Drop That Swatter--It's a Robot
People have long awaited the perfect docile machine that washes windows or shovels snow. Instead, technology has given us awkward brute robots good for building cars in a factory but terrible for moving and doing everyday chores.
Scientists long believed the solution lay in devising a robot brain that processed information sequentially. A robot would first analyze information its sensors picked up from its environment to identify where it is and what surrounds it. Then it would figure out what to do about that environment--in effect, make an action plan. Finally it would produce a set of directions to its motor controls to carry out the planned action.
In short, the robot would do a great deal of pseudo-thinking, making models of external realities and internal possibilities and comparing them against one another, before it could accomplish anything. That's slow and takes a lot of processing capability. Even with miniaturized electronics, this means a pretty big robot. But now a possible revolution in robotics is emerging from something known as the Insect Lab at the Massachusetts Institute of Technology. Despite the name, it has no real insects--only computer bugs.
Five years ago, an artificial-intelligence expert looking for something new to do decided the accepted view for a mobile robot's brain was too intelligent. Rodney Brooks realized that researchers had been thinking of mechanical people. He turned instead to insects.
Insects operate on reflex behavior--coherent, straightforward, but simple responses to what they sense. Turn on the kitchen light, and cockroaches scuttle for darkness. Whenever certain sensors are stimulated, specific motor controls direct immediate action.
Brooks and his students followed that idea in building a robot they called Allen. It looked like a tall footstool on wheels. Allen had three behaviors programmed in layers--parallel processing rather than sequential.
The most basic behavior was "avoid obstacles." The robot's sonar sensor thus let it rest in the middle of a room until someone--a mobile obstacle--came near. Then it would roll away. Its other two directives were "follow walls" and "go through gaps," which means that a pursued Allen seemed lifelike as it fled away along a wall and dashed through the first open door it came to.
Allen's successor, Herbert, had a gripper arm and a more complex set of layered behaviors. Herbert could roll quietly through the lab and steal cans of soda pop off desks.
So far, so good. The rolling robots demonstrated that the insect-model electronic brain, with tiny semiautonomous circuits triggered by sensed information, worked very well. Yet Allen, Herbert, and their experimental kin still needed perfectly smooth, flat surfaces on which to move. Real progress demanded a robot capable of dealing with uneven terrain.
Legs, as most land-based life forms have established, are a good way to do that. But legged robots have been enormously difficult to devise. They're given to falling over, usually breaking something in the process. Again Brooks turned to the insect model, for locomotion as well as brain. Insects don't have far to fall when they stumble, and tripping up seldom hurts them. So he designed and programmed a robot that really looked like a low-slung but oversized bug.
Built by Brooks' student Colin Angle, six-legged Genghis evokes a foot-long cockroach. Most of its 57 behavior circuits are dedicated to local activities in its individual legs, with only five required for overall coordination--two for walking, one for steering, and two for the infrared optical sensors that let the robot track moving objects.
Because so much brain power is given over to decentralized activity, Genghis can clamber over obstacles that have immobilized far more sophisticated robots. If one leg should catch as the robot tries to lift it over the edge of a step, for example, a "leg lifting" behavior causes Genghis to back down and try again, this time lifting that leg higher.
The new robots may sound creepy, but--going by their portraits in the 25 May 1990 issue of Science magazine, which reports on the work--they're cute. In fact, their charm has been a nuisance to the researchers. Grant examiners, for instance, reportedly find it hard to take the cute little bugs seriously.
And the researchers do want to be taken seriously. They're hard at work on a successor to Genghis, dubbed Attila. This model has knee joints, the ability to turn itself over if it falls on its back, and several other even more insect-like skills. But unlike any natural bug, Attila packs a camera and can carry other instruments on its back. The Insect Lab team thinks Attila is the perfect explorer to send on a Mars mission.
I wish them well. Think of how much snow a six-armed robot could shovel!
