1 FEBRUARY 2007
Insects are not very good at avoiding collisions with cars.
So it would seem like not such a smart idea to ask insects how to design automatic-collisions avoidance systems for automobiles. On the other hand, human drivers aren't so great at avoiding crashes, either, and perhaps need all the help they can get -- even from locusts.
As it turns out, locusts have some pretty skillful moves for avoiding collisions, such as with birds patrolling for yummy locust snacks. When a locust spots a bird approaching on a collision course, certain nerve cells start firing electrical signals like crazy, initiating evasive maneuvering.
Scientists have studied robotic visual systems that mimic the workings of these locust nerve cells, known as lobula giant movement detectors. Ideally, a car equipped with such a robotic system could automatically avoid collisions with other objects, or at least send a warning signal of an imminent crash. And such systems can detect approaching objects, research has shown. But in complicated situations (such as real traffic), these systems signal too many false alarms. Cars or pedestrians merely moving across the path are often mistaken for objects threatening to collide.
Attempts to improve warning performance are under way, though, as described in the current issue of the journal Biosystems. Researchers at the University of Newcastle upon Tyne have devised robotic models that combine the locusts' ability with skills found in a nerve cell system used by flies.
When flies land, these nerve cells respond to the growing image size of the landing surface, so the fly knows when to stop flapping its wings. Translated into a robotic visual system, that ability allows the robot to distinguish crosswise motion from a head-on approach. An approaching object gets bigger in all directions at once; a crossing object grows in one quadrant of the image and then another, not both at the same time.
Simple tests with videos of traffic conditions show that the system combining locust-type with fly-type detectors eliminated most of the false alarms, although sometimes at the cost of delaying true collision warnings. "The insect inspired neural network described in this study does provide an effective mechanism for detecting collisions," wrote Richard Stafford and collaborators Roger Santer and F. Claire Rind.
No doubt it will be a long time before robots using insect-based brains will be allowed to drive cars. But using computing strategies from real life to improve robotic artificial intelligence may be just what robots need to help them achieve the promise for them that has long been offered by science fiction. And such help can come not only from insects. Some researchers believe robotic intelligence could benefit from a symbiotic relationship with slime mold.
A slime mold typically takes the form of a giant amoeba, which responds to certain stimuli by oozing its protoplasm around. Clever researchers from Kobe University in Japan and the University of Southampton in England have grow slime mold in a six-armed star pattern (each arm meeting in a hub at the center). The trick is to provide a wet surface in the desired shape surrounded by dry plastic -- the slime mold grows to cover the moist areas and avoids the dry ones.
White light repels slime mold, causing its protoplasm to ooze back in forth, producing oscillations in the thickness of circular bulbs at the end of each arm. And so how, you might wonder, does that help out a robot? 
Well, in a paper also in Biosystems, the Kobe-Southampton researchers show that the slime mold's oscillations in response to light can guide the motion of a simple robot.
We're not talking about C-3PO here, but rather a small metallic bug-like robot with six legs and one light sensor per leg. Light sensed from the six legs is transmitted to a computer, which in turn activates a projector that shines light on the slime mold. The pattern of light will cause changes on the oscillation rate; the slime mold's response is transmitted back to the computer, which then instructs the robot about how to move its legs.
What's going on here? The robot is dumb. It senses light from six directions, but doesn't know what to do about it. The slime mold is smart. Shine a pattern of light on it, and it computes precisely how to ooze. So in effect, the slime mold can do the computation for how to move based on the pattern of light detected by the robot, and instruct the robot accordingly.
OK, so this is a rather crude beginning for bio-powered robotic brains. There's a lot more work to do before systems such as these become practical. But the basic idea makes sense. Nature has packed powerful computing ability into biological systems, for much cheaper than what it would take for human engineers to design similar complexity into silicon. (You do, however, have to feed the slime mold some oatmeal.)
In any case, biology offers engineers some spectacular information processing strategies that have stood the test of trial and error over eons of time. Scientists shouldn't be too proud to ask for help, even from amoeba or locusts. They've been around a lot longer than people have.
E-mail: tsiegfried@nasw.org
