Robots Take On the Home
Roomba - iRobot's $199.95 Intelligent Floorvac
iMow - Toro's $499.95 Robotic Lawnmower
September 25, 2002
iRobot Corporation, located "just outside Boston" has introduced a robotic vacuum cleaner that can vacuum your floors while you work on your computer (PRODUCT REVIEW- Vacuum's artificial intelligence a bona fide treat - Nando Times). What's unique about it is its price and its availability. At $199.95 it's a small fraction of the $2,995 price of Probotics Cye. Roomba is available from Hammacher-Schlemmer, The Sharper Image, Brookstone, and from iRobot itself. Looking around us, our nearest Brookstone store is at the Riverchase Galleria Mall in Birmingham, about two hours from here. In the meantime, the Hoover Company has just announced a partnership with Friendly Robotics to deliver a robotic vacuum cleaner by the end of this year. A price has not yet been set. Eureka's Robo Vac is apparently still unavailable, nor is there any indication when it might be. Electrolux is presumably still selling its Trilobyte robotic vacuum sweeper only in Sweden.
iRobot's Robovac Eureka's Robo Vac Electrolux' Trilobyte (Sweden)
One of Electrolux' subsidiaries, Husqvarna, has an "Auto Mower" for $1,499.95 (to be compared with Friendly Robotics' $499.95 "Robomower", or Toro's $499.95 iMow). Friendly Robotics Robomower uses a compass and an odometer to mow by dead reckoning, presumably backed up by the buried perimeter wire that, let's hope, keeps it from wandering into your neighbor's flower beds. (Robomower debuted in early 2000 in the Hammacher-Schlemmer catalog.)
Of course, these are robots in the sense that an automatic washing machine or an automatic car washer is a robot. Right now, they're probably still in the "early adopter" phase, but as more people see them, and as their capabilities improve, they'll gradually increase their market penetration.
Robomower is available from Central Lawn and Garden here in Huntsville. I'll have to visit Central Lawn and Garden and see what they say about Robomower.
Later: Well, I visited Central Lawn and Garden, and there were three Friendly Robotics Robomowers sitting along a wall. They're bigger than I pictured. They now come in two sizes: the RL 520 and the RL 820. The RL 850 cuts a 21" swath. The guys who were selling them didn't seem to be too thrilled about them. Central Lawn and Garden has sold a few of them, but Robomowers are obviously not a hot-ticket item. On the other hand, who knows about them? There certainly hasn't been any local publicity. The fellows there said that they were good for yards up to 6,000 sq. feet. It turns out that they'll mow 3,000 sq. feet (60' X 50') on a charge. Then the batteries have to be recharged, or you can use a set of spare batteries. (A spare set of batteries will set you back $130, or a fast external charger and extra battery bundle can be had for $190.)
It could be the cat's meow for someone who is physically handicapped... e. g., an 80-year-old lady.
The two guys at Central Lawn and Garden said that this might be the wave of the future once they're "perfected".
Robomower would seem to be a natural for a fuel cell. Then it could really tear up the pea patch... uh.... perhaps that was an unfortunate choice of words.
Twenty years from now, we'll take automowers and autovacs for granted, and think that these first models were terribly crude.
Husqvarna's Auto Mower Friendly Robotics' Robomower Toro's iMow
Dr. Hans Moravec's' Visually-Guided Robotics Platform
One drum I was beating back in early 2000 when I wrote, "Here Come the Robots!" was that of Dr. Hans Moravec's (and colleague's) visually-guided robotics platform, due to appear in the 2002-2003 time frame. Well, it's now late 2002. How is that project proceeding?
I get the impression that it's proceeding a little slower than was forecast but only because there's more to it than.... meets the eye? It sounds as though Dr. Moravec and his associates are making outstanding progress, and that their product, when it appears, will be very satisfying. He says in his most recent report,
"We, of course, continue to follow unexpected avenues the research presents, including many not mentioned."
That report, "Report 6 (Higher fidelity)", was written in February, 2002, so another report will probably be published sometime soon. Of course, writing reports takes time away from productive research.
As everyone knows, the impossible takes a little longer. Dr. Moravec and his affiliates are working "loaves and fishes" miracles, and taking the world where it's never been before. There will be no hard-and-fast point at which you can say, "Well, they're done now." So if it's 2004 or 2005 before a first edition of Dr. Moravec's promised visually navigated utility platform is ready for commercial delivery, we can be thrilled to have such a trail-blazing product whenever it appears. The important thing is that it's making progress as fast and well as it can, given the resources and constraints that weigh upon Dr. Moravec and his co-workers. In reality, new inventions develop slowly and steadily, Only in retrospect do we assign specific dates to what was a gradual process.
One Terops Computers by 2010? 2005? Now?
Happily, computer progress seems to be on track through at least 2010. Intel is making remarks about delivering 1 terops computers by 2010. In the meantime, IBM, Sony, and Toshiba have promised 1 terops computers by 2005. And nVidia is supposedly already delivering 1 terops graphics processors in its latest and greatest graphics cards. (These graphics engines are special purpose computers, but then, robotics and AI may lend themselves to special purpose computers.) .
RAM Memory for $250 a Gigabyte
If we needed large complements of RAM memory for artificial intelligence, right now, it would cost about $250 a gigabyte or $250,000 a terabyte. At that rate, it will cost about $2,500 a terabyte by 2012, or $250 a terabyte by 2017.
One Terabyte and Up Disk Drives
Disk drives were one domain in which we were supposedly hitting the technological wall, but that is no longer the case. Disk drives in the multi-terabyte range are already in the offing.
Human-Level Artificial Intelligence
I believe that computer hardware will reach human-level storage capacities and processing speeds before the present Moore's Law progression fades out. (Dr. Hans Moravec has estimated human-level intelligence to require processing speeds of 100 terops. I'm hoping that human-level intelligence can be achieved with processing speeds of no more than 10 terops. My reasoning is that human-level AI may not need to meet all the requirements for surviving in the wild. Also, I think that 1 petabyte... 1,000,000 gigabytes... of storage may suffice. My reasoning is that 1,000,000 gigabytes of storage could store a lifetime's worth of video footage. In reality, although we record many other details besides just the visual, we store memories in very abbreviated forms, and probably reconstruct them at need. Also, our ability to remember words and facts are a pale shadow of what a computer-based memory could provide. One hundred megabytes (1/10th of a gigabyte) could probably be encoded to hold a dictionary-thesaurus of the English language (or any other language).
By the end of this decade, I could imagine ten 5.25" disk drives storing 10 terabytes apiece, for a total of 100 terabytes of disk storage, a terabyte of RAM, and multiple, special-purpose processors producing a total of 10-or-more terops in a desktop enclosure. If push comes to shove, 8 terabytes of RAM, backed up by multiple disk drives, ought to provide the storage capacity for human-level intelligence. Of course, these are just hardware capacities. The software and structure of a human level artificial intelligence is quite another matter.
The Semiconductor Industry Association's fifteen year technology roadmap calls for 256-gigabit memory chips by 2012. Extrapolating today's price of $250 a gigabyte for RAM, based upon 256-megabit chips, we're led to projected price, using 256-gigabit chips, of $250 a terabyte by 2017. That might drop to less than $100 a terabyte by 2020.
A wild card in this is the introduction of molecular-level memory, whenever that occurs. There are something like 1015 simple molecules per square centimeter, so this would set an approxiamte upper limit upon "conventional" memory bit densities of the order of a petabit per square centimeter. However, supporting circuitry is required, and speed of access is a very important parameter. On the other hand, volumetric memory storage is certainly possible, especially for non-volatile memories, like flash memory. There are probably ways to achieve huge improvements that will surface over the coming decades.