MIT Reporter

James McLurkin's work space is cluttered with items typical of a research scientist working at MIT's Artificial Intelligence Laboratory: transistors, circuit boards, and the like. But there's also the large Tupperware container that sits near his computer and houses his inspiration--ants. For the past three years, beginning while an undergraduate, McLurkin has been building a community of robots modeled after a colony of the social insects.

The notion has practical potential: cooperating robots could handle tasks that are best for people to avoid, such as removing bomb fragments from war zones find hazardous waste from landfills. But McLurkin has also been driven by a more basic desire--simply to develop a community of very small robots.

Each of McLurkin's dozen cubic-inch, 1.2-ounce robotic ants is packed with gizmos. These include 17 sensors that allow each robot to recognize aspects of its environment, such as the location of other "ants" through "bump" sensors that resemble and act like whiskers. Moreover, two devices emit infrared (IR) signals--which can be sensed by other robots--while three "actuator" devices enable the machines to manipulate their surroundings. (For instance, a pair of "mandibles" can grasp items.) Also squeezed in with the requisite computer chip and a battery are two motors that enable each robot to move six inches a second on two treads plus a third motor that powers its mandibles.

All these mechanisms enable the robots to interact in a way that begins to mimic a true ant community. Consider, for instance, how McLurkin has programmed his machines to forage for "food," which consists of quarter-inch brass balls. When a machine's bump sensors, which carry a voltage, encounter brass, the resulting conductivity causes the robot to emit an IR signal. During a"foraging" period, the reception of that signal by the other machines is an indication that "food" has been found. Those robots then follow the signal to join the first machine.

To make the foraging activity better approximate what happens in nature, McLurkin intends to simulate the scent trail that real ants leave between food and their colony's nest. Robots that find brass balls might leave an ink trail that their cohort could detect with an appropriate optical sensor.

An important aspect of foraging and the other activities McLurkin is devising for his robots is the community's ability to succeed at a task even if one.individual fails--as in the style of natural ant colonies. Because success stems partly from the sheer number of working robots, the more machines he develops, the more often his colony will complete its tasks. The researcher is going for a total of 21 machines, which would make the robot community one of the largest in the world.

Another key to the robots' ability to work together smoothly lies in devising programs that are as simple as possible, according to McLurkin. Traditionally, programmers have written large, complicated robotic codes that process every sensory input before giving instructions on how to respond. Such complex programs are difficult to write and operate. By contrast, McLurkin designs numerous small programs that run concurrently and that each focus on only one or two possible inputs. Depending on the actual inputs, only the program deemed most appropriate by the programmer directs the robot's response. The programming technique, called subsumption architecture and developed by Rodney Brooks, professor of computer science at MIT and associate director of the Artificial Intelligence Lab, creates "complicated-looking behavior from simple instructions," McLurkin says.

For example, when his "ants" engage in a programmed game of tag, a robot that is not "it" typically starts by moving forward slowly, based on a program directing that action. A second program enables the robot to move away from objects into which it bumps. A third program enables the machine to note when it is tagged--through the transmission of an IR signal--by the robot that is "it." When tagging occurs, other programs direct the robot that is now "it" to move forward slowly and emit an IR signal upon bumping into an object.

Such interactive abilities have attracted the attention of the U.S. Central Intelligence Agency, which has expressed initial interest in miniature robots that could conduct a surveillance operation by carrying a group of cameras and microphones. And the Department of Defense (DOD) is funding McLurkin's research on foraging because DOD experts see possibilities for directing robots with mandibles to retrieve fragments of cluster bombs, a task now performed by people.

But before robot communities can move into the commercial mainstream, says Maja Mataric, an associate professor of computer science at Brandeis University whose doctoral work at MIT on robotic communities inspired McLurkin, researchers like him must develop better sensors and actuators. For example, the ability to develop a relatively small sensor that can sniff out many chemicals is important, according to McLurkin, who points out that "if we packed all the different chemical sensors into a robot now, it would be the size of a refrigerator."

While resolving such problems could lead to practical applications of significant social value, McLurkin also envisions some possible "lighter-weight" uses for the machines. A group of micro-robots could reside under a refrigerator during the day and pick up crumbs on the floor at night, he proposes, or could be marketed as a game-playing toy that "would be cool for kids."

PHOTOS (COLOR): James McLurkin's robots, designed to interact like ants in a colony, are crammed with numerous sensors and three monitors as well as a computer chip and a battery.

~~~~~~~~

By STEPHANIE V. GREPO