Engineers help biologists' dreams come true


ZebraNet to collect data bounty

bySara Peters

To be a wildlife biologist, one must also be a bit of a sneak. Creeping up on a group of animals, hiding in the tall grasses, hoping that the click of the camera, the hushed whispers, and the sound of a pen scribbling notes on a pad won't frighten them off. Because if it happens that an alarm goes off and the whole herd goes sprinting across the savanna, there's no telling how long it will take to find them again, or what intriguing data will be missed in the meantime.

It's not easy being a biologist. If only there were a way to observe animals closely, without invading their territory.

Soon there will be. Princeton scientists and engineers are now developing a sophisticated system to track animals--specifically zebras--in the wild: a wireless sensor network dubbed ZebraNet.

About ZebraNet

ZebraNet will be established in the Mpala Research Centre in Kenya. Mpala is a biology field station administered by Princeton University, the Kenya Wildlife Service, the National Museums of Kenya, the Mpala Wildlife Foundation, and the Smithsonian Institute.

ZebraNet will collect round-the-clock data on zebras over the course of one year through the use of a sophisticated sensor network.

A sensor network like ZebraNet consists of several data collecting locations, called nodes, which communicate with a base station that compiles the data from all nodes. In the case of ZebraNet, the base station will be mobile.

Scientists will just pack the computer into a truck or a plane. Each node is a tracking collar carried by an animal in the wild. These nodes contain a global positioning system (GPS), storage cells, a wireless transceiver, a CPU, solar cells, and batteries.

These nodes will periodically collect and store information about the animal's location, heart rate, body temperature, and feeding frequency. This information will be transmitted back to the base station when the researchers fly or drive by within a certain distance from the nodes. The plan is to collar 30 to 50 animals and use the data to paint a clearer picture of their migration patterns, eating habits, and inter- and intra-species interactions as well as the impact of human action on their activity. Because the collars will collect data all day and night, scientists can enjoy their observation efforts without disturbing the animals.

The ZebraNet project is a collaborative effort between Princeton's Departments of Electrical Engineering (EE) and Ecology and Evolutionary Biology (EEB).

"It's a great Princeton story," said Principal Investigator Margaret Martonosi, associate professor of electrical engineering.

Inspired by students

A few years ago, Professor Martonosi--along with Professors Perry Cook, David Dobkin, and Stephen Lyon--was advising a pair of undergraduates working on a senior project. Andrew Steiner '00 of EE and Daniel Davenport '00 of CS created an interactive campus tour by using a GPS hooked to a palm pilot. The GPS could detect a tourists' precise location and provide information about the buildings and the sculptures in that place.

The buzz about this project made its way to Professor Daniel Rubenstein, professor and chairman of the EEB department.

"Dan e-mailed me and said 'I should tell you about zebras some time,'" Professor Martonosi said. "I thought he was going to give me some tourism advice."

To her surprise, Professor Rubenstein wanted to discuss the possibility of applying this senior project prototype to wildlife research. The idea was that if the contraption could track people roving around campus perhaps it could track animals in the wild as well.

"Normally, when you're studying animals you have to find them before you can watch them, and they often run away from you," said Professor Rubenstein, one of the project's coprincipal investigators. "We'll be able to get a sense of how a whole population is using a landscape."

Intrigued by their conversations, Professor Martonosi stopped by the Mpala Research Centre during a vacation she made in Africa in the summer of 2001. During this visit she got a feel for the challenges inherent in the task Professor Rubenstein was suggesting and became inspired.

"This project is novel in an engineering sense, and novel in a biological sense. We scored big with NSF because we are solving needs in two very different fields."

-- Daniel Rubenstein, professor and chair, ecology and evolutionary biology

Biologist's dream

When it's up and running ZebraNet will be a biologist's dream. Over the course of one year the collars will take GPS samples every three minutes and collect detailed activity logs over a three-minute span once every hour.

After the animals are captured, collared, and released, the researchers will only require scant contact with them as they drive or fly by the base station.

From this extensive, 24-7 data, biologists' will make many inferences about zebras and their ecosystem.

Zebras are good animals to begin the project with be-cause the social structure of certain zebra species makes the engineers' job easier. Several animals move in a tightly knit group called a "harem," consisting of one male, several females, and their offspring. Since the harem is a long-term union and all the animals in a harem move together, the engineers only need to collar one animal in that particular group.

Periodically many harems will meet up, forming a large, temporary group termed a "herd." Scientists wonder: Why do the zebras do this? How do they collectively choose when to form larger groups and when to migrate? Are there any social or biological changes that occur in preparation for this event?

ZebraNet will help answer these questions. It will give the scientists better information about migration patterns, which will lend insights into these intricate herd dynamics, as well as give them information about speciation, and the mixing of gene pools.

Better migration pattern information will also help biologists observe how human action and expansion impacts the ecosystem. Professor Rubenstein hopes to eventually be able to collar more species throughout the food chain to learn about other interspecies interactions.

"There is a whole richness of biological questions we could answer," Professor Rubenstein said.

Engineer's nightmare?

While a biologist's dream, this project could be an electrical engineer's nightmare.

The task is to create a wireless network that will range over an undetermined number of miles, in the middle of the wilderness, in a region where no cellular network exists, attaching each node to the neck of a wild animal that will undoubtedly spend a considerable amount of its time standing in the rain and running from predators.

The engineers, however, are undaunted. Their design choices are based upon the environment, the particular needs of the biologists, and other situations that challenge the system.

One of the biologists' considerations is weight. If the collar is so heavy that a zebra couldn't wear one while continuing his normal, daily behavior, it would be of little use as a source of accurate data.

All components of the zebra collar must add up to no more than five pounds, which limits the number of weighty batteries that can be included in the node.

Thus, power becomes a major limiting factor for the engineers. The goal is to have a power supply system in which the battery will be able to operate for five full days between recharges by solar panels that line the collar.

Spatial range

Another major consideration is spatial range. This network must span the entire migration pattern of several harems in a zebra herd over the course of one year.

Since there is no existing cellular network established in the region that could be tapped into, a complete network must be set up.

The region, however, cannot support a fixed base station because of the harsh environment, as well as the chances of vandalism.

The alternative is to design a peer-to-peer network with a mobile, intermittent base station. What this means is that each node will collect and store the data for the individual animal wearing the collar. Each collar will be able to store up to 100 days of information for one animal.

Intermittently, scientists will drive by or fly over the region. When a node comes within a certain proximity of the base station, it will transmit all of the stored information. The scientists can then compile and observe the data at their leisure back at the lab.

Information swap

Another essential ingredient of the peer-to-peer network is that peers can swap information. Occasionally, all nodes in the network will perform what is called a "peer discovery."

At a given time, all nodes will search for other nodes within a certain range and transmit their own data to one another. This way, if one zebra who is particularly shy around humans never comes close enough to the base station to transmit his data directly, then a more gregarious friend of his may transmit the data on his behalf. This increases the data homing success rate.

There are two methods of swapping that the engineers are considering. One is called "flooding," a method in which a peer sends his data to every single peer within range. This is quite reliable, yet it does have a major drawback.

Data swaps use more power than any other operation the nodes perform. Since power is a major concern in ZebraNet, flooding may not be the best choice.

A more selective, power-efficient method is called a history-based method.

In this method, a node "chooses" which other nodes to send data to, based upon its track record of communicating data back to base. It wouldn't waste energy swapping data with a node worn by an animal that was notoriously bad at getting within range of the base station.

Impala software

Another concern for the engineers is that changes may need to be made to the nodes' software if errors are discovered and adjustments need to be made.

"One of the coolest developments of the past four months," Professor Martonosi said, "is a software system we call Impala. It allows for adaptation between different protocols, depending on how they are performing at that particular moment."

Ting Liu, a graduate student in the Department of Computer Science, has been committing most of her time to working on Impala.

"Since these collars are going to be on wild animals and you can't just press a reboot button when you want to upgrade the software, one of the things we've been thinking about is the ability to send out wireless software updates," Professor Martonosi said. "The Impala software also allows a hardware node to accept software over the radio so that we can do software updates or bug fixes remotely, without physically touching the collar."

Autonomous system

Wireless software updates, the peer-to-peer network, data swapping, and a rechargeable power source will make ZebraNet an almost entirely autonomous system.

Although ZebraNet is being designed for the specific purpose of biological research, Professor Martonosi believes that the engineering breakthroughs the team is making will have implications in many other fields.

"The notion of an auton-omous network system that needs little human intervention is really satisfying to anyone who's tinkered extensively with a computer or a home appliance," Professor Martonosi said. "These things really should be more auton-omous, even if you can easily reach the reboot button."

The attitude of the engineers is not that ZebraNet is a nightmare, but that it is merely an exciting set of challenges.

"The challenges have led to some really interesting research thoughts," Professor Martonosi said. "Wireless software updates are something we might not have thought about if the realities of the system we're trying to deploy didn't make us think about it.

"In fact, that's one of the neat things I find in a lot of engineering research. When you set out to build something there are a lot of corner cases that you have to deal with."

In addition to being a novel source of research ideas, ZebraNet is an enjoyable departure from the norm for Professor Martonosi.

"From a personal standpoint, this is a ton of fun," she said. "I've always enjoyed being outside, but being a computer engineer is one of those jobs that rarely lets you be outside. So the ability to combine what I enjoy as a hobby and what I enjoy as a professional is very exciting."

Looking ahead

In addition to Professors Martonosi and Rubenstein, the ZebraNet team includes EE Professors H. Vincent Poor *76 *77 and Stephen Lyon, EE Assistant Professor Li-Shiuan Peh, EE graduate students Hidekazu Oki, Chris Sadler, Pei Zhang, and Philo Juang, and CS graduate students Ting Liu and Yong Wang.

The team is currently working on prototypes. They just recently completed the design of the second generation of ZebraNet cards, which are smaller, more integrated, and use less power than the first generation (see photo on page 11).

They've got some collars able to communicate across campus and they hope to test them on horses at a local stable soon. The plan is to begin testing in Africa this summer and install the final system next year.

Professor Rubenstein is understandably anxious to put the system to use. He said his colleagues are chomping at the bit to put ZebraNet collars on the animals they study.

"So much of our work has always been hidden from us biologists," he said. "ZebraNet will give us a much larger window to peer through. The invisible will become visible."