Homer Hoover – robotic vacuum cleaner
Gregory T. Pope
Frank Jenkins wants to put a robot in your home. And what better way to win a place in your household than by doing your dirty work?
ROBOTS HAVE ALWAYS been seen as playthings with no real purpose,” says Frank Jenkins, a computer programmer in Menlo Park, California. “But when I show people Homer, which is designed to vacuum the house, suddenly their attitude changes.”
While the robotic housekeeper is a scifi staple, free-roaming robots have faltered outside the controlled confines of the laboratory. So a house, where things don’t stay put, might be the most unpredictable environment a robot could encounter.
Homer has at least two years to go before it can survive the average American home. Built in Jenkin’s garage around the guts of a Black & Decker upright vacuum, the robot is currently all broom and no brain. Eventually, Homer must be smart enough to vacuum all through the house without getting lost, and it must scoot back to a U-shaped unit to recharge its batteries. But the robot’s eyesight now consists of just four infrared sensors. These devices broadcast cones of light, at wavelengths beyond human perception, and sense the light’s reflection from objects. Because their range is only about four inches, Homer gropes around myopically by following walls.
But Jenkins aims to beef up his little robot’s senses. “In my opinion,” he says, “there’s not a robot built that has a tenth the sensory input it needs to operate in the real world.”
Homer will navigate with 12 ultrasonic sonar units and 24 light-level sensors. The sonars bounce inaudible sound pulses off the surroundings and gauge distances to objects by clocking the echoes of the pulses. The light-level sensors act in unison as a crude panoramic camera. In addition, built-in odometers will tell Homer how far it’s rolled, and four extra sonars, along with the infrared sensors, will spot nearby obstacles.
Sensory overload? It would be, without a mind to make sense of it all. Jenkins plans to provide Homer with sets of artificial neural networks–brainlike programs that winnow useful information from sensory input, even if the input includes some unforeseen data.
Traditional computer programs are structured to anticipate the sort of data they will receive and process into a solution. By contrast, a neural network requires training to come up with the right answer. The “trainer” exposes the neural network to various inputs, such as different views through the sensors, and tells the program what its output should be, such as “This is a wall.” After enough education, the program can extrapolate when faced with unfamiliar data–like a chair placed at an unusual angle, or a moving pet.
The neural network will identify sensory input as a doorway, a corner, a wall, or an obstacle such as a chair or a cat. A mapping function then puts these interpretations to use: walls, corners, and doorways make good landmarks because they never move, whereas chairs and cats are best left off the map.
The neural network alone won’t make Homer foolproof. More mundane help comes in the form of bumper rings and a microphone to detect the sounds of objects bouncing around inside the beater apparatus, signaling that the vaccum has sucked up a marble or a sock; Homer can spit out the offending object and vaccum around it.
Homer will need this edge to outperform the competition from Japan. Both the Matsushita Electric Industrial Company and the Sanyo Corporation are developing their own robotic vacuum cleaners. Although neither boasts Homer’s sensory arsenal or its neural network, both are smart machines.
They rely on “fuzzy logic,” computer instructions more forgiving than conventional programs. But they use their fuzzy brains merely to spot obstacles and decide how dirty the floor is. Their owners would teach them the layout of the house by operating them manually the first time. The robots would later navigate alone simply by retracing these initial paths. Although the Matsushita machine, like Homer, must find its charging plug to stay alive, Sanyo’s robot trails a power cord.
Jenkins expects a Homer-derived product to sell for three to four thousand dollars. At the moment, however, Homer’s development is restricted to evening and weekend spurts. “We’re talking about a part-time project here,” says Jenkins, “but more than just a hobby. It’s a way of trying out my software ideas, but ultimately I’d like to have the robot go into production.”
COPYRIGHT 1993 Discover
COPYRIGHT 2004 Gale Group