On Maine’s Great Duck Island, scientists studying small sea birds called Leach’s storm petrels are using a network of tiny wireless sensors embedded in the birds’ nests to gather information.
The sensors, or “motes,” are used to monitor environmental conditions around the nesting burrows. One day, they could form the basis for intelligent wireless networks capable of harvesting a wide range of information from their surroundings. Applications could be as diverse as agricultural management, earthquake monitoring and military operations.
A group of researchers at an Intel Corp.-funded lab at the University of California, Berkeley, is investigating that possibility and many others. “We are trying to prototype a future in which there are many, many small sensor devices that are able to get a very fine physical sensing of the real world,” says professor Joseph Hellerstein, director of the Intel lab at Berkeley.
Intel Research Berkeley is one of four university “lablets” that the chip maker set up to identify technologies worthy of “acceleration and amplification,” says Kevin Teixeria, a spokesman for Intel Research.
The other three labs are located at Carnegie Mellon University in Pittsburgh, the University of Washington in Seattle and the University of Cambridge in England.
Each Intel lablet has a specific focus. At Berkeley, the focus is on what Intel calls “extremely networked systems,” and the lab is developing operating systems and programming tools for wireless sensor networks of the kind being tested on Great Duck Island. The laboratory has already developed an operating system called TinyOS and a query-processing technology called TinyDB, which is designed to simplify the task of data collection.
Berkeley researchers are now building a suite of tools called the Tiny Application Sensor Kit (TASK), which Hellerstein says will make it easy for even nontechnical users to deploy applications using sensor networks.
TASK components include TinyDB and several client-side tools for helping users set data collection, filtering and monitoring rules.
TinyOS runs on the sensor motes, with TinyDB acting as a sort of proxy gateway to the sensor network. TinyDB provides a SQL-like interface through which users specify the data they want extracted from the sensor network and how often they want the data.
TinyDB collects raw information from the sensors and then filters and aggregates it and sends it to a server or PC for further processing. The sensor network technology is also being tested at various other locations, including on an oil rig off the coast of northern Scotland, where it is being used to study vibration of onboard equipment.
The focus of the Cambridge lab is on highly distributed applications. Derek McAuley, the lab’s director, likes to describe his work as “turning-over-the-rocks” research. “You never know what you’re going to find underneath,” he says.
For instance, one Cambridge research team is investigating a “virtual-channel processing” technology called Xen that will enable one system to support multiple operating systems and users more efficiently than current software-based virtualization approaches can, McAuley says. Another project involves the use of optical components for connecting internal parts in future computer systems.
Such photonics-based interconnects could offer far greater scalability and bandwidth than current Peripheral Component Interconnect-based technologies, McAuley says. “If you look at the bandwidth that’s going to be required for applications 10 years from now, going optical seems like a good idea,” he says.
Intel’s lab at the University of Washington is developing what researchers call the System for Human Activity Recognition and Prediction, or SHARP, which is designed to predict human activity by observing the objects a person touches and the context in which they are used. The research uses radio frequency identification (RFID) technologies and data mining software to gather physical information and infer human activity from it.
To test the concept, the university has developed an RFID-enabled glove called iGlove that reads information from objects that are embedded with RFID tags. Scientists hope to be able to accurately infer what an iGlove wearer is doing if they know what objects he touches and the order in which he touches them. That kind of technology might be used by elder-care professionals to monitor the activities of older adults in their homes to infer the state of their health.
The Carnegie Mellon Intel lablet is investigating software for widely distributed storage systems. Researchers working with Seagate Technology LLC are trying to enable interactive searching of terabyte-size collections of nonindexed data.
As part of that effort, researchers are studying how to speed up searches and make them more accurate by embedding processors either close to or on storage devices so they can examine and discard irrelevant data close to the source.
Although Intel funds the labs, it doesn’t own the intellectual property, and the research is widely shared and published, Teixeria says. Intel won’t disclose how much it’s spending on its university research projects, but its overall R&D budget is expected to exceed $5 billion this year.
The real goal, Teixeria says, is to see if the labs can unearth something that Intel might then be able to take in-house and develop further. “It’s this notion of both helping to grow the technology and seeing where there is a usage for it within Intel,” he says.