Smell chips for robots – is that me that’s on fire?
A comapny called Cyrano Sciences
has developed an odor-sensitive chip called the Nosechip
which should find applications in an artificial nose. While the sensitivity is relatively low, the system detects the overall pattern of smell (which may be due to many smell molecules) rather than a few specific compounds. The supplied software uses pattern recognition algorithms to classify overall smell environments. At present, the chips cost thousands of dollars, but costs will come down and sensitivity will go up.
Not a typical image – but real robots that jump will most certainly be able to smell us – and themselves.
Adding smell is a good thing for robots. In the past, the problem with robots is that they responded to a tiny bit of their overall environment – 2D vision for example. However, in biology is is evident that the smartest animals don’t optimize one sense, but rather build models of their world from a variety of sense information. Thus, a cat might integrate a mix of vision (a dark patch of leaves with some motion), sound (rustling), smell (mousey fragrance), kinesthetic sense (I am crouched for a spring) and history (I followed this noise here and it has continued) into an overall model of a mouse. In contrast, robo-engineers have spent fruitless decades trying to develop algorithms that would correctly classify the leaf movement as caused by a mouse – something a human being might have trouble doing.
Adding smell, even at low-resolution, offers a robot a change to tie more closely into its environment and be less of an disembodied machine.
Use of smell in robots that jump need not be complex. The robot doesn’t have to decompose a smell into all its components in order to use a sense of smell. For example, if the pattern of “locker room smell” can be detected, it would make the robot’s navigation more effective – when asked to go to the locker room it would be reassured by the smell that it had arrived in the right place! Moisture and temperature sensors would also be useful. Smell also might be part of a self-diagnostic routine – an electrical problem might be identified by burning insulation smell, causing the robot to “safe” itself. Finally, with a more sensitive nose the robot would have an easier time telling a (smelly) human from a (smell-free) dummy.
A long time ago, scientists used detergents on a woman in a way to totally remove her natural smell for a short period. The extremely clean subject reported the unnatural feelings this caused. Robots need to smell as well.
While the NoseChip is currently the only option, other researchers are busily working on tiny chem labs on a chip. These chips contain tiny tubes, valves, and reservoirs. It seems that it will be a cinch to create pin-sized gas chromatographs, resulting in ultra-sensitive – and cheap smells. Like the human eye, the robot could have a coarse, low-sensitive “pattern recognition” smell coupled with a high-res smell allowing individual compounds to be picked out.
If there was any chip that belongs in Sony’s dog-robot Aibo, this is it!
– posted by Pete @ 2:30 PM
US versus Japan in robots
An interesting new online magazine has opened at Robotics Trends
. This ambitious magazine plans to cover all aspects of the ‘new’ non-industrial robotics – personal robots, service robots, security and defense robots, sports and entertainment robots. In addition to the typical newsfeeds the magazine features interviews with various hotshots in the new robotics industry.
One of these is Colin Angle, the founder and CEO of iRobot, which makes the wildly successful Roomba robotic vacumn cleaner. In fact, at present iRobot is the only US robotics company selling successfuly to consumers – others are at the “incubation” stage or are exclusively selling to military and defense. In part of the Robotics Trends interview, Angle describes his desire to make a real mass-market robot rather than the impractical “one-off” demonstration robots found in most academic departments. He also decries the exclusive reliance on government/military money for robot funding found in the US – the Roomba is a self-conscious effort to make a robot work as a real consumer product.
So far so good – this in a nutshell describes what has been wrong with robotics research during the past several decades in the U.S.
However, later in the article, Angle comments unfavorably on the advanced humanoid robots being developed in Japan and other Asian countries. From the discussion it is clear that he considers systems like the Asimo impractical for everyday use, similar to the “university robots” he encountered at MIT. The assumption appears to be that they are too complex and do too little. He also discounts the technical lead in producing complex robots Japan has over the US, insisting that when they do become practical the money spent to be the “first mover” won’t matter – anyone will be able to duplicate the Japanese systems. Finally, he pooh-poohs the goal of the Asimo, Aibo, and QURO robts – as entertainment toys they will never be more than “mascots” for the companies producing them, namely Honda and Sony.
I don’t completely buy either idea. Sure, the Asimo is not a personal robot, but the QURO and Aibo are. Despite their far greater complexity, they are consumer products and therefore have a reliability comparable to the Roomba.
What Angle misses – or doesn’t say – is that a lot of the big bucks Japan has put into its robots has been into reliability. The new Aibo has over 40 sensors – comparable to a MIT robot Angle felt was impractical – but it is being sold to consumers. In contrast, the Roomba only has a couple of sensors. In other words, the Japanese are learning how to produce robots that are both complex and reliable, unlike the US.
This is not due to the difference in US versus Japanese “robot culture” but a simple matter of money – Honda had 100x the money and 10x the time to create the Asimo that Angle had to create the Roomba. The Roomba’s development had cost constraints not experience by the Japanese group. That isn’t to say that the Roomba is not an amazing accomplishment – a $200 robot that is actually selling. I suppose I understand that Angle doesn’t want to come off like he’s whining for support, but the fact is that the US at present doesn’t support non-university applied research into making practical robots.
Angle’s second point can also be questioned. Sure, if the Asimo is built into a truly practical robot during the next decade it will be possible for US companies to do the same in relatively short order. But this is unlikely. Instead, we will see a mirror image of the 1980s during which Japanese companies focused on simple PC components (memory chips) and ultimately dominated the market. Today, we don’t buy our computers from Asia – instead, the components sans the Pentium/PowerPC are created there, assembled there, and shipped here. If Japan produces advanced robots our likely contribution will most likely be to create advanced components for their systems. When the super-Asimos circa 2015 are walking off their container ships in Long Beach they will containe American components. Some comfort. Even this is optimistic.
At present, the US has “hollowed out” its manufacturing base and produces very little physical product. Now, IT jobs and tech-related service jobs (e.g. call centers) are also moving offshore. What we’re left with is a huge number of lower-paid “service jobs” – flipping burgers and doing each other’s nails. Ou other big export is media (movies, music and television) and financial services (e.g. the US dollar as the world’s currency reserve). Unfortunately, there’s no sign of this trend reversing – during the next decade, millions of jobs in software, design, and tech support are going to move offshore.
So, when the time comes to produce advanced robots, we won’t be making the parts. We could grab the idea and make the parts, but we won’t have factories to make them. We could finance the development of new systems but it is unlikely that the US dollar will have the power it has today. Some other currency (Euro, Yuan) is likely to take its place.
So what will we do? We will buy robots from Japan and sell services back and forth to each other related to them. We’ll pay for this by continuing to export whatever remains of our technical know-how. Remember that the post-GenX generation hitting college during the next couple of decades is rejecting science and technology as a career. The decades long decline in science and engineering as a US career is likely to continue. As present, the gap is closed by the large number of non-US citizen coming to the US, but will this continue forever. Unless US universities maintain their science/tech lead in resources it will not.
To summarize, by 2015, iRobot will probably be producing its robots outside the US for non-US consumers. If it has work in the US it will be in the defense/security industry which has been the only source of US funding for four decades. The US will remain a backwater for consumer robots, and most likely the current US standard of living will have dropped to levels comparable to Asia, eliminating the financial angle. In short, it is not clear that there will be any point to the US copying the Japanese robotics industry.
Angle’s final point in the Robot Trends article is that current Japanese entertainment robots are nothing more than overpriced PR stunts for their companies. Here I’ll cut him some slack, at least for the Asimo. For the moment it does little more than walk around and amaze schoolkids. But Angle implies that, being complex, it is unreliable. This isn’t right. Honda isn’t afraid of having several Asimos stomp around onstage, without falling or suffering mechanical breakdowns. Honda is clearly applying its huge resources for creating efficient, complex consumer products (namely cars) to Asimo. The Asimo is not a one-off wonder like US university robots. It is a product – early generation, to be sure, but real. It can move through human environment better than any robot ever created. If robot senses (e.g. vision) improve enough it will be a platform for a variety of real, non-PR projects.
What goes for Asimo goes double for Aibo and QURO. Sony has created pricey, but real consumer entertainment robots. The reliability of an Aibo is comparable to many consumer electronic devices. $1,500 is pricey, but compare that to the $400,000 robots being sold by Neiman-Marcus. Hey, the latter are less sense-studded and capable than the Aibo. Why? The weren’t manufactured in Asia.
I very much doubt that Sony is creating robots simply as PR toys. They may be serving this role now, but consider IBM’s original attitude to the PC. After Apple made the cover of national magazines in 1980s, IBM decided to produce its own PC to keep its hand in. The original IBM PC wasn’t meant as more than proof that IBM was in the PC market. The Intel/PC computer ultimately became dominant because third-party computer companies emulated the IBM PC and swept the market. If IBM had maintained its interest it would have controlled the PC world.
So, the only danger for Sony and Honda is that they grow tired of being out in front in robotics. If Honda suddenly decides to give away its research on the Asimo for free (which is exactly what IBM did) US companies may have a fighting chance. Otherwise, they will fade, or move to Asia to be part of the action.