Why Consider Many-Robot Systems?

Application Requirements

The general notion of "swarms" of "insect robots" has received attention in the popular and semipopular press [NYT91], [FREE91], [RICH92], and "microrobots" have even been joined by "nanorobots" in the popular imagination [DREX86]. The use of large numbers of small robots (with varying degrees of "large" and "small") has been seriously proposed for a wide variety of applications, including intelligent land mine deployment [MACE88], behind-the-lines military communications relaying [MEGA78], warehouse security sentry [FLYN87], ship hull cleaning [FLYN87], warehouse material handling [RICH92], lunar base construction [BROO90], gathering oceanographic data [STOM89], planetary surface exploration [MILL90], [BON90], [FLAM92], and aircraft carrier deck foreign object debris (FOD) disposal [GAGE92a].

The Mine Countermeasures (MCM) Application

Perhaps the ideal application for many-robot systems technology, however, is that of Mine Countermeasures (MCM). The detection and clearance of mines both ashore and in the surf zone constitute problems which have not yet been satisfactorily addressed, from minefield reconnaissance prior to an amphibious assault, through breaching operations, to humanitarian demining after hostilities have ceased. The obvious dangers to personnel, the problems associated with detecting non-metallic mines, and the difficulties of navigation and communication in the surf zone present difficult challenges to the technologist who would build an MCM system. However, recent (and continuing) advances in a number of technology areas offer hope that tools can now be developed to make MCM operations safer and more effective. For many applications, a system comprising a large number of identical and very inexpensive robotic search vehicles may provide an appropriate solution.

MCM operations is an application area which appears to be perfectly matched to the many-robot systems concept in the following respects:

The MCM environment is dangerous to humans; a robotic solution allows MCM operators to be physically removed from the hazardous area.
The MCM environment is also dangerous to machines; the use of multiple inexpensive robotic search elements minimizes the cost of lost system assets, and allows the mission to be performed by the remaining elements.
One important MCM task is the destruction of mines; using very cheap, deliberately expendible elements allows a one-element-per-mine approach.
Many mines must be dealt with; the use of many robots allows these targets to be prosecuted in parallel, rather than one at a time.

Note that the claim here is not that a many-robot approach is necessarily the best way to solve the MCM problem, but that the MCM problem is one of the best ones for many-robot technologists to consider addressing.

Biological Models

Nature provides outstanding models of functioning systems consisting of large numbers of more or less intelligent and mobile elements. One example is found in the flocking, herding, and schooling behaviors observed in many different types of vertebrates. While the literature on mammal herds, bird flocks, and fish schools is mostly descriptive or coarsely analytic (e.g., [PART80], [VANO70]) a synthetic approach to the study of such group behaviors was pursued by Craig Reynolds [REYN87] in order to develop a realistic-looking animation sequence of a flock of birds.

More interesting than vertebrate flocking, however, are the behaviors of the social insects: ants, bees, and termites; the observed aggregate behaviors exhibit a greater complexity, while the individual animals are much simpler. Through experimental manipulation of insect colonies and computer simulations, researchers have elucidated some of the mechanisms by which these colonies survive and grow by adapting to their changing environment. For example, Deneubourg [DEN90] has demonstrated via simulation that sorting behaviors observed in ants can be produced by the simplest possible biasing of random behavior by environmental cues, while Franks [FRAN89] has used simulation to explore the changing raiding patterns of army ants. Seeley [SEEL89] has investigated how worker honey bees appropriately initiate various productive activities in response to quite simple signals and cues. Honey bee colonies thus provide a model for achieving "purposeful" coordinated group action, responsive to changing environmental conditions, without employing a world model -- in fact, without explicit global decision making of any sort.

With various individual and group animal behaviors serving as "existence proofs", quasi-intelligent "emergent behavior" resulting from the interaction of simple reactive planners has been proposed as the basis for the intelligent control of individual robots, in the development of usefully complex systems [BROO86] as well as simple conceptual vehicles [BRAI84]. The term "Swarm Intelligence" has been used to describe the application of this approach to distributed systems consisting of perhaps hundreds of elements [BENI91]. Biological models are explicitly acknowledged as the motivation for much of this work [ARK92a], [KUBE92].

The limitation of nature's "existence proofs", however, is that the "purpose" of a natural system is to survive and reproduce, so that specific behaviors that appear purposeful to the observer merely represent larger or smaller hills in the topographical fitness map continuously processed by the forces of natural selection. The realization of emergent behaviors to allow an artificial system to achieve an a priori specified purpose may not always prove to be a straightforward matter.

Technological Opportunity: Inevitable, Yes, but When?

Eight years have now passed since Anita Flynn described a vision of swarms of "gnat robots" performing useful tasks for humankind [FLYN87], and they're not here yet. While MEMS (Micro Electro Mechanical Systems) as a manufacturing technology has moved steadily forward, the principal commercially successful applications for MEMS devices have been components such as pressure sensors and accelerometers for the automobile market. In fact, robots of any size have yet to appear in our daily lives -- arguably the closest thing to a recognizably-robotic sensor-actuator combination we see is the proximity-sensing automatic door opener, together with its descendants, the toilet flushers and faucet controls now seen in many airport restrooms.

Nevertheless, robotics researchers persist, and there is little doubt that the continuing exponential improvement of microelectronic processing price/performance, coupled with continuing developments in MEMS and other sensor and actuator technologies, will eventually yield successfully mass-marketed autonomous mobile robots -- for example, an early possibility might be toy "pets" capable of displaying interesting behaviors in their interactions with their owners and with each other. Even currently available toy cars can be (and often are) easily coupled with fairly simple electronic sensor/processor/control appliques to provide affordable robotic research platforms.

The mass production of robots will certainly trigger dramatic unit-price reductions, and it is these reduced costs that will finally permit the implementation of "swarms" of "mini" (and, ultimately, "micro") robots to handle real-world applications in both the military and civilian worlds. But the realization of practical systems comprising large numbers of mobile robotic elements which are capable of performing useful tasks requires much more than just the cost-effective manufacture of the robots themselves -- for example, the prospective user must understand what the system is capable of doing in order to know when to deploy it, must know how to tell the system to do the specific task required, and must be able to assess how well the system is doing or has done its job. And the system development process must ensure that the system will actually achieve its advertised level of performance across the full range of specified manufacturing tolerances and intended operational environments.

Up to Many Robot Systems

Robotics home page

Please address all questions and comments to: robo-web@spawar.navy.mil
Last update: 1 December 1998.