Sensors of the World, Unite!

Continued from page 1

By Technology Review

June 27, 2003

Poor: Say you're on your way to work Let's put a wireless embedded node in every streetlamp. Why would you do that? The town of Brookline, MA, spends about $200,000 a year to drive someone around in a truck. And they have to do it both day and night, because in the day they're looking for lights that are stuck on, which is costing them money, and at night they're looking for lights that are burned out. Both are problems. If all these lights were networked, they could just trickle their data back slowly to the control center and provide an alert as to which lamps are burnt out or stuck on.

TR: That would certainly help the public works department, but how would it affect the rest of us?
Poor: Once you've got the nodes in place, the entire town of Brookline has a wireless mesh spread over it, which can be used for other purposes. This is Metcalfe's Law-that is, a network's value increases with the number of things connected to it. So you put a wireless network node in every bus and as it drives past a street lamp, it notes that the bus is there and it can zip the information four blocks ahead to the next bus kiosk that says that #39 will be here in three minutes. And put one on the kids who are on the bus, and it can make sure they got on the bus at the right place and got to school at the right place and nothing happened in between. We can skirt the issues of privacy here, because this is something their parents want them to have. And this is all overlaying a network that was installed for an entirely different reason.  I like to call it Moore's Law meets Metcalfe's Law.  Silicon has gotten cheap, wiring has not. So if you build networks that make their connections via silicon radios rather than wire, they're going to get cheaper, which means they can get more pervasive. You can't do that if it's a wired network.

TR: What's the cost of setting up one of your embedded device networks?
Poor: Right now, if you add one of these $5 devices to a $2 light switch, you've got on the order of $10 cost. And installing it really means taking a piece of double-stick tape or Velcro and thwacking it on the wall. The fact that you make a self-organizing network simplifies things, so there's really not much additional labor associated with that.

TR: What do you see as the big challenges in the next year, and next five years?
Poor: It's difficult to predict where the real problems are going to lie, because it's really going to be opportunistic-different applications will catch on at different times. The requirements of a system are highly dependent on the actual application, and it will be a delectably messy environment for a while. Coming up with scalable and adaptable architectures that work across multiple applications is going to be critical.

TR: Scaling up a self-organizing device network from a few nodes to a hundred is a problem that lots of people are trying to solve. How big are your networks?
Poor: We're deploying multiple-hundred node networks now. We have simulated up to multiple thousands and one of our challenges has been to find a physical space to lay out that large a network.
 
TR: Are you working toward making the network itself into some sort of collective consciousness or intelligence?
Poor: I think it was [MIT Artificial Intelligence Lab Director] Rod Brooks who used to promote behavior over intelligence. You know, he said, insects display behavior and when you put enough insects together, then you start to see something that resembles intelligence. The classic example is termites. One termite by itself runs around randomly, 10 termites will roll dirt into a ball. But you get a couple million of them together and they build these incredibly complex nests. Where's the intelligence there? So I think it's misleading to talk about things on the order of light switches and thermostats and strain gauges and tank level centers and such. It may be a mistake to talk about intelligence per se, but each of those points on a network needs to have enough behavior that they will self assemble into a large and scalable network. And then we point to the aggregation. There'll be some nodes that have more computational power than others, which will do some kind of analysis, and distill the data down to more useful information. Is that intelligence? A little bit. It's more like behavior.

TR: If you had to boil down Ember's contribution to one idea, what would it be?
Poor: We took the gutsy move of going from what was a fairly pure research topic into commercial systems. There are other people that were working in defense systems, and that's nice because there are always clear funding paths for that. So I think we were one of the first to actually open up embedded device networking to a commercial audience. The gratifying thing is that there's now a lot of other people in the space, and we think that has validated the market.