By J.R. Wilson
The U.S. Defense Advanced Research Projects Agency (DARPA) in Arlington, Va., is celebrating its 50th anniversary, and several reviews are being written of its contributions to various areas of military and civilian science and technology. It is a history that includes the creation of numerous technologies, many developed by DARPA decades before they became 21st century standards, such as stealth, global positioning, and navigation satellites, and nanotechnology.
Created by a U.S. Department of Defense (DOD) directive on 7 Feb. 1958, DARPA initially was chartered in response to the Soviet launch of Sputnik and directed to get the U.S. moving in space-a task largely transferred to NASA (which came into being a few months later), as well as to the Air Force Army before the end of the Fifties.
A second primary task was to find ways to detect nuclear tests in the atmosphere, space, underground, and underwater, which DARPA experts pursued with sufficient success that a series of test ban treaties could be signed and monitored. Eventually, in what has become a DARPA tradition, success led to the effort being transferred elsewhere for full implementation.
The overriding directive for DARPA, however, was common to those tasks and all that followed-to prevent the U.S. from ever again being surprised by someone else’s technology. That quickly expanded to include becoming the creator of technological surprise for others.
To accomplish this, DARPA was given a key capability unique to government-or civilian-agencies: It was allowed to fail, providing the payoff for success was high enough.
Without being able to fail and not face funding cuts or worse, DARPA could not have taken on what have become known as “DARPA-hard” challenges, the vast majority requiring significant advances is computing capability. As a result, the agency became a world leader in advancing the technology of smart computers-sometimes referred to as artificial intelligence (AI)-especially in the area of information and signal processing.
DARPA’s role in furthering computer and related technologies has been direct and indirect, sometimes resulting from a requirement contained within a program in which that technology plays a supporting role, such as charged coupled devices to support NASA satellites or working with the Department of Energy on biological microchips as part of an integrated system to automatically detect, acquire and process complex environmental samples.
From its creation through the late 1980s, DARPA was a powerful force in the development of microelectronics, computing, and network communications. As commercial use of and demand for those and related technologies became a multibillion-dollar global enterprise, the U.S. military became more and more reliant on commercial off-the-shelf technology. Even so, with its free hand in pursuing ideas too revolutionary for almost any corporate or government lab to justify, DARPA has retained an ability to significantly influence the directions of the broader market, thus assuring DOD access to and the ability to use new technology.
“DARPA is in and out of programs over the years; we fund something until we feel we have done all we can, then we go back to basic components until we feel we’ve done enough to actually put something together to demonstrate it in a system,” says DARPA director Dr. Tony Tether. “When somebody says something is a great idea-but it can’t be done-that makes it a challenge for us here at DARPA. Our job is to show something can be done and kick it out so we can go off and do something else people don’t think can be done.”
Sometimes this can include things DARPA itself may have determined can’t be done-at the time it was attempted, at least-which emphasizes the importance of yet another DARPA-unique factor: lack of corporate memory.
“You can’t kill good ideas, although you can stop them for a while. But they come back, especially at a place like DARPA,” Tether adds. “We rotate our people through here in no more than six years, which is great because that means 10 years from now someone can come in with an idea we tried before that didn’t work, but there won’t be anyone around to say that. That’s what makes us different from anyplace else.”
DARPA has always relied on short-term (usually two to four years) program managers brought in from throughout government, industry, and academia, as well as cooperative efforts with other government and private labs and universities. AI was a prime topic for such cooperation, often leading into areas not specifically pursued by the agency in its early years, such as robotics, but nonetheless supported in academic endeavors by its Information Processing Technologies Office (IPTO). One of the early examples of this was Project MAC (Machine Aided Cognition-later renamed the Laboratory for Computer Science) at MIT and its director from 1974 until his death in 2001, Dr. Michael L. Dertouzos.
In his book, “What Will Be” (Harper Collins, 1997), Dertouzos described DARPA’s extensive role in creating and expanding the Computer Age as spectacular, accounting for “somewhere between a third and a half of the major innovations in computer science and technology. These include time sharing, computer networks, programming languages like Lisp, operating systems like Multics (which led to Unix), virtual memory, computer security systems, parallel computer systems, distributed computer systems, computers that understand human speech, vision systems, and artificial intelligence, an endeavor responsible for understanding and emulating human intelligence by machine.”
From Reduced Instruction Set Computing (RISC) and Redundant Array of Inexpensive Devices (RAID) to Massively Parallel Systems that set the stage for industry-developed high-end computers, DARPA’s hardware efforts took the world from a handful of building-sized, extremely expensive, environmentally super-sensitive and ultimately limited capability computers in the early 1960s to a new millennium in which high-speed microchips and connectivity are pervasive throughout military and civilian life.
The Strategic Computing Initiative (SCI), created in the 1980s to fund all DARPA computer work, was cited as an essential agent in the evolution of AI, robotics, advanced computing, and autonomous capabilities by the then-director of DARPA’s Information Technology Office (ITO), said Dr. S. Shankar Sastry, at DARPATech 2000.
“(SCI) produced its share of disruptive technology in developing reduced instruction set processors, specialized graphics engines, RAID disks, robotics, and AI tools which are now currently mainstream. The investments by DARPA in these technologies in their early and formative years have paid rich dividends,” he told the agency’s biennial gathering of program managers and industry. “More importantly, even in the face of the most recent worldwide surge in the commoditization of IT, it is important to note that DARPA investments have maintained superiority in national security needs.”
By 1985, much of DARPA’s experience and research in this arena was applied to a new Tactical Technology Office (TTO) Smart Weapons Program, with a mandate to use advanced computer hardware and software to develop a modular family of smart, autonomous weapons that would make AI a battlefield reality.
That effort also highlighted the importance DARPA placed on advancing software as well as hardware. From the Ada and High-Performance Fortran programming languages to Unix and Mach operating systems to microkernel operating system technology to wavelet-based algorithms for the Longbow Fire Control Radar System’s automatic target recognition, DARPA has pushed the capabilities envelope and opened the door to further software revolutions by industry.
DARPA research into Very Large-Scale Integration, begun in the late 1970s, is widely credited with influencing VLSI design, expanding its availability and ultimately leading to new innovations in computer-aided design, integrated circuit fabrication and low-cost access to submicron complementary metal oxide semiconductor (CMOS) technology.
In short, much of what became taken for granted in the ever-quickening pace of the computer revolution of the 1990s and today can be traced to DARPA research dating back half a century into concepts and technologies that, at the time, seemed at best far-fetched or mere oddities.
For example, Pilot’s Associate, begun in 1986 in cooperation with the Air Force, combined DARPA’s experience with advanced computer systems and AI to enable the pilot of an advanced single-seat jet fighter, such as the F-22, to receive integrated, prioritized information for rapid decision-making, based on a flood of data from several sensors and sources. The results of that effort can be seen in the F-22 and F-35 avionics and missions systems suites.
Another application of smart computing can be seen in DARPA’s research into robotic autonavigation, a prominent example being the Grand Challenge and Urban Challenge competitions DARPA sponsored in recent years. The first offered a $2 million prize to the first team to build a fully automated vehicle that could cross 132 miles of California desert, with no human interaction, in less than 10 hours. The second, in the Fall of 2007, raised the ante to $2 million for the first vehicle, $1 million for the second, and $500,000 for the third to complete a 60-mile urban environment course, including other traffic, without violating any California traffic laws, in less than 6 hours. In both cases, several teams met the challenge.
It was an unusually public effort by an agency that typically shuns the spotlight, but it also met a number of traditional DARPA goals: Both challenges advanced the application of computing and AI to autonomous vehicle operations, intimately involved industry and academia, encouraged the interest of students and young engineers, and demonstrated the potential of autonomous robotic vehicles to the military. As it had with UAVs, even when the military showed little or no interest, DARPA managed to bring another form of smart robot to a level of prominence that is likely to change the face of future battlefields-and, as a typical side effect, civilian vehicles as well.
Even as decades of R&D are now putting smart weapons, UAVs and ground robots into the battlespace mix, DARPA is moving forward with the more complex application of AI to walking robots. While C3P0 and the legged mechanical giants of Star Wars are still a long way from reality, DARPA’s ability to not only use the continuing evolution in computers and other technologies but foment technological revolution makes the advent of a walking robot on the battlefield-and elsewhere-far more likely far sooner than science fiction has predicted.
“Right now, all robots are programmed and I think we are running into the limits of programming languages and the flexibility we can program into these systems. Being able to have robots learn, whether autonavigation or better precision or fuel efficiency, and break the reliance on human programming is a new frontier,” notes Eric Krotkov, president of Griffin Technologies and a former TTO robotics program manager.
“I believe we have the computers and hardware we need to do that now; what is missing is the learning methodology to take a bunch of examples and turn it into a program that runs on the robot. At Urban Challenge, most teams were changing their codes right up to the last minute, with human programmers trying to cut down on bugs. If we can eliminate that and have the robot do that itself, it would be a big win.”
DARPA also has taken the concept of AI and autonav to space with its successful Orbital Express Space Operations Architecture program in 2007 to validate the technical feasibility of robotic, autonomous on-orbit refueling and upgrading of satellites. Two satellites in low Earth orbit performed a complex series of rendezvouses and matings, exchanging fuel and components, with no human intervention.
At DARPATech 2007, IPTO director Dr. Charlie Holland described DARPA’s past, present, and future contributions to the broad field of information technology and the provision of decision dominance on the battlefield through computerized information processing.
“IPTO helped invent large parts of the computing field in the usual DARPA way. We took risks, betting on packet switching when the rest of the world used circuits. We expanded traditional notions of information processing to include human-computer interaction, robotics and more,” he said. “What is needed [now] are computing systems that learn and reason, can structure massive amounts of raw data into useful, organized knowledge with a minimum of human assistance and adapt to new situations.
“DARPA has played a major role in developing specialized hardware, multiple processors, parallel algorithms, and other techniques to improve computation efficiency. But we now recognize that increases in hardware speed are not enough to break the barriers computing faces. There is a huge need for improvements in software development. We need at least an order of magnitude improvement in our ability to quickly develop efficient, problem-solving software and code for flexible, adaptable applications. We also need new run-time systems that execute code an order of magnitude more efficiently on the heterogeneous hardware that is becoming increasingly common in computers and microprocessors.”
That DARPA will continue to pursue those new demands can be seen in programs already underway or in the proposal stage. Whether it can match or exceed its first 50 years in the decades to come remains to be seen, but its unique position as an agency in which failure is accepted as a requirement to breaking through boundaries, where today’s work may not see practical application for decades, where there is no “corporate memory” to prevent a new generation from tackling problems their predecessors could not resolve has proven a strong foundation for creating much of the technology the world now takes for granted.