The Robots Are Getting
Here Come the Robots -
- Kevin Kearney,
and Bob Seitz
June 14, 2003
Have the Robots Arrived?
--Dr. Moravec promised the first cost-effective visual navigation subsystem by the end of 2002.
Three years ago, in January, 2000, one of us published a paper entitled, "Here Come the Robots!", predicated upon Dr. Hans Moravec's developmental work on a visually guided platform. Dr. Moravec (http://www.frc.ri.cmu.edu/~hpm/), promised the delivery of a prototype "power head" by the end of 2002 that could be retrofitted to existing industrial robotics devices to permit them to visually navigate in real-world environments.
--Toward the end of 2002, we began to think he wouldn't make it.
Toward the end of last year, knowing how long it takes to accomplish anything, and knowing that Dr. Moravec was supported by a couple of graduate students, one of whom (Martin C. Martin) has been awarded his Ph. D. and has moved on, we expected that Dr. Moravec might require another year to make good on his promise.
It's now the middle of 2003. How are Dr. Moravec and his co-workers coming?
The answer is that they have delivered their product as promised! Dr. Moravec announced at the end of December, 2002, that he has reached his goal, and that he and a few friends have organized a small company called Botfactory to bring his powerhead to market.
Dr. Moravec currently estimates the current cost of the components in his subsystem at a bit more than $5,000, with an estimated $1,000 price tag by 2007.
--Dr. Moravec's Long-Term Timetable for Robotic Development
What's particularly thought-provoking about Dr. Moravec's timely meeting of his milestone is the rest of his forecast... most notably, his prediction that robots will rival humans by 2030, and will surpass them (and perhaps, supplant them) by 2050.
In 1991, Dr. Moravec published an article on robotics in the "fact" section of Analog Science Fiction and Fact in which he explained why we didn't already have intelligent robots, and how soon we might expect them. In the article, he forecast the year 2000 as the time frame in which the one-billion-floating-point-operations-per-second speeds that would permit computers to emulate human vision. He estimated at that time, based upon a performance comparison of nervous tissue in the retina with the computing resources required to perform the same low-level visual functions, that a 10-teraflops computer would be required to match the computing power of the human brain. He predicted that by the year 2000, computer speeds and storage capacities would reach the lower end of the envelope (one gigaflops) for visual guidance of automatic devices.
He also set forth the following timetable for robotic development.
-Automatic Lawn Mowers (Automowers)
In 1999, an Israeli company, "Friendly Robotics" (www.friendlyrobotics.com), introduced RoboMower for $700. Toro licensed the design from Friendly Robotics and brought out its own $500 automatic lawn mower in 2000, called "iMow", forcing Friendly Robotics to lower its price to $500 to compete.
RoboMower incorporates a magnetic compass, and after mowing along a perimeter-wire outline of the yard, begins to cut diagonally across it.
--Pros and Cons
Recently (May, 2003), one of us visited our local John Deere dealer (which carries RoboMower in the city of Huntsville) where he discussed Robomower's pros and cons. The man at the counter said that RoboMower does a surprisingly good job of mowing yards, but that its principal problem is its poor durability. He said that it's constructed with parts that wear after two or three years. I mentioned that Friendly Robotics has just come out with a new, second-generation model that costs $700 (while still offering their original $500 model). I speculated that this new design might have beefed-up critical parts to address the durability problem, and he agreed that this might well be the case.
What’s particularly interesting about Friendly Robotics is that they seem to be the first company to successfully produce and sell a consumer-level automower. Will they become the “Apple Computer Company of the Robotics industry?
(Toro, by contrast, appears to be keeping a low profile. There are no iMow automowers for sale in Huntsville, although iMow is still listed prominently on Toro’s website. We suspect that Toro is playing it cagey, using iMow as a stalking horse until automowers have established themselves as a viable niche in the marketplace).
In the meantime, Friendly Robotics is working on “a fully automated indoor surface cleaner intended for domestic use. Future plans include additions and enhancements to the Robomower and surface cleaner product lines, and, later on, additional domestic chore busters such as a robotic snow thrower, an in-house security and surveillance robot, and perhaps one day a ‘personal helper’ that can bring your meals to the table.”
Friendly Robotics’ future “personal helper” will presumably possess arms and manipulators.
Friendly Robotics is also working with Hoover to develop and market an automatic carpet sweeper.
The website cited below, “Review: A Robot for the Lawn Lazy”,
(http://abcnews.go.com/sections/scitech/TechTV/techtv_robomowerReview0304010.html), presents a (favorable) review for RoboMower.
Stay tuned for further announcements.
-Automatic Vacuum Cleaners
In 2000, there were big industrial automatic carpet sweepers, but only experimental-prototype, consumer-level automatic vacuum cleaners… most notably, Eureka’s Trilobyte., shown below.
Eureka has just released the Trilobyte for sale in the UK for the princely price of £999. Trilobyte uses ultrasonic sensors to map out a room, spending up to 15 minutes in this mapping mode before sweeping the carpet.
It will automatically return to a recharging station when its batteries run low.
In the meantime, last fall, an MIT-spinoff company, iRobot (Rodney Brooks, et al; http://www.irobot.com/home/default.asp), began selling Roomba, a $199.95 autosweeper that is playing to excellent reviews: See two of them below).
of Reviews: 6
thing really works!
rschneider (firstname.lastname@example.org) from St. Louis , MO
January 30, 2003
bought a Roomba as a novelty/conversation item, but as it turns out this
is a bona fide functional vacuum cleaner. I was suprised that a 1st
generation robot vacuum would work this well, but sure enough they seem to
have most of the bugs worked out on the first try.
-- Dust bunnies beware....
Matt (email@example.com) from Alexandria, VA
January 14, 2003
simply works. Cut our heavy vacuuming down to once a month and put my
daughter, much to her delight, out of the weekly vacuuming business. Im
very impressed with both the quality of the product and the engineering
that went into the design and logic. Ive had mine since Early November and
it now has probably 50 to 60 hours on the clock. It successfully survived
both the Thanksgiving and Christmas cleaning frenzy. It works best on our
short pile carpet and hardwood floors. Keep it way from fringed carpets
and fragile items. While its gentle, it cant tell a Ming Vase from a wall.
Keeping it away from electrical cords is also recommend it will attempt to
eat the. Stairs and drop offs are no problem It just turns away and
presses on. The brushes do an excellent job on picking up larger items (to
include pine needles) while the mini-vac easily sucks up dust and smaller
Like RoboMower, Roomba is also for sale here in Huntsville
at Linens’n Things.
word-of mouth reviews are so good that one of us is planning to buy one on
Monday (before you read this). (We’re normally late-adopters who don’t buy
new devices until they’re on year-end sale at Walmart for $59.95, but
there’s a 60-day trial period for this, so we’re going to give Roomba a
--Automatic lawn mowers and automatic carpet sweepers are appearing on schedule.
In any case, automatic lawn mowers and automatic carpet sweepers are appearing in accordance with Dr. Moravec's timetable.
Looking to the Future:
Once something becomes market-driven, it’s hard to forecast specifically, since developments depend upon market feedback as well as technological progress.
-The Rest of This Decade (2002 to 2010)
--For the rest of this decade, most robots will probably be unrecognizable as robots..
They will simply be automatic machines of steadily increasing sophistication, like robotic car washes and automatic washing machines. Most robots will probably be found in commercial settings where they may be cheaper and less error-prone than humans. Tasks that are dull, dirty, and/or dangerous will be prime candidates for replacement by automatic or semi-automatic machinery.
What we’ll see now will be additional companies getting into the act, with new, improved models appearing every two or three years. (Ten years from now, people are still going to own manual floor sweepers, and some people are going to say, "You can't beat a manual vacuum sweeper for picking up dirt, and for getting into corners. I may get an auto-sweeper when this carpet sweeper wears out, but my old one isn't that old, and I don't use it that often. I'll just stick with what I've got until they get all the bugs ironed out of these automatics.")
In the meantime, automatic trimmers and edgers, sentry “robots”, snow blowers, leaf mulchers, floor scrubbers, and sentry robots will probably follow automowers and autosweepers into the marketplace.
For the next few years, these devices will probably use “touchy, feely” navigation to grope their way around, and to map their two-dimensional work areas. And in fact, visual navigation may not be needed for two-dimensional navigation. However, it may be more necessary for 3-D sensing of the environment once robots acquire robotic arms and manipulators.
--Batteries are a bottleneck
When low-cost, light-weight fuel cells become a practical reality, which might also happen toward the end of this decade, these machines may play a more significant role in consumer affairs. It’s hard for battery-operated mowers and vacuum sweepers to compete with high-horsepower, manually controlled mowers and vacuum cleaners.
--Dr. Moravec’s visual navigation “Powerhead”: several years of testing, evaluation, and product engineering comes now
Dr. Moravec has 'linked up with a small, six-member company (incorporated as 'Botfactory') pursuing the commercialization of his CMU mobile navigation system, targeting "a prototype navigation unit in two years, and a first salable product a year after that".... in other words, a marketable visual navigation subsystem by the end of 2005.
Visual navigation systems
--Working down the price curve.
In order to embed Dr. Moravec's visual navigation subsystems in consumer equipment, I would suspect that such subsystems mustn't add more than $100 to the prices of these devices. Given the current rate of semiconductor progress in which the number of transistors on a chip doubles every two years, I would project such a price premium to fall to $100 in another 12 years, meaning that it would be 2015 before visual navigation subsystems appear in consumer devices.
We suspect that this will happen well before then, as special purpose hardware and software makes its debut.
--The Cell: a one-teraflops chip by 2005.
This discussion wouldn’t be complete without mentioning the “Cell”. About two years ago, IBM, Sony and Toshiba announced plans to deliver, by 2005, a microprocessor chip to be used in Sony’s Playstation 3, as well as in mainstream computer applications, that would deliver 1,000,000,000,000 floating point operations per second. (That would make the Cell about 500 times as fast as the fastest Pentium 4 in March, 2001.) This development program appears to be coming along on schedule.
. The Cell is so-named because it is designed to be networked, with other Cell chips to create very low-cost supercomputers. Such chips might advance the development of artificial intelligence by up to ten years. Also, there may be further improvements in the Cell as time goes on, and microprocessors continue to improve across the board.
Intel has said that it will introduce a one-teraflops microprocessor by approximately 2010.
All along, computer researchers have been developing enabling technologies for speech recognition, speech synthesis, optical character recognition, facial recognition, limited specialized-domain voice-response systems, bipedal locomotion, and other humanoid capabilities that will support anthropomorphic robots.
Alternatively, “virtual robots” that appear in 3-D on computer screens may combine such functions. These will only get better over time.
One other enabling technology needs to be mentioned: computing technology.
As he discusses in his articles, Dr. Moravec estimates that human-level intelligence will require computing speeds of the order of 10-to-100 trillion floating point operations per second. To that, we would add an estimate of, perhaps, 100-to-1,000 trillion bytes of computer disk storage, and 1-to-10 trillion bytes of RAM storage.
When the Cell arrives in 2005, we can expect to achieve 1 trillion floating-point operations per second per chip. Disk capacities of ½ trillion bytes should be available at a cost of $250, so that 100 trillion bytes of disk storage would cost about $50,000. RAM should cost about $100 a gigabyte, or about $100,000 a terabyte.
Intel is projecting at least 15 more years of progress at Moore’s-Law rates, and disk drives can probably also look forward to at least 15 years of comparable improvement.
That means that by 2010, 10-to-100 terabytes of disk storage should cost about $5,000, with RAM running around $10,000.
By 2015, prices should have dropped to about $500 for disk storage, and $1,000 for RAM.
By 2020, disk storage should cost about $50, with RAM running around $100.
Even if these goals aren’t precisely reached, there ought to be manufacturing economies that would carry prices down into that range.
Even if we couldn’t reach our goal that way, it should be possible to link robots to larger central computers over wireless links, and to draw upon heavy computing power for only the heavier tasks, while performing lesser tasks locally.
Eventually, semiconductor prices may decline because of higher levels of automation.
We believe the computer industry will make it on the hardware front.
2010 to 2020:
--Visually guided automatic machines/robots enter the home.
The period between 2010 and 2020 might be a time when visually guided equipment will be entering the home. This is when real excitement will begin, as devices are introduce that begin to look and function more like 3CPO or the Jetsons’ “Rosie, the Robot”.
Sometime during this period between 2010 and 2020, Dr. Moravec anticipates the advent of "universal robots"... robots with the intelligence of a small mammal, and the ability to cook, load and unload the dishwasher, clean, bring in the mail, and perform other routine household chores. The figure below shows an artist's conception of such a robot cooking eggs for breakfast (taken from Dr. Moravec's book, ROBOT, Moravec, Oxford, 1998, Chapter 4: Universal Robots, page 124.
The addition of arms and manipulators to automatic devices would seem to me to entail an order-of-magnitude increase in mechanical complexity and software sophistication. Early implementations of such capabilities may begin to appear during this decade, improving steadily with time.
Robotic companions for the elderly and robotic toys might be increasing in sophistication throughout this decade. We might see robots that trim shrubs, and deliver mail and food trays in nursing homes. They might be voice-activated, and capable of voice responses. There might be versions that could move magazines and papers out of the way while vacuuming, or scrubbing the kitchen floor. At some point, kitchen robots capable of preparing food and loading and unloading the dishwasher will appear.
Their first applications will logically be as surrogates for paid human employees. ("May I take your order, please?")
2020 to 2030:
This might be a time when universal robots begin to show up in various domestic settings.
These developments will be gradual, appearing slowly enough that we gradually adjust to what are only more and more complex machines. An example of this kind of evolutionary development is the Windows operating system. The version of Word we’re using to prepare this has automatic spell checking and grammar checking built in to it, but in no way do we consider it anything more than a tool.
2030 to 2040
By the year 2030, Dr. Moravec predicts that robots will approximate human mental capabilities for performing in the real world. (Hans Moravec isn't saying that robots will emulate humans, but that they will perform in ways that allow them to perform some of the same tasks.) It will be at this point (and possibly, earlier) that serious questions about the roles of robots will have to be considered.
It might be the 30's before anthropomorphic robots are common in homes. (A "gynoid" that could bathe the children while reading them a bedtime story or showing them a TV movie might be popular, especially if "she" looks, talks, and acts like a mommy.)
--Animatronic "Life Dolls"?
Quite a bit of technological effort has been invested in dolls that are gradually transitioning from passive to active (with the ultimate goal of androids that can emulate human behavior up to, but short of fighting) I could imagine that a great deal of money will be made this way. At hundreds to thousands of dollars a doll, with thousands-to-millions sold, the revenues could be substantial.
During this period, animatronic robots might act as guides and escorts, might check out groceries, and might partially or largely take over farming operations.
2040 to 2050
One could imagine a situation in which an android is kept warm by its fuel cell, maintaining a body temperature of about 98 degrees. The fuel cell might be cooled by a piston pump or squeeze pump that would act like a heart, and would circulate red liquid (water or alcohol) as a coolant. The android might "drink" alcohol as a fuel. The android might "breathe" in air, and "breathe" out carbon dioxide and water vapor. One could also imagine that the android would have a small battery and/or capacitor bank to even out surge energy requirements, with the fuel cell recharging the battery and/or the capacitor bank.
These design tricks would make androids seem more human, which you could imagine being an engineering challenge, and of value for human companionship (especially for children). (I can see it now: a Stephen Spielberg movie in which the children's robotic playmate turns on them when robots try to take over the world.)
During this period, humaniform robots might replace humans in performing dull, dangerous or dirty tasks.
By 2050, Dr. Moravec envisions robots surpassing humans, perhaps migrating into space, for which they can be well adapted, leaving humanity behind.
Hans Moravec thinks that competitive commercial pressures will lead ineluctably to this result.
It is this prediction of superhuman robotic mentalities that shakes us up. However, computers are already enormously “smarter” than humans in certain areas of endeavor (e. g. arithmetic), but they certainly couldn’t be called self-aware, and so far, they’re certainly no threat to human sovereignty.
One very important point to consider is that computational speed alone won’t bring about the “humanization” of computers. The computer on which we’re preparing this article is already about 1,000,000 times faster than the fastest computers of half a century ago, when there was so much concern about computers outthinking human beings, and yet, it’s absolutely a passive machine. Even though it automatically checks our spelling, there’s no thought involved in that process… merely rote-mechanical checking with entirely predictable outcomes.
Perhaps, the crucial Rubicon will be crossed if computers are given the power (and let’s hope this never happens) to set their own goals, based upon real-world experience, and are equipped with urges to attain their goals through the generation of, and exercise of their own strategies… in other words, if they act with purpose.
There's no reason why robots should have any qualms about their inactivation even if they were self-aware. Lacking the instinct for self-preservation and other biological imperatives, they might have values that were more relaxed than our own. ("I could, but why should I?") Then, too, using RF links, their personae could be stored in an up-to-the-minute central database, so that they would have the promise of reactivation. There’s no reason why they should have a desire to take over from humans or to advance robotic causes over human causes
But this is a topic that will have to be considered in depth as such a possibility looms larger on the horizon.
Humanoid robots would probably want lower-level machines that would perform lower-level functions just as we do. We don't need highly intelligent thermostats. They'd be overqualified for the job.
--A hierarchy of processors?
One interesting difference between artificial intelligence and human intelligence is that there could be a hierarchy of intelligence, ranging from elements that would essentially be purely mechanistic to networked processors that would pool problem-solving power to tackle major problems. (Some processors might run at low clock speeds with low power consumptions until the need arose to greatly increase their clock speeds and power inputs.)
Also machine intelligences could be wirelessly (telepathically) linked in a way that isn't currently feasible with humans.
It’s going to get exciting, particularly two or three decades from now.