The prospect of computers moving "out of the box" into the physical world to form intelligent, or "embedded," systems, such as robotic surgery tools, oil refineries and airplane avionics systems, is exciting and certainly makes for great newspaper copy. But experts in the intelligent systems community know that the current state-of-the-art in embedded systems will only take us so far. As intelligent systems proliferate more profoundly into physical reality,

they will be required to adapt rapidly to an almost infinite degree of complexity in the environment. This complexity is already threatening to overwhelm embedded systems technology.

Vanderbilt's Xenofon D. Koutsoukos is helping to whittle this problem down to size.

The assistant professor of electrical engineering and computer science has won the prestigious CAREER Award from the National Science Foundation for his research on next-generation computer design techniques and tools for "smart" systems. Professor Koutsoukos is taking a new approach to solving the problems encountered in making complex computer systems interact autonomously and accurately in real-life situations. He's using formal analysis and verification methods drawn from probabilistic and applied mathematics to model, analyze and design embedded computer systems.

He believes that this approach will ultimately result in more autonomous and robust systems that can manage the uncertainty and variability of the real world.

"Using formal methods, we can develop systems with a higher degree of autonomy and ensure that they will work correctly," Professor Koutsoukos says. "These methods are much more definitive than computer simulations and expensive physical testing that are currently being used in designing embedded systems." The CAREER Award will fund five years of the research. During that time, Professor Koutsoukos and his associates will develop and adapt formal methods and test them on a mobile sensor network consisting of dozens of computer nodes equipped with cameras, microphones and laser sensors to autonomously monitor a floor of a building.

Professor Koutsoukos and his associates have modeled the network components' capabilities and constraints as well as the current and potential interactions among them. This model continuously assesses real-life conditions reported by the nodes and directs the nodes to make specific changes to best respond to the environment. The network is thus able to sense and respond on the fly, in real-time.

"Ultimately the system will be able to meaningfully monitor the area, without human intervention, and could send a robot to respond to different situations," Professor Koutsoukos says.

New Depths

Koutsoukos is also a senior research scientist in the Institute for Software Integrated Systems (ISIS), a research institute at Vanderbilt. He and his colleagues are exploring and developing a profoundly new framework for the full scope of embedded systems software.

"We need new research to obtain a much deeper understanding of the nature of embedded software design, and we need to use this understanding to develop new programming abstractions and new design technologies and tools," he says. "Otherwise, we will not only be unable to take technology to the next level, but we will also be plagued by 'glitches' that can incapacitate whole systems or cause disasters."

To put that into perspective, take the example of the ABS (anti-lock brake system) on a car. The ABS computer can sense when a wheel is locked and electronically pumps the brakes 10-20 times faster than any human foot could do the job, which is a handy feature when you hit an unexpected patch of ice at night, for instance.

But not such a good feature if the ABS computer "crashes." In this and other safety-critical systems, smart software has to get it right the first time, and every time.

Complicating this challenge is the fact that smart devices and components are made by different companies, most of which used different software packages built in-house or purchased from third-party vendors. Many of these systems are being brought together to create multi-faceted systems that must communicate across a variety of platforms and protocols. Doing the Math

Then there's the problem with the math. Physics and computer science parted ways some 30 years ago, each group humming along separately, quite peacefully and respectfully, but with each discipline growing exponentially in complexity.

Enter computer-controlled physical systems, and suddenly you need computers that can deal with the "analog" world of physical reality. Humans have built-in abilities to sense and respond, sorting through a vast amount of sensory information just to walk to a door and turn the knob. So far, no robot has been able to accomplish this feat, which is literally child's play for your average toddler. Translating the real world into mathematical descriptions and then into code that a computer can read and respond to it requires nothing less than a completely new mathematical model and tool infrastructure.

"Many of the abstractions that have been so effective at improving our computational capabilities are either indifferent to or at odds with the requirements of software that interacts with physical processes," Professor Sztipanovits explains.

"That worked fine as long as the system stayed in the box. But physical reality is less tidy and predictable."

Professor Sztipanovits and his colleagues consider tackling the daunting task of re-integrating computing with physics to be truly "one of the grand challenges in computer science today," he says.

Drawing from ESCHER

Professor Sztipanovits is one of the leaders in developing a private, non-profit organization to address this grand challenge and to advance embedded computing. That organization, ESCHER (Embedded Systems Consortium for Hybrid and Embedded Research) http://www.escherinstitute.org/ was created to harness the efforts of government, academia, and industry to facilitate the development of next-generation design technology for embedded systems.

ESCHER was established by an academic/industry/government consortium, spearheaded by Professor Sztipanovits, Vanderbilt Professor of Computer Science Douglas C. Schmidt, and Professor Shankar Sastry, chair of the Electrical Engineering Department at UC Berkley.

"For industry, ESCHER will help define common research needs and act as a clearinghouse for collaborative efforts," Professor Sztipanovits says. "It will establish processes to transfer the research results quickly and efficiently to the industries that will be transformed by these efforts.

"Industry and government will find that ESCHER has a tremendous value in transitioning embedded systems technology to practical use," Professor Sztipanovits says.

On the Defense

ISIS has already put some of these developing technologies to practical use. ISIS researchers have developed a computerized system that can detect the precise location of a sniper in a noisy, crowded city; unmanned flying vehicles that can locate the source of enemy radio waves; software that can navigate fighter planes safely through hostile airspace and cut the task of scheduling air operations and aircraft maintenance from hours to minutes. Assistant Professor of Computer Engineering T. John Koo is working with researchers from UC Berkeley, Southwest Research Institute, and ISIS to explore the feasibility of using lightweight and highly maneuverable Micro Unmanned Aerial Vehicles (UAVs) for intercepting and locating enemy broadcast-communications systems in urban environments.

Equipped with radio receivers designed by Theodore "Ted" Bapty, Research Assistant Professor of Electrical Engineering at Vanderbilt, the UAVs can detect transmissions from broadcast communications systems such as radios, as well as radar and other electronic systems. Intercepting such transmissions can provide information on the type and location of even low-power transmitters, such as hand-held radios.

ISIS has developed a similar distributed network acoustic-localization system to pinpoint the location of a sniper. A team under the direction of Research Assistant Professor of Electrical Engineering Akos Ledeczi developed the technology behind the "shooter localization" application, which uses multiple small networked sensor nodes that can be scattered randomly on the ground, on the fly. The nodes enable the system to locate a shooter's position along with the projectile trajectory, by tracking the muzzle blast and the ballistic shockwave.

"Our system contains a large number of cheap sensors-possibly hundreds or thousands-communicating through a self-assembling wireless network," Ledeczi says. "By using a large number of simultaneous measurements at different locations, we can achieve much better coverage than with just a few sensor units."

Another defense application of ISIS software research is the Coherent Analytical Computing Environment (CACE) system used by Marine Corps to schedule air operations and aircraft maintenance, reducing the time required from a full day to four minutes. The Office of Naval Research recently announced that it will spend $5.74 million to expand the use of CACE to the entire inventory of Marine Corps tactical aircraft, as well as the next-generation Lockheed-Martin Joint Strike Fighter.

"We are very excited about this opportunity that allows the application of our research ideas to a wide variety of aircraft," says Gabor Karsai, Associate Professor of Electrical Engineering and Computer Engineering, who directs the effort at Vanderbilt. "We envision that our technology will also be extended to commercial aviation."

CACE was developed by researchers at ISIS and UC Berkeley, a partnership that Karsai expects to strengthen through similar projects as well as collaboration within ESCHER.

Straight to the Source

Cross-country collaborations such as this one are signs of the times in the computer world, if the "open source" movement is any indication. Open source systems (such as the Linux operating system and Apache web server) and applications make computer code accessible and allow programmers to read, redistribute, and modify source code, without locking them into expensive, proprietary solutions. Unlike closed models of conventional software development, where code is proprietary and inaccessible, open source software evolves through standardization, constant improvement, and adaptation by the worldwide programming and user communities.

Vanderbilt Professor of Computer Science Douglas C. Schmidt has been one of the pioneers in the open source movement since 1987. He currently leads the development of software at ISIS that helps a wide range of computing systems work together more smoothly and efficiently.

Professor Schmidt says that he was motivated early in his career by "the countless hours and dollars companies and agencies spend to write software from scratch for applications, despite the fact that most of the software functions share many common features that were not being exploited.

"What was needed was a core software structure-called middleware because it sits in the middle layers of complex distributed systems-that could be reused over and over again with diverse operating systems, networks, and hardware," he says.

Professor Schmidt and his colleagues around the world have developed two widely used middleware packages: ACE and TAO. With the help of 30 core developers, more than 1,850 contributors, and eight companies that provide commercial support via an open source business model, the ACE and TAO middleware is being used by developers, and end-users all over the world in domains ranging from telecom, medical, aerospace, defense, and financial services "Everything we've done with ACE and TAO and our other R&D projects is available on the Internet at www.dre.vanderbilt.edu," Professor Schmidt says. "People regularly download the software we've built and make their own adjustments. We then integrate the best of their contributions, resulting in better research and products. We work with some of the brightest software developers around the world in every time zone, every hour of the day."