Tracking Zebras in the Wiredlessness

ZebraNet Wireless Devices
Prof. Margaret Martonosi and students Hidekazu Oki (foreground) and Philo Juang (background) display ZebraNet wireless devices.

Though it may seem counterintuitive to scientific rigor, serendipity can play an important role in promoting progress in research. In Calit², we tend to think this is most likely to occur at the boundaries where disciplines meet, where differing cultures are shared and new languages learned.

This truth of this philosophy was reinforced recently in a phone conversation with Margaret Martonosi, a professor of Electrical Engineering at Princeton. She has been invited to UCSD Friday, November 22 to give a seminar about a project at one of those boundaries.

In her case, the boundary, which she happily points to now as an "overlap," is between electrical engineering and biology. In a three-year project funded by NSF, groups in the two disciplines are focused on using energy-efficient computing to track wildlife. The project, in collaboration with EE colleagues Li-Shiuan Peh and Vince Poor, and Dan Rubenstein, chair of Ecology and Evolutionary Biology at Princeton, is known as "ZebraNet."

How did this project come about? "I specialize in power-efficient computer architectures and had been working with some seniors on an independent project," says Martonosi. "They were studying energy dissipation of a wirelessly enabled Palm Pilot system. The point of the system was to couple GPS and a Palm Pilot together in an application to offer a campus tour."

In chatting with her colleague Rubenstein about the seniors' project, the two realized that, if they could track people, they could track zebras, chosen because zebras and related equines are Rubenstein's research specialty. "I just happened to be going to Africa on vacation shortly thereafter, and the rest is history," she says. Serendipity, indeed.

That formative conversation took place some 18 months ago. Last fall, an M.Eng. student started developing hardware prototypes, and both undergrads and Ph.D. students have been working on simulation and communication protocol software.

"From my engineering point of view," says Martonosi, "this project is about wireless sensornets and mobility - not only are the sensors themselves mobile, but the base station is mobile too. We're also using this project to study large-scale, long-lived sensornets and, of particular interest to me, energy conservation."

The biologists, for their part, are interested in applying the technology to study patterns of animal behavior that has not been possible before, in particular, long-distance migration patterns, how encroaching development appears to affect animal populations, and how flexible the animals are in co-existence with increasingly urbanized human populations. In time, the biologists hope to extend the work to study inter-species interactions.

According to Martonosi, current wildlife tracking technology is more primitive than you might guess. The most common way to track wild animal populations is by equipping them with collars with VHF transmitters. "The main problem," she says, "is that the only way you can capture the collar data is by flyover or by using an unwieldy antenna to triangulate off the VHF collar's signal. So you can only get very occasional samples of what looks, from a data perspective, as static. The results must be like watching a very old and very jerky movie when what you want is smooth, continuous representation of movement so you can understand the subtleties." Moreover, she points out, the flyovers can only be done during daylight hours, so half the animals' behavior remains unobserved using this methodology.

Obviously, the researchers can't count on cellular coverage in the wild, so they plan to equip the zebras with radio-enabled collars, making them individual ad hoc sensors that will form a "network" by "discovering" their neighbors and self-organizing to perform the routing of the data. The architectural model is based on a many-to-one (zebras-to-base station) communication model with the communication component leveraging the animals' instinctual interest in interacting with each other. The zebras themselves would probably be amused at this "discovery" model since, being herd animals, they obviously already know their neighbors.

Martonosi is quick to point out that this peer-to-peer data forwarding architecture is distinct from data forwarding through a central service. Martonosi likens her architecture to one in which your cell phone serves as a carrier for others' calls.

"We don't use any broadcast communication - satellite or cell phone network - but we do use satellite for GPS, and the rest of the communication occurs via peer-to-peer data transmission. Each zebra will have a transmitter capable of five miles. As they range, they will periodically go into peer discovery mode: The transmitters will wake up, query other collars within range, and, if they find one, share the data both ways. The zebras will keep wandering and swapping data until one gets in range of the base station to download all the collected data."

According to Martonosi, the peer-to-peer model has at least two other benefits: It enables the transfer of data to be shorter, which, in turn, saves energy, and it enables recovery of data from more reclusive animals that would be less likely to come in range of the base station.

The GPS component enables sampling zebra position every three minutes, with more detail captured once every hour. "What the biologists also want," says Martonosi, "is to be able to sample whether the head is up or down, which has implications for grazing behavior, the ambient temperature, and the zebra's body temperature, heart rate, and interactions with other species."

"We've had no end of design challenges," laughs Martonosi. The animals range over 100-500 kilometers often over difficult terrain; there are no fixed base stations to capture the data ; they have to plan for power generation and storage with special attention to conservation; and the equipment needs to be sufficiently reliable, fault tolerant, and rugged.

"The equipment," says Martonosi, "can only be replaced by tranquilizing the affected animal and that, in turn, is likely to affect the natural movement pattern of herd mates. The devices need to be waterproof, shock-proof, and, even bite-proof."

The size and weight of the devices, of course, have to take into account the size of the animals and their ability to carry them. The device the researchers have prototyped integrates a GPS chip and CPU, short- and long-range radios, a packet modem, a battery, and a solar cell, all weighing about 2.4 pounds.

"This size is not a problem for an animal as large as a zebra, I'm told," says Martonosi, "but for smaller animals - such as hyenas and wild dogs, which are of interest to the biologists - the device needs to be smaller."

Turning back to the engineering, she goes on to explain that such weight limits constrain battery capacity, which, in turn, constrains the energy available to fuel the system. ZebraNet is designed to operate for five days without solar recharge.

The hardware engineering has been complemented by software developed, including a probabilistic simulator to evaluate protocol tradeoffs under different mobility assumptions, mobility models based on biological observational data, and even a "visualizer" of zebra motion.

The research plan calls for implementing the collars first on domesticated horses in New Jersey. "We have also applied to the State of New Jersey for related research support, but of course New Jersey has no zebras," says Martonosi. "So we structured that proposal around deer management, which is important to suburban areas, especially in the northeast. Deer are a good interim substitute for zebras because they manifest particular movement patterns you can model - speed, clustering patterns, and so forth."

Next the researchers will turn to outfitting wild horses on the barrier islands off the coast of North Carolina and Virginia. "Then," says Martonosi, "we're going to the Mpala Research Centre in Kenya for the real thing on zebras, probably next summer."

When asked what they've learned, Martonosi responds, "We have a good idea of the types of protocols that work well, both in software and communications, in transferring data around the system. We also have a good understanding of the radio ranges appropriate for this scale of network: We're trying to cover hundreds of kilometers using 30 to 50 collars, so the network is sparsely connected."

Potential applications of this work might be wildlife management, for example, where increasing urbanization can lead to unusual interactions between people and wild animal populations. More commercially still, one could imagine using this work as the foundation for wireless computing for publish and subscribe models - where data, such as sports scores, financial data, and other data people need frequent updates on, is pushed around. It might also be applied to wilderness search and rescue, such as in the event of an avalanche.

The DoD is interested in similar systems to support the communication needs of forces fighting in remote areas. One of their problems now, according to Martonosi, is that their architectures often rely on a central static base station, which poses extreme vulnerability, and, you can imagine, potential troop deaths as a result. "Instead," says Martonosi, "soldiers and tanks could be the mobile nodes used to push data around a peer-to-peer network. This would be a big win in terms of resilience of the system to a single point of failure."

One of the values of this project for Martonosi has been the opportunity to adapt ideas from mobile, ad hoc sensornets to concrete applications: "moving past the abstractions common in research," as she puts it. "While previous work in sensornets focused on cases where sensors are static, here we're looking at mobile sensors. Mobility is an extra challenge. We're also looking at systems with very sparse connectivity and networking coverage, which are challenges for energy efficiency and information recovery." Martonosi's group is also designing ways to conduct software updates automatically, rather than communicating with every zebra collar individually, so that they can achieve some level of autonomous operation.

Martonosi, not being a biologist, could imagine extending her system to elephants. "They are particularly exciting because they cover huge distances, making them good data carriers, and there are the obvious benefits of people knowing exactly where the elephants are to avoid danger."

But what about the ivory trade, which shows recent signs of being re-legitimized? What if poachers were to hack the system and gain access to the data? "Luckily, the poaching community is not very high-tech," she replies. "At least, not yet."

When asked about the implications of extending this work to the human population, Martonosi says, "One nice aspect of this project is that we haven't yet had to deal head-on with privacy issues. People clearly worry more than zebras about their privacy. If a similar system were developed for human applications, we would necessarily place much more emphasis on encrypting data and on authenticating senders and recipients during data swaps."

Don't forget Martonosi's seminar: Friday, November 22, 2:00 PM (with reception following) at UCSD in the Applied Physics & Math Building, #4301. The title of her presentation is "High-tech Wildlife: Power-aware Computing, Biocomplexity, and the Princeton ZebraNet Project." This seminar is co-sponsored by Calit², the UCSD Jacobs School of Engineering, the Committee on the Status of Women in Computing Research, and the University of San Diego.

For more information, see Martonosi's [Web page], which includes a recent article on ZebraNet from the Princeton Weekly Bulletin (November 11, 2002).