Learning from Nature: Applying Biomimetic Approaches to Military Tactics and Concepts

Lashon Booker

orrowing design ideas from Mother Nature, medical technicians fabricate artificial limbs and organs. Using the principles of variation and natural selection, computer programmers solve complex problems by creating algorithms that can be used for optimization, scheduling, and engineering design. Copying the infection-fighting techniques of their biological cousins, artificial immune systems supply new approaches to computer security.

These applications are all examples of "biomimetics," the design and implementation of artificial processes, devices, or systems that imitate some aspects of biological systems. MITRE is employing biomimetics to meet some of the most difficult challenges facing our customers. When traditional problem-solving techniques come up short, biological systems can provide metaphors that point the way toward innovative solutions.

Many of the difficulties facing MITRE's Department of Defense (DOD) customers arise from the need for timely and affordable design, analysis, and evaluation of new military systems and operational concepts. Modeling and simulation are becoming increasingly important technologies for addressing these problems. One of the key factors in making simulation technology effective in this role, though, is the ability of modelers and analysts to generate innovative twists on operational concepts, tactics, and possible threat responses. Any technique that enables humans to systematically examine a broader range of possible innovations, or suggests alternatives that may not have been considered, would greatly increase the effectiveness of these simulation-based activities. While it is not likely that the process of exploring new tactical behaviors can be completely automated in the near term, substantial progress has been made in automating some aspects of this process. Biomimetic computational techniques have proven to be useful tools in helping to achieve that progress.

A recent example can be found in the research into swarming tactics conducted by the Joint Experimentation Directorate (J9) of the U.S. Joint Forces Command. Military swarming refers to a maneuver in which forces and firepower converge on a target force simultaneously from all directions. The DOD expects swarming tactics to play an important role in future battlefield environments involving small, dispersed, autonomous units operating independently using networked, decentralized command and control. This tactic is especially important for rapid reaction forces that need to avoid sustained direct-fire battles and rely on elusiveness for survivability.

Solutions from Swarming Ants

To understand how to achieve the necessary coordination for a swarming maneuver, military researchers are studying the behaviors of bird flocks, animal herds, and insect colonies. For example, a close study of how ants deposit chemical pheromones to coordinate foraging and other activities has proved extremely valuable. Through a series of simulation-based experiments, the DOD discovered that pheromone-based algorithms achieve impressive results in controlling a swarm of unmanned aerial vehicles conducting an attack against critical mobile targets.

We are using our biomimetic expertise to take a step beyond studies of swarming techniques like those conducted by J9. It is a daunting challenge to find efficient ways to specify the individual behaviors for each of the entities in a swarm. The pheromone algorithms, for example, are specified by a complex set of equations controlling the dispersal and interactions of the simulated chemical pheromones. It is not always easy to find the "right" settings for these control equations because the emergent behaviors of the swarm as a whole can be hard to anticipate. Recent MITRE research, however, suggests that machine learning techniques may eventually be capable of generating the required behaviors automatically.

Previously, in a project called Automated Discovery of Innovative Tactics and Behaviors, a MITRE team developed and demonstrated a new technology for learning tactics and behaviors automatically based on experiences in a simulation. We addressed the learning problem by focusing on behavior sequences and performance feedback. Learning can be accomplished by trying different combinations of actions, observing the behavior sequences generated, and then choosing new combinations of actions in ways that make desirable behavior sequences more likely. This approach transforms the problem of behavior discovery into a form that can be addressed by adaptive control and evolutionary computation methods.

The initial focus of our swarming research was on learning behavior tasks in lattice swarm models. In lattice swarm models, individual agents are constrained to behave in a three-dimensional space defined by a discrete lattice. Each agent has a repertoire of actions it can use to move through this space and modify the environment. An agent's sensors detect information derived from local properties of the agent's current position in the lattice and the positions directly adjacent to it. Since each agent has only a local view of the overall activity in the swarm, some additional mechanism must be available to coordinate the collective behavior of the swarm. Pheromones are one example of such a mechanism.

Wasps' Nests

The behavior of wasps as they construct their sophisticated nests is a well-studied example of swarming behavior that can be modeled in the lattice swarm framework. The building activities of each individual wasp are directed purely by observing the nest components constructed by the other wasps in adjacent locations. (This phenomenon, dubbed "stigmergy" in 1959 by French scientist Pierre-Paul Grasse, explains how wasps, termites, and other insects can build relatively complex structures by taking their cues from the structure itself. As the nest is built, the insects observe its current state and change their behavior accordingly to build the next piece.)

Computer simulations designed to study this behavior have shown that lattice swarm models, in which insects are modeled as simple agents controlled by simple sets of rules, can generate structures resembling known wasp-nest architectures. We used one of these simulations to investigate the "reverse engineering" problem: learning rules to direct the simulated wasps to build structures having some desired properties. Preliminary experiments show that this can be done successfully. MITRE-developed algorithms enabled the wasps to learn nest-building behaviors that were more successful than previous attempts to solve this difficult problem.

We have also investigated the application of these ideas for DOD purposes. For example, we've explored how to use these learning techniques to generate behaviors for a group of simulated micro air vehicles conducting simple reconnaissance and surveillance tasks. While the results of these experiments were not conclusive, they did show that the learning algorithms are capable of generating useful swarming behaviors to solve challenging tasks having military significance.

Overall, MITRE's biomimetic research has provided an important foundation for enabling its sponsors to conduct more careful investigation and testing of innovative tactical concepts, such as swarming.

For more information, please contact Lashon Booker using the employee directory.

Solutions That Make a Difference.®
Copyright © 1997-2013, The MITRE Corporation. All rights reserved.
MITRE is a registered trademark of The MITRE Corporation.
Material on this site may be copied and distributed with permission only.

Privacy Policy | Contact Us