Sunday, June 22, 2008

Grant Supports "Intelligent Binoculars" that Mimic the Human Brain

The Georgia Institute of Technology has received a contract to help develop “intelligent binoculars” that mimic the low-level image processing done by the human brain. Called the Cognitive Technology Threat Warning System (CT2WS), the device is expected to be far more capable than portable visual threat-warning equipment currently used by the U.S. military.

A research team headed by Paul Hasler, an associate professor in the Georgia Tech School of Electrical and Computer Engineering (ECE), will investigate the use of neuromorphic engineering to enable a CT2WS device. Neuromorphic techniques use innovative hardware and software approaches to emulate human intelligence.

The Georgia Tech team will be working with Hughes Research Laboratory and Northrop Grumman Corp., which have won CT2WS contracts from the Defense Advanced Research Projects Agency (DARPA).

“Realistically, this is probably not something that’s going to be one project and then it’s done,” said Hasler, who is a research team leader at the Georgia Electronic Design Center (GEDC), a 200-researcher group at Georgia Tech focused on mixed-signal (analog-digital) microelectronics. “This opens up a whole avenue of thought about neural-inspired approaches to new applications. There are likely to be a tremendous number of opportunities here going forward.”

Neuromorphic engineering is interdisciplinary, using fields that include biology, physics, mathematics and computer science, as well as electrical and other types of engineering. Its aim is to develop artificial systems—such as vision devices or auditory processors or even robotic systems—based to a degree on natural biological systems.

The CT2WS program’s aim is to deliver several prototype units small enough to be carried by solders in the field. DARPA has stated that these intelligent visual devices are now possible because of recent discoveries in a variety of fields, including wide-angle optics, digital imaging, cognitive visual-processing algorithms, neurally-based target detection processing and ultra low power analog-digital signal processing electronics.

A successful visual-warning device of this type could produce revolutionary capabilities for the combat soldier.

“The idea of this project is to build a visual device that is attentive, that can do the kind of low-level visual processing that your eyes do naturally,” Hasler explained. “You would see a certain picture in your field of view, but the device would actually be looking over a much wider space—and if it found something interesting it would present you with that picture as well.”

Hasler specializes in designing analog microelectronic circuits, in particular, novel analog designs that possess some of the configurability of their digital counterparts and yet use far less power.

Working with Associate Professor David Anderson, also in the Georgia Tech School of Electrical and Computer Engineering, Hasler has built mixed-signal analog-digital circuits that can match the performance of all-digital designs while using hundreds of times less power and producing less heat.

Hasler has already performed research relevant to the design of intelligent visual devices. With $1.2 million in funding, he and Anderson worked on an earlier DARPA program that focused on new approaches to integrating the sensing and processing of images. And when Hasler won the Office of Naval Research Young Investigator Award, he used that funding to examine frameworks for building neural circuits.

Hasler explains that power issues are vital to a project like CT2WS, which must be portable and yet have enough processing power to emulate human cognition.

“This is a highly interesting platform problem because you really are trying to build, at least initially, a simple representation of one layer of human brain cortex, maybe multiple layers if you can,” Hasler said. “You need a great deal of computational power, robustness and performance, and it becomes possible because a configurable analog-digital system has a power efficiency improvement of up to 10,000 times compared to an all-digital system.”

Hasler adds that the new DARPA program can also be expected to stimulate speculation about whether this technology has commercial promise.

“This kind of problem can serve as a roadmap for future applications,” Hasler said. “It can get people from the commercial side interested because it demonstrates some very real possibilities.”

No comments: