Duke Leads Defense Department-Funded Studies to Detect Hidden Targets With Robotic Sensors
DURHAM, N.C. -- A vision of futuristic robotic aircraft and land vehicles that can sense and close in on targets hidden in trees, caves or bunkers is being explored by a new four-university research initiative led by Duke University's Pratt School of Engineering. The hunt would begin over a wide area, using stationary and moving sensors that might scan for communications signals emanating from a bunker, or the different kinds of electromagnetic signatures put out by machinery, or the infrared waves emitted by a heated object. Another tool would be radar.
Other sensors, perhaps installed aboard airborne or surface vehicles, would autonomously coordinate their activities with minimal intervention from humans. Those would narrow down the search using a special kind of back-tracing mathematics to locate the "fields" in space where the tell-tale waves vibrate.
"The idea of doing multi-sensing on multiple unmanned platforms is new," said Lawrence Carin, a Pratt School professor of electrical and computer engineering who is leading the effort. "It hasn't been done before. Almost everything we have proposed is new."
The technology being developed would allow increasingly localized sensor searches for quarry so hidden that "you don't even know where to start to look without this technology," Carin said in an interview. "This is a very challenging problem. It will constitute a big leap ahead from where things are today."
Researchers at Duke, Georgia Institute of Technology, Stanford University and the University of Michigan will each take on different parts of developing the enabling mathematical underpinnings of this technology with $6 million in Defense Advanced Research Projects Agency (DARPA) funding, which will be administered through the U.S. Army over five years. The award was announced March 20.
The objective, according to the language of the award, is the development of "detection and classification algorithms for multi-modal inverse problems." That means developing mathematical rules -- called algorithms -- to "train" and control multiple sensors that, with increasing precision, could detect invisible signals emanating from such targets, and trace those signals back to their sources -- a technique called inversion.
"The targets could be land mines, targets under trees like tanks or troops, or targets in underground bunkers or caves," said Carin, who is the overall administrator of the DARPA Multi-University Research Initiative (MURI) grant to Duke, Georgia Tech and Stanford. The Michigan work, while funded separately, is being coordinated with the MURI project.
Once potential targets are perceived, the search might be pinpointed using a different mix of sensors. "For instance, if vehicles are moving through trees, you could actually sense the motion," he added. "A hole in the ground will cause perturbation to the gravity that's observed on the surface, so you can detect that." Other sensors, he noted, might likewise register acoustic vibrations.
These arrays would do more than just sense passively. They would also have to infer from these different signals what their sources are through the inversion process. "In the campaign in Afghanistan, they're using a lot of unmanned Predator drone aircraft," Carin said. "Our vision is that in the future you could have multiple drones out there, not just collecting data but actually doing the inversion."
"You would like them to be self-controlled," he added. That means they would not need detailed updating to tell them what to do after being sent out to a certain area. "They would make decisions on their own," he said. "A sensor would have to be able to think."
"You could have multiple drones and multiple land robots that communicate with one another and do the inversion on the fly, because it is too complicated for them to communicate back, and moreover through communications traffic they will reveal themselves," Carin said.
"What would happen if you have several robots? Let's say they are sensing for electromagnetic radiation due to traffic or machinery. Let's assume they sense something interesting. They may then turn on a seismic sensor or gravity sensor. As they learn information, that will guide how to use other sensors."
This MURI project builds on a previous one, also directed by Carin, that coordinated research into land-mine detection by investigators at Duke, Georgia Tech, Stanford, California Institute of Technology and Ohio State University.
That previous work gave the new MURI's team members a considerable track record working with inversions algorithms that both deal with "noisy" field data and "a multiplicity of sensors and applications," Carin noted in his application for DARPA support.
"Our inversion algorithms now represent the state of the art in this field and are deployed in systems being developed for the Army Countermine Office," he wrote. In the process of that research, which still continues through separate grants, "we developed a close working relationship with the Army Night Vision Laboratory and associated contractors," Carin's application continued.
A key researcher from Georgia Tech also involved in the previous MURI project is Waymond Scott, an electrical and computer engineering professor who is a top expert on the effects of vibrations on soils and buried objects. Scott's work will focus on detecting obscured structures, such as buried land mines, with seismic and electromagnetic waves. His approach will combine both experiment and theory.
"Our major emphasis is understanding the physics of how these waves interact so we can design better sensors, improve signal processing algorithms and put some bounds on what will be practical in sensing systems," Scott said. "We will also have a strong coupling between our experimental work and our numerical modeling efforts. Having better models will help us understand the complexities of the sensors."
He noted that detection of buried land mines and underground structures such as tunnels or bunkers is made more difficult because soil, unlike air or water, is not a homogeneous material. Layers of different soil types, and objects such as rocks and sticks, affect the transmissions of waves. So complicated analysis techniques are needed to pick out the targets.
In past work, Scott and his Georgia Tech collaborators have been successful in detecting unique resonances created by land mines and the complicated mechanical structures inside them.
Stanford mathematician George Papanicolau will be working on "reverse time migration," a method to both measure invisible waves, forces or vibrations and, in effect, "reverse them in time" so they lead back to their sources.
Alfred Hero, a professor of electrical engineering and computer science, will lead development of new statistical signal processing and classification algorithms at the University of Michigan.
At Duke, Leslie Collins, an assistant professor of electrical and computer engineering who has worked closely with Carin on the mathematical and statistical underpinnings for Duke's land-mine detection advances, also will collaborate in the new MURI project. So will Qing Liu, an associate professor of electrical and computer engineering.
The MURI program also will feature a three-week workshop each summer at Stanford where "mathematicians from around the country and perhaps around the world will join us to try to tackle some of the most difficult problems we've come across," Carin said.