Nice neighbors (of the animal kind)
May 14th 2004WallabyGeneral & nerd alert
Saw a deer on the way home from work today. I was walking along through the engineering campus, which still lots of nice woodsy areas, and there was a youngish-looking white-tailed deer by the side of the road. He (?) let me get to within about 25 paces before he sort of strolled off into the woods. Sometimes I really really like this town. (Mostly in the summers.)
Speaking of work, finished up most of the rewrite for the control software this week. The first time in probably two years that I actually did something useful of a programming nature… fun. Then I started pondering a general-purpose programming environment (can’t really call it a language, exactly) just like LabView but not so narrowly focused on sciencey and engineering stuff. (Not quite sure what else it should have, actually.) Then I decided it’d just be too much pain to do anything much bigger than this with wires and blocks.
Features I did *not* add (yet):
* It does not play music based on how well the experiment is going. (No real support for mp3 playback – as far as I know – and I have no idea what an appropriate soundtrack would be anyway. Something about bones.)
* It doesn’t send me email when it’s all done. (This would actually be useful, since I don’t have to be watching it anymore. Maybe combine this with an email-to-SMS gateway so it can “call” me to let me know.)
* It still doesn’t collect the specimens or put them into the tester for me. Nor does it do any of the data analysis. *sigh*
Noses that glow in the dark…?
Seminar yesterday was someone talking about building an artificial nose using massive arrays of fluorescent sensors. (Extremely cool.) There was a question at the end about quantitative response (i.e., “how much of this dangerous gas is in this room?”), and the speaker made the point in his answer that sensor arrays really aren’t good at quantitative response but that’s okay – because it really is an artificial nose. It even made mistakes that seemed very human – it was trained on two perfumes, and tended to identify one perfume as the other (still >90% accuracy, though). It was trained on a solution of explosive-analogues in water and tended to identify test solutions of explosives in water as, well, water.
People can recognize one compound at a time, or usually a mix of two or perhaps even three different compounds. But at some point, we don’t recognize individual compounds any more – we recognize the whole mixture and say “oh, that smells like apples”. And that’s just what the sensor arrays do: they can’t explain why the mixture smells like apples, perhaps, or quantify individual components of it, but they sure can tell (with careful training) that it’s an apple instead of a banana, or even a ripe apple instead of a green or overripe apple. And it all just happens very automatically, at least after unconscious training, although of course there’s retrieval problems when we can’t remember what that smell reminds us of.
These particular devices can be made pretty small, although miniaturization hasn’t really been a big focus (at least, he didn’t talk about it, which he would have if it were especially tiny). Which somehow suggests armies of robots sniffing for things – land mines? earthquake survivors trapped in fallen buildings? bombs? Especially interesting to combine a squad of robot sniffers with a base station that handles most of the processing and helps them swarm to the right spot…
