Mobile Robot Lab Home

HUNT Project

Directions to the Lab

Overview | Biological Basis | References | In Progress | Paper | Figures |

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.

  Biological Basis

A. Individual Properties

Figure 1:  Example of a wolf hunt.

   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.

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:  Transition between foraging states diagram.

Table 1:  Foraging states for wolf hunting behavior

   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 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.

Table 2:  Probabilities of transitions between states

   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 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].


[1] B. A. Duncan, P. D. Ulam, R.C. Arkin, “Lek Behavior as a Model for Multi-Robot Systems” [2] D. Smith, “Wolves Chasing an Elk,” National Park Service. Dec. 1,2007,
[3] 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.
[4] 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.
[5] 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.
[6] 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.
[7] R. Abrantes, “The Evolution of Canine Social Behavior 2nd Edition,” Ann Arbor, MI: Wakan Tanka Publishers, 2005.
[8] L.D. Mech, L. Boitani, “Wolves: Behavior, Ecology, and Conservation,” University of Chicago Press. 2003.
[9] Georgia Tech Mobile Robotics Laboratory, “Manual for MissionLab,” Version 7.0, 2007.
[10] 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.
[11] R.C. Arkin, “Behavior-based Robotics,” MIT Press, 1998.
[12] R.C. Arkin, “Motor Schema-Based Mobile Robot Navigation,” International Journal of Robotics Research, Vol. 8, No. 4, August 1989, pp. 92-112.
[13] 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.

  In Progress

We are currently working on developing and running real robot experiments.


Our submission to ROBIO 2009 can be found at:
Doc version
PDF version


Several videos of simulation runs:

Simulation: One wolf, one elk, elk stopping behavior
Simulation: One wolf, one elk, elk runaway behavior
Simulation: One wolf, Three elk, elk runaway behavior
Simulation: Two wolves, Three elk, elk runaway behavior


The figures and tables from the paper can be found in the list below:

Figure 1:  Example of a wolf hunt.

Figure 2:  Transition between foraging states diagram.

Figure 3:  FSA for wolf behavior.

Figure 4:  FSA for elk behavior.

Figure 5:  Description of map and one on one hunt.

Figures 6 and 7:  Transitions for one on one and one on three hunts.

Figure 8:  Transitions for two on three hunt

Table 1:  Foraging states for wolf hunting behavior

Table 2:  Probabilities of transitions between states

Table 3:  List of releasers and transitions

Table 4:  Results from wolf simulations