“It was pretty pathetic,” recalls electrical engineering student Pei Zhang, describing the scene. In the middle of the Kenyan plains, three Princeton University professors and four Ph.D. students—Zhang among them—all hovered over a metal box, waiting for a green light to flash.

But this was more than an ordinary LED. It was the culmination of three years spent developing and testing a wireless network like none attempted before: a network of computers carried in collars worn by wild zebras. Backed by a $1.3 million grant from the National Science Foundation, Princeton’s ZebraNet project sought breakthroughs in two rather disparate fields: zebra behavior and wireless sensor-network computing.
ZebraNet wouldn’t be able to advance either field unless the little green light glowed, showing the first communication between the base station in their hands and the computer that had been wrapped around a zebra’s neck that morning.

With the zebra in sight, the seven academics anxiously watched the metal box. The light remained dim. Then, finally, a beam of green light appeared, and the group erupted in cheers. The startled zebra took off at a gallop.

Since that first glimmer of success in January 2004, ZebraNet has accumulated an impressive list of accomplishments. It has overturned long-held ideas about zebra behavior. It has met the ambitious goal of creating a wireless network of computers that can process position data and communicate with GPS satellites, a base station, and each other—all powered by a small bank of solar cells rated at 0.4 watts per computer. It has proven the potential to wirelessly update software on a network using peer-to-peer communication. And perhaps most importantly, it has challenged a group of graduate students and their professors to take their research out of the laboratory and into one of the most demanding field environments imaginable. “Engineers spend a lot of time inside in a lab,” says project leader Margaret Martonosi. “This was a once-in-a-lifetime opportunity.”

The project grew out of electrical engineer Martonosi’s work with portable, low-power computers. In one senior thesis project, her students developed a system for GPS-enabled Palm computers to give an automated tour of the Princeton campus, with information provided according to the user’s location.

Princeton zoologist Dan Rubenstein, an international authority on zebras, learned about the automated tour and immediately saw possibilities for his own research. Before ZebraNet, Rubenstein had monitored zebra movements by learning the stripe patterns of individual animals and recording sightings. Other biologists use VHF collars that emit a “ping” signal to track large animals. A researcher takes several readings with an antenna and triangulates those readings on a map to home in on an animal. “It’s slow, it’s not particularly accurate, and it’s very labor intensive,” Rubenstein says.

Rubenstein and Martinosi wanted ZebraNet collars to collect GPS readings several times an hour and to store the readings in flash memory. Even more amazingly, they wanted two computers to be able to swap data whenever one collared zebra came within a kilometer or so of another one. This would mean that if a researcher could find just one collared zebra, he could wirelessly upload GPS data from several zebras. In the same manner, the researcher could perform a software upgrade to the entire system simply by getting within range of one zebra. This capability is essential, says Martonosi, “because it’s extremely difficult to reboot a zebra.”

Sleepless Nights

Moving from these dreams to ZebraNet’s reality would require dozens of technological advances and many sleepless nights, however. Martonosi put several electrical engineering undergraduates to work on small aspects of the project, while relying on four core graduate students who could commit to years of focus on ZebraNet. “We started from scratch,” says one of those grad students, Chris Sadler. “We built everything.”

The group’s first priority had to be power efficiency. Sensor networks like ZebraNet are usually made up of remote nodes that must run on their own power for months or years. A sensor network, in the words of Sadler, is “the poster child for power supply difficulties.” ZebraNet’s computers are built around a microcontroller that can switch between two different clock speeds. The faster, 8-megahertz clock is only used for brief bursts of computing power, such as when receiving a fix from GPS satellites. The slower, 32-kilohertz clock handles routine functions at half the power consumption. ZebraNet’s software turns on the battery-gobbling radio and GPS chip only when required.

As the group refined their system, they tested collars on bicycles, cars and horses. “Someone suggested we put the collars on grad students, but Margaret shot it down,” recalls Zhang. “She said she won’t be known as the professor who used GPS to track her grad students.”

In the third year of the project, the ZebraNet team began preparing for their first deployment in Kenya. Two hundred components had to be soldered onto each of 20 circuit boards, often with the aid of a microscope. “I didn’t sleep for a whole week,” says Zhang.

Despite the thrill of seeing that initial green light, the first deployment was not an unqualified success. Heavy rains shorted the solar cells on the collars, and the five-mile radios turned out to have a range of a kilometer or less. Still, Rubenstein collected 24 hours of tantalizing data suggesting that everything biologists thought they knew about the nighttime activity of zebras might be wrong. And the team had proof that the concept of ZebraNet was valid.

Perhaps equally important was the experience. Zhang notes that when signing up for a graduate degree in electrical engineering, “I envisioned myself in a lab in a basement all day long, and here I was, outside, with giraffes walking by while I was using a soldering gun.” Working among both biologists and zebras, Sadler came to realize that “it’s not just ones and zeros floating around from collar to collar; it actually means something.”

Shades of Gray

Back in New Jersey, the ZebraNet team had 15 months to polish the rough edges of their system before the second deployment. They gave the mobile base station the capability to initiate communication with collars on demand, rather than waiting for a scheduled radio transmission every two hours. “We had figured in the worst-case scenario the biologist would have to follow an animal for two hours,” says Sadler. “In the lab that sounded fine, but when we were out there we saw there are no roads, they’re dodging ditches, animals run away, and you lose track of them.”

Software was improved to automatically produce maps of zebra movements from the raw data without assistance from the researcher. “Not all biologists are computer whizzes,” says Sadler, “so we spent more time on user interfaces.” Since the radios had proven to be weak and inefficient, Sadler wrote new compression algorithms to reduce substantially the quantity of data to be transmitted, shifting the burden onto the more effective number-crunching power of the microcontroller. And much time went into waterproofing the solar cells.

By the second deployment, in mid-2005, the system proved its worth. With 15 days of data showing unprecedented detail, Rubenstein could say with confidence what he had first suspected after seeing the 2004 results: Plains zebras do not spend their nights quietly resting in the middle of open expanses, as previously thought. They head for ravines and woodlands for much of the night—presumably to avoid lions—and move more quickly when they do pass through the grasslands.

Today Martonosi and her team are finishing their last papers on the project and wrapping up ZebraNet. The project never quite achieved its goal of producing a collar that would survive under its own power for a full year. And radio range remained a disappointment. But she doesn’t mind leaving these technical wrinkles to be ironed out by private companies that want to commercialize GPS radio collars and related sensor networks. As it is, Martonosi worries about the amount of time her students spent fiddling with waterproof seals and soldering guns.

“I want to give grad students good guidance about the best use of their time,” says Martonosi, who has five patents and dozens of publications. “Grad students are the ones in the trenches, and they want to publish and do dissertation research. If they just stayed in the lab and wrote papers about simulations, they might have less real-world experience but more papers.”

So do she and her students regret their time out of the lab? Not a bit. Martonosi believes that her ZebraNet students now have an advantage, because “they’re not just simulator jockeys. They understand how real system design constraints come into play.” Sadler appreciates that advantage now during job interviews with corporations such as Microsoft and Google. “ZebraNet comes up constantly,” says the 27-year-old. “I guarantee it gives me a leg up.” And Zhang, who wants to stay in academia, says that ZebraNet has changed his whole approach to research. “Any work I think of now, I first think how to implement it in life,” he says. “Because I know what’s needed in the real world, people can trust the work.”

Don Boroughs is a freelance writer in Johannesburg, South Africa.

TOPˆ