Slashdot Mirror
Flexible Sensors Make Robot Skin
Roland Piquepaille writes "In recent years, lots of efforts have been made to give robots the ability to hear and see. But what about the sense of touch? Unlike us, robots don't have sensitive skin. But this is about to change. By using organic, or plastic, field-effect transistors as pressure sensors deposited on a flexible material, researchers at the University of Tokyo have created an artificial skin which will give robots the sense of touch. The prototype has a density of 16 sensors per square centimeter, far from the 1,500 of our fingertips. When this density increases and when the problem of the reliability of this kind of transistors is solved, the researchers say this artificial skin will also be used for car seats or gym carpets. Expect to see them in four or five years. More details and a picture of a robotic hand using organic transistors as pressure sensors."
More real realdoll.
*big smile*
Oh yeah
KITT: "You know, Mike, we need to talk about how you're doing on your diet."
tasks(723) drafts(105) languages(484) examples(29106)
"Look at yourself, standing there, cradling the new flesh I've given you. If it means nothing to you, why protect it?"
"I... I am simply imitating the behavior of humans."
"You're becoming more human all the time. . .Now you're learning how to lie."
"My programming was not designed to process these sensations."
"Then tear the skin from your limb as you would a defective circuit...Go ahead...! We won't stop you! Do it! Don't be tempted by flesh!"
The only acceptable defense of scientific results is to say that they were the product of the Scientific Method.
It could work, but other technology needs to catch up first. Fairly detailed sensors could be installed in current prosthetics, I'm sure, but the machine-nerve interface just doesn't carry enough data yet. It doesn't matter whether we know what that data means, since the brain can probably learn to interpret it on its own, but we just don't have the fine control over the interface that we would need. In related news, an article in this month's Discover (full text viewable to subscribers) discusses a lot of these limitations, although it comes at it from the angle of whether mind-reading (or controlling) computer chips are possible.
Everybody complaining about Slashdot becoming Piquepaille's personal soapbox for plagiarisim seems to get instantly modded down. Is he a pseudonym for one of the Slashdot editors or something?
Anyway, what is the robot ability up to now?
* Has skin
* Eats flies
* Can transform into other robots
* Walks on water
It sounds like the plans are coming together nicely for overlord robots.
how does one overcome the sensor-sensation gap?
In the case of a sex android one does not even bother trying, as it's only the sensations produced in the human componant that matter.
All the android has be able to do is adeptly fake it, making it more of an android wife than an android girlfriend.
KFG
I started a Masters degree on this issue in the 1980's and it's sad to see the same *wrong* approach to touch still being applied if the end use is a robotic hand/finger. At the time MIT was doing work on this, as were a few other places, all with the wrong approach. Here's the problem:
It's not the sensors or the density or how long they last or their accuracy or anything like that, even though these are real problems. The big killer problem is wiring. You get all these signals and at some point you need to get the wiring over joints that have to bend a real lot. And the more sensors you have the wires your typically going to have. Eventually you end up with bundles of wires and the simple fact is bundles of wires do not like being bent repeatedly, apart from which fingers need to be skinny to be useful and this is at odds with fat bundles of wires.
One solution however is physically simple and was presented at a National robotics conference in Australia in 1990. In summary I proposed and had made a working 2D slice of finger that used only 4 sensors. A 3D finger tip would require about 9 sensors, and by finger tip I mean measuring the major contact, magnitude and direction anywhere beyond the joint. The method was based on normal engineering and had the 4 sensors buried into a compliant skin. An external force caused a reading on all 4 strain gauges. From this small amount of data a PC worked out the magnitude, position and direction of the applied force using data collected from earlier testing. As a 2D finger slice it could successfully follow an edge when attached to a robot arm. I can scan and email the paper (this was pre net days) if any researchers want to extend this work and come up with practical robotic fingers. Email me.
Another solution is to put the smarts into the skin so only a "summary" signal needs to go back through the various joints. This couldn't be done in the 80's but could be now?