behavior University of Minnesota, 1987 Upper Buford Circle, St. Paul,
MN, 55108. email: [email protected]

978-1-4244-9318-0/10/$26.00 2010 IEEE

A. Individual Properties
Wolves are able to consume a variety of prey from mice
to moose because of their generalized skull morphology.
And the apparent lack of coordination could be an advantage
in that it allows wolves to hunt over a range of conditions
irrespective of any requirement to coordinate. They use a
few basic heuristics (rules of thumb), e.g., attack while
minimizing the risk of injury with no overall hard behavioral
constraints on actions [4]. This makes their behavior very
flexible and allows them to quickly and easily make the
transition between different species of prey, such as elk in
the summer and bison in winter when elk migrate. One
observation in Yellowstone National Park, involved a pack
of wolves that had been hunting bison, moved into a new
valley, and immediately started hunting elk. This serves as a
testament to the adaptability of wolf hunting behavior, and a
powerful clue regarding their success in such varied
environments.

1043

B. Breakdown of a Wolf Hunt

As is the case for most large carnivores, the predatory
behavior of wolves is composed of multiple phases of
behavior or foraging states. Traditionally, only three states
are considered: search, pursuit, and capture [5]. In this
research, however, a modified ethogram with six states has
been adopted: search, approach, watch, group attack,
individual attack, and capture, as proposed in [6]. Here,
MacNulty concluded that the additional states represent
functionally important behaviors, and for robotics, this
more detailed ethogram lends itself more easily to software
implementation. The wolf packs studied to form this
ethogram were located in Yellowstone National Park,
hunting elk and American Bison. The focus of the first phase
of the robotics implementation involves a model of wolves
hunting elk. The following is a description of a typical hunt
with wolves and elk. A diagram showing the typical
progression through foraging states is given below in Figure
2.

Figure 2. The progression of transitions between states seen in a typical

hunt with wolves and elk.

TABLE 1
Foraging States for Wolf Hunting Behavior
Foraging State

Description

Search

Traveling without fixating on and moving

toward prey

Approach

Fixating on and traveling toward prey

Attack Group

Running after a fleeing group or lunging at a

standing group while glancing about at different
group members (i.e., scanning)

Attack Individual

Running after or lunging at a solitary individual

or a single member of a group while ignoring all
other group members

Capture

Biting and restraining prey

When a hunt is initiated, the wolf pack heads out from its
den or resting site and begins searching for prey. Hunger
motivates the initiation of a hunt [8]. What direction the
wolves go and to what extent they are willing to travel are
dependent on their experience of prior successes and
failures. As they search they make use of their strong senses,
using the wide range of their lateral vision and their movable
ears, to scan the landscape for potential prey. Once prey has
been located, they start approaching.
Assuming that that the pack has located a relatively
stationary herd of elk, the wolves approach at moderate
speed. In general, wolves do not sneak up on their prey, nor
do they target a specific individual from the herd until after
the herd begins running. Species that use this approach
strategy are known as cursorial predators and it is the
principal difference separating their hunting behavior from
that of other large predators such as lions [6]. In response to
approaching wolves elk will either stand their ground or to
run away. Elk most commonly run away which usually leads
to the attack group state.
As the prey quarry run away, they split up into groups
headed in different directions and the wolves must also split
up to follow as many as they can. During this stage of the
hunt the wolves are scanning through the groups of prey,
trying to locate the weakest individual that will provide the
best opportunity for a kill. An advantage of running the
animals to exhaustion is that it creates opportunities for the
prey animals to make a fatal mistake (i.e., tripping). It also
provides a useful test of performance by which the wolves
can evaluate which animal is the weakest [8]. When a weak
animal is detected by a wolf, that wolf then transitions to the
attack individual state.
The attack individual state is characterized by
intensified pursuit and greater focus on the targeted prey
individual. Other wolves may see the pursuit of this wolf
and join in, but that is not necessarily the case. Coordination
of multiple wolves (or lack thereof) is discussed in the next
section. The goal of this behavioral state is for the wolf to
get close enough to the prey to begin biting it in an attempt
to bring it down. Whether it is a single wolf or a number of
wolves, biting the prey signifies a transition to the capture
state.
The ultimate goal of the capture state is killing the prey.
If the prey animal is small (i.e., a calf) the first wolf may
attack the throat directly since it can easily handle the animal
by itself. If the prey is larger and there are many wolves,
they will often bite at the hind legs and rump attempting to
slow their prey down before grabbing the neck. This project
is not concerned with the mechanics of how wolves bring
down prey but it is important to note that there are
differences in attacking different prey. If the prey truly was a
weak individual, the wolves will most likely complete a
successful kill, but if they had misperceived a strong animal
as weak, they may fail and either give up on the hunt or
transition back to an earlier state.
The narrative of a hunt that has just been related gives a
general idea of how many specific individual hunts progress
through these foraging states; however, it is often not this
clear cut. Many other transitions are possible aside from the

1044

seemingly linear straightforward progression from search, to

approach, then attack group, attack individual, and finally
capture. For instance, wolves primarily attacked groups after
approaching but they also sometimes attacked elk groups
immediately after discovering or watching the group [6].
MacNulty et al. compiled their statistical observational data
of state transitions (Table 2) where the tabular values
represent the probability of transition between states. Notice
that the transitions chosen for the description of the linear
hunt above are those of highest probability in the table.

unaware that they need the others to help them take down
the large prey because this is most often the case. It may not,
however, be required that wolves need help to take down
any of their usual prey. It is proposed that one of the biggest
reasons that large terrestrial predators do not use group
coordination is that they do not necessarily need it. Solitary
hunters have a high success rate, roughly 21% for most large
carnivores [MacNulty unpublished data].

III. IMPLEMENTATION OF WOLF BEHAVIOR

TABLE 2
Probabilities of Transitions Between States From [6]
Following State
Preceding
State

Search

Approach

Watch

Attack
Group

Attack
Individual

Capture

Search
Approach

.00
.09

.68
.00

.00
.12

.31
.69

.01
.09

.00
.01

Watch

.32

.35

.00

.27

.06

.00

.24

.09

.03

.13

.51

.00

.16

.06

.02

.16

.08

.52

Attack
Group
Attack
Individual

Thus far, the watch state has been neglected as it is a

rare state for wolves to enter when attacking elk; as seen in
the table above, the highest probability of entering the
watch state is 12% from approach. For this reason, the
watch state has been left out of the ethogram for our
robotics implementation described in Section III.
C. Coordination or Lack Thereof
Wolves are generally perceived by the public to be highly
coordinated hunters using strategies and teamwork to bring
down large prey. Over two thousand hours of observed wolf
behavior in Yellowstone Park seem to prove otherwise [4].
According to these observations, wolves not only show no
signs of planned strategies but also little to no noticeable
communication while hunting. This is evidenced by the fact
that wolves hunting the same herd do not make transitions
between states together (i.e., one may find a weak prey and
transition to attack that individual while the others remain in
an attack group). The disparity in these transitions goes so
far as to see one wolf having killed an animal and begin
eating it while the others persist in the attack group state.
Furthermore, in this last example, the wolf that made the kill
did not appear to make any attempt to signal the others of its
success.
The seemingly coordinated wolf hunting behavior is most
likely the result of byproduct mutualism where each
individual is simply trying to maximize its own utility. It is
hypothesized that wolves see the fact that other wolves are
chasing an elk as a sign of weakness of that prey animal and
from that stimulus determine that they have the best chance
of a meal if they join in the pursuit of that animal. Even far
greater size of their prey does not force wolves to rely on
teamwork; according to MacNulty, some aggressive wolves
would attack even large bison alone. It is possible that such
wolves simply assume the others will help them, or they are

Experiments for this project were conducted with

simulated robots in MissionLab1, a software package
developed by the Mobile Robotics Laboratory at Georgia
Tech [9,10]. MissionLab provides a graphical user interface
where the user specifies behavioral states that control each
robots actions, and perceptual triggers that control the
transitions between states, yielding a finite state acceptor
(FSA). The behaviors created for the current project can be
combined with other pre-existing behaviors such as obstacle
avoidance, moving toward an object, or noise (random
wandering). This allows for assemblages of behaviors to be
created and connected in the FSA to create arbitrarily
complex missions [11,12].
Reducing the overall hunting behavior of wolves into the
five foraging states related earlier, facilitated implementation
where each behavioral state represents a corresponding state
for the robot. These states, together with a few others added
for initial configuration and termination of experiments,
were used to create the FSA shown in Figure 3. The
perceptual conditions that must be met in order for a wolf to
transition from one state to another are known as releasers.
For instance, for the wolf to switch from the search state to
the approach state the wolf must of necessity have found
prey to approach; therefore, we say the presence of prey is a
releaser to transition to the approach state. These are
encoded as perceptual triggers in MissionLab. A list of the
releasers used in this implementation and the transitions they
facilitate are given in Table 3. The system of releasers would
normally be enough to define the transitions in a MissionLab
FSA except that often, multiple transitions are possible from
the same state to many others. In nature, what decides which
transition is chosen is a combination of situational factors
such as the number of wolves in the pack, the number of
prey individuals in the herd, terrain features, as well as the
wolfs individual attributes such as age, weight, and
personality (i.e., aggressive individuals are more likely to
move more quickly toward capture). While these factors will
be incorporated directly in later work, for now their affect
was indirectly computed by using the probabilities of
transitions of observed wolf behavior described earlier in
Table 2.

1045

MissionLab is freely available for research and educational purposes at:

www.cc.gatech.edu/ai/robot-lab/research/MissionLab/

Figure 3. Finite State Acceptor for Wolf behavvior with five foraging state (watch state removed) as well as initial and final states for
experimentation purposes. The stop states con
nnected to each foraging state are dummy states to facilitatte the probabilistic trigger.

TABLE 3
List of Releasers and Transitioons
Releaser

Transitions poossible

Prey Found

S A, G, I

Prey Lost

A, G, I, C S

Multiple Prey

S, A, I, C G

Prey Running

A G, I

Prey Stopped

G, I A ; A, I C

Weak Individual Identified

GI

Prey Close

G, I C

Search, Approach, Attack Group, Attack Individual, and Capture

abbreviated: S, A, G, I, C, respectively.

These probabilistic triggers for the FSA in Figure 3 were

created in the following fashion: for eachh state there is a
Control trigger leading from that state too the search state
and then another trigger leading to every otther state to yield
a complete graph encompassing all posssible transitions
between the 5 major behavioral states. T
The probabilities
from Table 2 were entered as parameters iinto the Control
trigger at build time, and at run time this triggger would check
which transitions had their releasers satisffied. A weighted
roulette wheel was then created by normaliizing the satisfied
probabilities such that they added to one annd then a random
number generated between zero and one w
would decide the

transition to take. Once a transition is selected, the condition

for the corresponding trigger is satisfied and that behavioral
transition occurs. If the transition selected was from any
t
condition would
state to the search state, the control triggers
be satisfied and a transition to search would occur. Also
incorporated into the control triggerr was a timer to force the
wolves to stay in a state for a minim
mum length of time based
on the observed average time wollves spent in each state
(D.R. MacNulty, unpublished data)). Without this hysteresis
feature the wolves would constantly
y alternate back and forth
between states for which the releaseers are present (a form of
behavioral dithering).
Each state in the diagram above is a combination of the
constituent pre-existing behaviors: move to object, wander,
and avoid obstacles. Move to object creates an attraction
vector from the robot to the objecct selected. In the search
state the selected object was friend
dly robots, which in this
case represent other wolves, in all other states the selected
object was elk. MissionLab usees built-in vector and
simulation specific functions, imp
plemented in C++ code.
The move to object behavior createes a vector directed from
the center of the robot to the center of the selected object so
long as the selected object is within range to be detected.
The wander behavior creates vecttors of random direction.
While in the search state, this was useful
u
to give the wolves
the ability to explore their environm
ment for prey. In all other
states the wander behavior was used to help the wolf
overcome situations of indecision which may occur, for
instance, when a wolf is exactly thee same distance between
two elk.
vior was added to prevent
Finally, the avoid obstacles behav

1046

collisions. As the robot detects obstacles, repellant vectors

are created, radiating away from the obstaacle. This allows
the robot to move around obstacles while searching and in
pursuit of prey. The final vector that determ
mines the robots
movement is the resultant of the vectors created by these
constituent behaviors. The degree to whicch each behavior
had an effect on the resultant vector is ddependent on the
gain, entered by the user, for each behaavior. All of the
parameters, including gains, for each state ccan be found in a
table in the appendix.
Although the elk being preyed upon maay have defensive
behavioral strategy and coordination, thhe focus of the
research to date has been on the hunting behhavior of wolves.
Therefore, the behavior of the elk was sim
mplified with their
reaction to the approaching wolves as simplly either stopping
or running away in a direction oppositee to the wolves
approach. To create a range of test sceenarios, the elks
behavior before the approach of wolvess was varied to
simulate situations where the elk are iniitially stationary,
moving back and forth between multiple grazing areas, or
wandering around. An example of the FSA to control the elk
behavior for moving back and forth betw
ween two grazing
areas is given in Figure 4, showing a sim
mple modification
that switches between an elk stopping upoon seeing wolves,
or running away.

Figure 4. Example photo of potential wo

olf hunt.

IV. SIMULATION RESULTS

R
The MissionLab wolf pack simulaations examined multiple
scenarios that were commonly obseerved with wolf hunts in
Yellowstone National Park. The underlay used for the
simulations, is from the Lamar Valley in Yellowstone
National Park where many of the acctual observations of wolf
hunts were taken.

Figure 4. Finite State Acceptor of Elk behavior. If the run away

behavior is desired, remove the dotted trigger to thee final stop state.

This project was conducted entirely with simulated robots;

however, future work expects to movee the system to
physical robots. The platforms expected too be used are: (1)
WowWee Rovio Wi-fi robots, and (2) iRoobot Creates with
the Element BAM (Bluetooth Adapter Moodule) as pictured
below.

Figure 5. (Top): The hunting area used for

f experiments with starting
location on the left, and grazing areas in the right and bottom of the
map. (Bottom): Close-up showing transittions from search to capture
for one wolf hunting one elk with stop beh
havior. The Straight Diagonal
line is the preys path between grazing
g areas. Underlay used for
simulation of wolves in the Lamar Valley
y, Yellowstone National Park
where many observations were recorded [1
13].

1047

The first scenario was one on one betweeen a wolf and an

elk, with the elk moving back and forth between grazing
areas and stopping when it perceived the wolf. The
progression of the wolf through the foraaging states was
recorded for each run so that frequency off transitions could
be tabulated for comparison between stattes and with the
original observed probabilities. An exampple of a hunt run
with these parameters is given in Figuree 5 with starting
location, grazing areas, and transition pooints in the hunt
labeled. The next scenario was created bby modifying the
preys behavior to run away, rather thaan stop, when it
perceived the wolf. An example of this, aalso showing the
wandering pattern of the wolf, is given in Figure 6. A third
scenario had the same run away behaviorr for the prey but
with one wolf and three elk Figure 7. A final scenario
involved two wolves and three elk as seenn in Figure 8. For
this scenario, the number of runs that both wolves ended up
killing the same elk is compared to that of tthe wolves killing
different elk. This comparison is made foor both situations
when the wolves discovered the elk togethher and when the
wolves discovered elk separately. Tweenty runs were
completed for each scenario and the tabbulated results of
these runs are given in Table 4.

TABLE 4
Results From Wolf Siimulations
One Wolf and One Elk (Stop)
Preceding
State
Search
Approach
Attack
Group
Attack
Individual

Search
-.00

Follow
wing State
Attack
Approach Atttack
Grroup Individual
.05
.95
.00
.88
-.00

Capture
.00
.12

.00

.00

.00

.00

.00

.08

.12

.00

.15

.67

One Wolf and One Elk (Run Away)

Preceding
State
Search
Approach
Attack
Group
Attack
Individual

Search
-.07

wing State
Follow
Attack
Approach Atttack
Grroup Individual
.00
1.00
.00
.93
-.00

Figure 6. Transitions for one wolf hunting one elk with behavior set to
run away from wolf. Random behavior of wolf seaarch can also be seen
previous to transition to approach.

.00

.00

.00

.00

.07

.21

.00

.23

.49

1048

Capture
.00
.00

.07

.26

.13

.54

.00

.09

.15

.22

.13

.41

Result of hunt
Discover elk together, kill same elk
Discover elk together, kill different elk
Discover elk separately, kill same elk
Discover elk separately, kill different elk

Figure 7. Close-up showing transitions for one woolf hunting three elk
with behavior set to run away from wolf.

.00
.00

.00

One Wolf and Three Elk (Run Away)

wing State
Follow
Preceding
Search Approach Atttack
Attack
State
Grroup Individual
.03
-.73
.24
Search
.21
.04
-.75
Approach
Attack
Group
Attack
Individual

Capture

Runs
4
9
3
4

%
20
45
15
20

w
and three elk. The
The final scenario involved two wolves
purpose of this scenario was to examine how multiple
wolves react to multiple prey. Th
he wolves were given a
slight attraction to one another thro
ough the move-to-object
behavior to simulate actual hunts wh
here the wolves generally
start relatively together. This behaviior made the discovery of
prey with the wolves together mo
ore common. When the
wolves came across the prey togetheer, they most often killed
different prey individuals. This maay have been due to the
confined space of the map alllowing the capture of
individuals they may have otherwisee been lost and forced the
pursuer to join their pack memberr. When they discovered
the prey separately the results were split for killing the same
or different individuals.
VI. SUMMARY AND CONCLUSIONS
O

Figure 8. (Top): Close-up of transitions with two wolves approaching

three elk together and separating during pursuit. (Bottom): Complete
view of hunt after the two wolves split up and kill different elk.

V. DISCUSSION OF RESULTTS
Many of the resulting probabilities of traansitions from the
first scenario vary from the observed probabbilities because in
this scenario the prey would never run awaay whereas in the
wild, running is the most common reactiion for the prey.
Comparison of the resulting transition probbabilities from the
first and second scenarios show that by chhanging the prey
behavior from stopping when approacheed by a wolf to
running away, the change in transitions iss most notable in
the transitions leading to attack. The first annd third scenarios
show similar differences for the same reason. Comparison of
the second and third scenario reveals thatt adding multiple
elk to the hunt has a large effect on the trransitions for the
obvious reason that the attack group state iss only possible in
the third scenario where there are multiplee elk. This is the
most realistic scenario as the vast m
majority of the
observations in YNP were wolves hunting multiple elk. For
this reason, the results from only the thhird scenario are
compared to the observed data. The probabilities of
transitions were similar to those in the obbserved data with
the error for the primary four transitionns (SA, AG,
GI, IC) at 5%, 6%, 3%, and 11% respectively. Some
transitions showed higher errors. Simulatioon results showed
much lower probabilities for all transitionns leading to the
search state. This is most likely due to the hunts being
confined within boundaries that often afffected the elks
attempts to run away.

By modeling wolf hunting behav

vior as a set of foraging
states per the ethogram of MacNulty
y et al. 2007, and using a
system of releasers and roulette wh
heel with probabilities of
transitions, we were able to simulate interactions between
wolves and elk with a relatively high fidelity to what is
observed in the wild. The probabilitties of transitions over all
scenarios were similar to those in th
he observed data. This is
not surprising as the observed dataa was used in the system
that generated these results. Althou
ugh there would seem to
be an advantage in structured atttack strategies, the high
variability in behavior of the wolfs prey as well as the chaos
hase down, and kill one
inherent in attempting to locate, ch
from a herd of hundreds of running
g elk would quickly cause
strict strategies to breakdown. The loose
l
adoption of general
rules gives wolves the ability to react quickly and
effectively. The results of simulations done in this project
showed that the wolves were in faact reacting to the preys
behavior as evidenced by the changee in transitions due to the
prey stopping or running when attaccked by the wolf.
The purpose of this study is to determine
d
if high fidelity
biological models can provide utiliity for a range of multirobot applications, in this case with
h an emphasis on pursuitevasion tasks. In particular, noting that byproduct mutualism
can produce very robust results fo
or biological groups has
implications for the ability to red
duce communication and
planning requirements for robot groups, while still achieving
ur results to date support
purposeful missions. We believe ou
this goal. It is also worth noting thaat while the probabilities
for transitions are currently invariable as defined directly
r
could address
from the wolf model, future research
variations that may adapt to a range of different situations.
APPENDIX

mulas for behaviors and

This appendix provides the form
transitions as well as the associated
d parameters used in the
simulated wolves.
a) Probabilistic Transition: Traansitions based on the
existence of releasers and probabilities
p
to simulate
factors such as prey group size, wolf pack size, and
environmental properties.

1049





where:
= Probability of transitioning from the
current state to Staten
= Input probability of the above transition

taken from MacNulty et al. 2007

= Sum of probabilities of transitions with
releasers satisfied
= Resulting state of the transition
= Random number between 0.00 and 1.00
b) Move-to-Object: Variable attraction to selected object.
Used for attraction between pack members and
attraction from wolf toward prey.
Vmagnitude = Adjustable gain value
Vdirection = Direction from the center of the robot to the
center of the object, moving toward the object
c)

Avoid-obstacle: Repel from object with variable gain

and sphere of influence. Used for collision avoidance.







Vdirection = Direction from the center of the robot to the center

of the obstacle, moving away from obstacle
where:
max = Maximum obstacle detection sphere
d = Distance of robot to obstacle
r = Radius of obstacle
d) Noise: Random wander with variable gain and
persistence. Used for exploration and to overcome local
maxima, and minima.
Vmagnitude = Adjustable gain value
Vdirection = Random direction that persists for
specified number of steps

.7
Enemy
.4
.5
10
4

m
m

Wolf attack group assemblage

Move to object gain
Selected object
Wander gain
Avoid obstacle gain
Avoid obstacle sphere
Avoid obstacle safety margin

.8
Enemy
.3
.3
10
4

m
m

Wolf attack individual assemblage

Move to object gain
Selected object
Wander gain
Avoid obstacle gain
Avoid obstacle sphere
Avoid obstacle safety margin

1
Enemy
.1
.3
10
4

m
m

Wolf capture assemblage

Move to object gain
Selected object
Wander gain
Avoid obstacle gain
Avoid obstacle sphere
Avoid obstacle safety margin

.1
Enemy
0
.1
1.5
.5

m
m

Wolf approach assemblage

Move to object gain
Selected object
Wander gain
Avoid obstacle gain
Avoid obstacle sphere
Avoid obstacle safety margin

REFERENCES
[1]
[2]

[3]

[4]

[5]

[6]

[7]
[8]
[9]

Parameter
Wolf search assemblage
Move to object gain
Selected object
Wander gain
Secondary wander gain
Avoid obstacle gain
Avoid obstacle sphere
Avoid obstacle safety margin

Value
.2
Friendly
.7
.3
.5
10
4

Units

[10]

[11]
[12]
m
m

[13]

1050

B. A. Duncan, P. D. Ulam, R.C. Arkin, Lek Behavior as a Model for

Multi-Robot Systems
D. Smith, Wolves Chasing an Elk, National Park Service. Dec. 1,
2007,
<http://www.nps.gov/yell/photosmultimedia/photogallery.htm?eid=37
9961&aid=547&root_aid=547&sort=title&startRow=10#e_379961>
A. Weitzenfeld, A. Vallesa, H. Flores A Biologically-Inspired Wolf
Pack Multiple Robot Hunting Model, in Latin American Robotics
Symposium and Intelligent Robotic Meeting (LARS 2006), Santiago,
Chile, 26-27 Oct. 2006.
D.R. MacNulty, Smith, D.W., Vucetich, J.A., and Packer, C.,
Nonlinear Effects of Group Size on the Success of Wolves Hunting
Elk, (in review), 2010.
C.S. Holling, The Functional Response of Predators to Prey Density
and its Role in Mimicry and Population regulation, Memoirs of the
Entomological Society of Canada, 45: 1-62.
D.R. MacNulty, L.D. Mech, D.W. Smith, A Proposed Ethogram of
Large-carnivore Predatory Behavior, Exemplified by the Wolf,
Journal of Mammalogy, Vol. 88, No. 3 pp.595-605, June 2007.
R. Abrantes, The Evolution of Canine Social Behavior 2nd Edition,
Ann Arbor, MI: Wakan Tanka Publishers, 2005.
L.D. Mech, L. Boitani, Wolves: Behavior, Ecology, and
Conservation, University of Chicago Press. 2003.
Georgia Tech Mobile Robotics Laboratory, Manual for MissionLab,
Version 7.0, 2007.
D. MacKenzie, R.C. Arkin, J. Cameron, Multiagent Mission
Specification and Execution, Autonomous Robots, Vol. 4, No. 1, Jan.
1997, pp.29-57. Also appears in Robot Colonies, (eds.) R.C. Arkin, G.
Bekey, Kluwer Academic Publishers, 1997.
R.C. Arkin, Behavior-based Robotics, MIT Press, 1998.
R.C. Arkin, Motor Schema-Based Mobile Robot Navigation,
International Journal of Robotics Research, Vol. 8, No. 4, August
1989, pp. 92-112.
Figure reprinted from Google, in accordance with their guidelines
posted online. 2009 Google Imagery 2009 DigitalGlobe,
GeoEye, USDA Farm Service Agency, Map data 2009 Tele Atlas.