Attack of the killer vacuum cleaners

Remember when robots were going to take over the world or, at the very least, bring you a cup of tea in bed? Still hasn't happened. But, says Charles Arthur, their time is finally about to come. Just not in the form we always imagined

Robots hold a persistent fascination in our culture. Ever since Rossum's Universal Robots by Karel Capek in 1920, and the more enduring Metropolis directed by Fritz Lang in 1926, we have wanted robots to do more of the hard work - yet been continually disappointed by what they can actually do.

Robots hold a persistent fascination in our culture. Ever since Rossum's Universal Robots by Karel Capek in 1920, and the more enduring Metropolis directed by Fritz Lang in 1926, we have wanted robots to do more of the hard work - yet been continually disappointed by what they can actually do.

For example, 2001 came and went without any sign of a computer with anything like the intelligence or awareness of Hal from 2001: A Space Odyssey, or C-3PO from Star Wars. Instead, what did we get? Sony's Aibo - a rather limited robot dog. If you ever needed a metaphor for how technological reality lags behind dreams, there it is.

But there are other signs that the robots are rising. Yes, yes, I know about the I, Robot film with Will Smith that is coming up, but it's not to do with that. Nor with Arnold Schwarzenegger in his Terminator incarnation. Instead, the robots of the near future are going to be a lot less visible, and their usefulness much more tightly focused. We already have room-cleaning robot vacuum cleaners - the iRobot Roomba and the Electrolux Trilobite - whose ranks will soon be swelled by one from the Dyson stable. And the Aibo has been joined, in North America, by machines called RoboSapiens, manufactured by Wow Wee, which are "walking" toy robots.

Things are about to happen with robots, because the element they need to make them truly useful - the software, which needs to be able to adapt to a wide range of situations - is getting cheaper all the time. Future Horizons, a semiconductor analyst based in Kent, forecasts that by 2010 there will be 55.5 million robots, in a world market worth £30bn - up from £2.4bn last year. "The electronics industry is on the cusp of a robotics wave, a period in which applications are aimed at labour-saving and extending human skills," it reports. Of those, it says that 39 million will be domestic robots, and 10.5 million "domestic intelligent service" robots.

That is because there's a growing need for robots to help the elderly and handicapped. Why? Because they can, and because there are more people who can benefit from that help. Researchers in Japan and the US in particular are developing robots that can help out around the house: not by vacuum-cleaning the room, necessarily, but just by noticing what you do and when you do something unusual. That doesn't even take dog-level intelligence. For example, say that your daily routine consists of getting up, having a hot drink, taking some tablets, and going out for a half-hour walk. One day, you get up and have a hot drink and go out for your walk. Your domestic robot should spot what you missed out and nag you about it. Or, if you don't get up at all, bother you a bit and then contact an emergency line. Such robotic Lassies could be just the ticket for the future elderly, since they would allow them to live in their own home yet provide the sort of ongoing attention that would otherwise require comparatively expensive human time.

But the real explosion in robotics is coming among the "immobots" - or, more simply, just "bots". These are bits of software that are incorporated into larger objects, and that remove a lot of the strain of having to decide what to do next. We're getting glimpses of how good these could be at present: the tiny number of Britons with a TiVo personal video recorder have something that decides, based on the programmes they choose to record, what other programmes they might like to see, and records those, too. LG Electronics of Taiwan is advertising an "internet fridge", which, it claims, will know when you're running out of eggs, say, and reorder them from the supermarket. Presently, it's only a concept, but an interesting one (although one can immediately see the obstacles: how would it know which online supermarket you favour? How would you get the information in, except through some horrendous computer-to-fridge interface?).

The reason why we can't yet declare "The Year of the Robot", however, is that researchers are still fundamentally split about how robots should behave and learn. One group favours the "top-down" approach, in which all the behaviour of the robot is mapped out, and its software is written to fill out that behaviour. The Roomba vacuum cleaner is a classic example: it has certain functions, such as cleaning and recharging, and it will work pretty much from the moment you plug it in. And you can be confident that it's never going to do anything more than trundle around your room vacuuming it; it's not going to start climbing the walls or shampooing the dog.

The alternative is something assembled from smaller, self-contained units, which creates a gestalt of behaviour based on that. Thus the system that controls the legs learns to "walk" independently, and the "eyes" make sense of what they're seeing by getting feedback from other sensors. Given the limited range of things it can do, your self-learning robot will quickly discover those limits, and the best way to work within them. That's the "self-organising system" approach, which seems to have greater potential. Sony's Aibo draws on a form of this, inasmuch as its behaviour is not completely decided when you buy it; you can train it to recognise your voice, for example.

But both systems have the drawback of not being completely predictable. That is why there is still no sign of robot "butlers". Given how hard we find it to write software that works perfectly when it has only to sit on a computer, a system that has to operate in real time and keep a robot from falling over while it tries to pass you a plate is a far bigger challenge. Even the technology to hold a plate just tightly enough not to drop it, yet not break it, and to know when to let go so that a human can take it, without the two of you dropping it, is a big challenge. That is the sort of skill that can't be programmed by brute force; you need software systems that can learn. Those exist, but there'll be some broken plates first.

Similarly, we're always on the verge of voice-operated speech-recognition services. In fact, they're all around us, but in limited areas, such as when you give your debit-card number, say, to a system that repeats back: "I heard... seven... eight..."

What is truly holding robots back from becoming even more prevalent is that while the price of the software and silicon is dropping all the time, the costs of the hardware and assembly aren't falling so quickly. Ideally, of course, you'd get robots to do it. But the reality today is that getting people in China to do it is preferable: they're cheaper, and more consistent. But that won't last forever.

I've finally succumbed to the temptation to extend the commentary I can make, and started a blog. You'll find it at www.charlesarthur.com/blog. Comments are welcome

network@independent.co.uk