October 15, 2002
NASHVILLE, Tenn. – Today, it seems as if all the devices we use are becoming “smart.”
Smart is the popular term for a device with a computer chip embedded in it that allows it do things that were impossible a decade or two ago: From toasters that automatically sense when the toast is done, to cellular phones that can recognize voices, to refrigerators that keep track of the types and amounts of found that they hold, smart devices are all the rage.
Yet a number of challenges remain before computers can achieve their full potential for controlling mechanical systems. “This is where the world of programming code hits the laws of physics,” says Shankar Sastry, professor and chair of UC Berkeley’s Department of Electrical Engineering and Computer Sciences. He heads up a new $13 million, five-year research effort funded by the National Science Foundation that is designed to attack these problems. The multi-institutional effort includes researchers from the School of Engineering at Vanderbilt University and the University of Memphis.
The program will be addressing two fundamental problems: the need for new mathematical techniques to model and analyze the complex interactions between computers and physical systems and the development of computer hardware and software that are cost effective yet have the extremely high reliability required for operations where human lives are at risk, such as operating automobiles and aircraft.
According to the researchers, current methods of software engineering are not adequately meeting the needs of today’s embedded software systems, “The field of computer science has become increasingly idealized and specialized, and the field of engineering has moved away from the fundamentals of programming,” says Sastry, who founded and directs UC Berkeley’s Center for Hybrid and Embedded Software Systems (CHESS). As a result, many of the advances in software engineering are not making it into embedded software systems because they do not accommodate physical constraints.
“Computing has a rapidly increasing role as a vital component of physical systems around us, such as cars, airplanes, mobile phones. In fact, these embedded computing systems represent over 98 percent of all computer applications,” says Janos Sztipanovits, professor of electrical engineering and computer science at Vanderbilt and principal investigator of the campus portion of the project, which totals $4.1 million.
“The so-called ‘drive-by-wire’ steering system that will replace the automobile’s steering column with digital controls is one example of the type of embedded system we can expect in the near future,” adds Gabor Karsai, Vanderbilt associate professor of electrical engineering and computer science, who is also a key participant. “We already have ‘drive-by-wire’ systems in experimental cars.”
According to Karsai, the problem with embedded systems is that the computerized equipment must function adequately in real time, facing physical constraints and meeting physical demands. “Physical processes are highly complex, and it is not obvious how they mesh with computational processes,” Karsai points out.
In order to make computers practical partners in controlling physical reality, software has to be as reliable as human operators. When it’s time to hit the brakes, the computer does not have time to freeze and reboot.
Previous efforts to make embedded systems reliable have depended on building in redundancy and adding mechanical back-up systems. Spacecraft and jet aircraft, for example, typically use three independent computers that “vote” on every action. In that way, if one processor begins to malfunction, it is over-ridden by the other two. But this approach doesn’t deliver either cost or productivity savings.
A major component of the new engineering approach is the Vanderbilt-developed Model-Integrated Computing, pioneered by Sztipanovits, director of the Institute for Software Integrated Systems. Model-Integrated Computing is based on the systematic use of models in the analysis and design of software programs and the automatic generation of computer code to implement operations and ensure reliability.
The project also includes a strong educational component that promises to reap long-term rewards in better-trained engineers and programmers. The NSF award will support the creation of a Summer Internship Program in Hybrid and Embedded Software Research (SIPHER), headed by Sztipanovits. “We will experiment with new forms of student involvement in research and training via the program, which will provide funding for students of underrepresented groups and their teachers to interact with researchers at Berkeley and Vanderbilt.”
The award will also support significant revisions to the undergraduate curriculum at participating institutions by introducing crossover courses for majors in computer sciences and electrical engineering. The researchers expect to include new elective courses relating to embedded systems design over the next three to four years.
In the near term, the project will focus on the development of reusable open-source software that supports embedded systems. Future applications could include anti-terrorism technologies, aircraft and vehicle electronics and autonomous robots. Other applications that stand to benefit are use of sensor networks to monitor an office building’s structural integrity or a home’s energy consumption.
Contact: David Salisbury, 615-322-NEWS, david.salisbury@vanderbilt.edu