Autonomous technology cometh
The advent of machines that act and decide will benefit from the lessons humanity has learned from itself.
By J.R. Wilson
Autonomy. Until the 21st century, it almost always referred to humans, either through individual freedoms or regional/national self-governance, sovereignty, independence, and freedom.
Today, however, autonomy is more likely to raise images of robots, self-driving vehicles, unmanned aerial vehicles (UAVs), and, more recently, their unmanned ground (UGV), water surface (USV), underwater (UUV), and even space counterparts. For DARPA, autonomy is a context-specific descriptor that can vary by domain, according to Timothy Chung, a program manager in the agency’s Tactical Technology Office (TTO), where he focuses on autonomous systems.
“For terrestrial robot systems, working in structured environments, even dynamic ones like streets, we’re starting to see a maturation, because we have the compute power, different sensing modalities, GPS, and a lot of data to help inform the autonomous behaviors. That also applies to the air, which, despite being 3-D, has fewer obstacles,” he said. With the ocean, there are environmental challenges, whether surface or subsurface, for getting air- or terrestrial-grade autonomy into those environments.
“Depending on how you define it, autonomy already is out there,” Chung continued. “Think about adaptive cruise control in cars or self-parking vehicles. A smartphone might be a conduit for some autonomy to exist, such as doing machine-learning off-board, then using it for facial recognition or augmented reality, for example, where you might interact with digital characters in a virtual world.”
Building on early and dramatic successes of UAV technologies in Operation Desert Storm and their deployment as essential elements of U.S. and coalition operations in Iraq and Afghanistan a decade later, the agency has continued to push UAV technologies forward. At the same time, the commercial and industrial R&D sectors have been building up their own momentum in the UAV space to generate the current worldwide explosion of personal and commercial UAVs.
Pervasive as unmanned systems have become, however, the vast majority of them remain connected to a human being with a joystick, even if that human is safely operating the system from continents away. This connection of one unmanned system to one (and usually more than one) human operator places constraints upon the system, limiting the effectiveness of unmanned systems and increasing operational costs. “The No. 1 manning problem in our Air Force is manning our unmanned platforms,” Gen. Philip Breedlove said in November 2011, when he was leading the Air Force’s operations, plans, and requirements directorate. The Air Force estimated at the time that each operational MQ-1 Predator required a crew of about 168 personnel, the MQ-9 Reaper approximately 180, and the RQ-4 Global Hawk some 300 people.
One of DARPA’s newest programs – the Collaborative Operations in Denied Environment (CODE) program – is designed to extend the capability of the U.S. military’s existing UAVs to conduct long-distance engagements of highly mobile ground and maritime targets in contested, denied, and constantly changing battlespaces. If the program proves successful, the resulting collaborative autonomy technology could enable groups of unmanned aircraft to work together under a single person’s supervision rather than through the current system of continuous control by a pilot and sensor operator, each supported by numerous analysts. UAS could navigate to their destinations, find, track, and identify targets, and then present recommendations for coordinated actions to a mission supervisor, who would approve or disapprove such team actions and direct any mission changes from hundreds or even thousands of miles away.
Meanwhile, DARPA’s OFFensive Swarm-Enabled Tactics (OFFSET) program focuses on the future ability of small infantry units to deploy swarms of 250 or more mini- or micro-UAVs and/or UGVs to help accomplish diverse missions in complex urban environments. Contracts to deliver on the pro- gram’s Phase 1 challenges – to design, develop, and demonstrate a swarm system architecture to advance the innovation, interaction, and integration of novel swarm tactics – were awarded in February 2018 to Raytheon BBN and Northrop Grumman Mission Systems.
DARPA also is moving forward with a larger portfolio of autonomy projects and programs:
- The Assured Autonomy Program, a new research effort that builds on recent breakthroughs in autonomous cyber systems and formal methods (reliable and telling mathematical models of complex systems), aims to advance how computing systems can learn and evolve to better manage variations in the environment and enhance the predictability of autonomous systems, such as driverless vehicles and UAVs.
- The Gremlins program aims to build UAVs that can carry 60-pound payloads up to 300 nautical miles and be launched from and recovered by fighter aircraft, bombers, or C-130 manned transports. The roles of these versatile UAVs could include intelligence, surveillance, and reconnaissance, signals intelligence, and electronic warfare.
- The Fast Lightweight Autonomy (FLA) program, now in Phase 2, aims to explore nontraditional machine-vision and autonomy methods that could empower a new class of algorithms for high-speed navigation in cluttered environments such as streets or inside buildings. FLA aims to develop and demonstrate the capability for autonomous UAVs small enough to fit through windows and fly at speeds up to 45 mph without human control or GPS guidance.
- The Experimental Spaceplane program, with Boeing, aims to build and fly an entirely new class of fully reusable hypersonic craft (Mach 5 or higher) that can be launched into low Earth orbit 10 times in 10 days; reduce launch costs by 90 percent; and deploy a satellite of up to 3,000 pounds. The goal of the program is to provide the nation with an unprecedented ability to quickly recover from a catastrophic loss of critical military or commercial satellites. Automated flight-termination and other technologies for autonomous flight and operations, including some developed by DARPA’s Airborne Launch Assist Space Access (ALASA) program, are part of the program. The first test flights are scheduled for 2020.
“In essence, we’re still conducting applications that people can or currently do, and seeing how autonomous systems can supplant the human actors to save lives, mitigate risk, or increase efficiency. There also may be breakthroughs enabling wholly new capabilities, perhaps some not yet envisioned, that humans do not currently do,” Chung explained. “Whether subjective or not, we have a baseline of human performance, which tends to be more forgiving for humans than it is for machines. We hold our autonomous systems to a higher bar of performance and, as a result, we need to understand the type of reliability, predictability, and trustworthiness they can have. These are the same traits we would want from a human teammate.”
The technical tasks for achieving this kind of autonomy, Chung said, involve advances in human-robot interfaces, techniques of coding the autonomy in the software, and improving reliability throughout the system, from its chips to its mobility and other performance parameters.
“On a factory floor, we can pretty much trust what robots and automated machinery do,” Chung continued. “This is not the case for something like unpiloted transport of passengers or deliveries, where we need software methods to determine and certify that the system is reliable. Part of the solution will reside in software tools for realistic simulation studies of autonomy.
Said Chung: “Those simulation tools will dramatically change how we feel about those autonomous systems in use, either alongside or remote from humans going forward.”
Historically, DARPA often has conducted research into technologies that neither are understood nor yet in demand, revisiting those same technology categories periodically as new program managers come aboard with fresh viewpoints and new developments to help move a “fringe” technology closer to being useful. Although the agency continues to pursue such research, the speed and global extent with which technology research and deployment is advancing has spurred DARPA to redouble its own signature knack for pushing the envelope without being deterred by short-term setbacks.
The agency’s most coveted sign of success is when it transitions the capabilities its R&D projects support into downstream development, testing, and operational use pathways.
On Jan. 30, 2018, for example, DARPA officially concluded about three years of collaborative development with the Navy by transitioning the Sea Hunter, a demonstration vessel that emerged from its Anti-Submarine Warfare (ASW) Continuous Trail Unmanned Vessel (ACTUV) program, to the Office of Naval Research (ONR). ONR now is continuing development of the revolutionary prototype vehicle, the first of what ultimately could become an entirely new class of ocean-going vessel – the Medium Displacement Unmanned Surface Vehicle (MDUSV) – that traverses thousands of miles of open seas for months at a time, without a single human crewmember aboard.
“ACTUV’s move from DARPA to ONR marks a significant milestone in developing large-scale USV technology and autonomy capabilities,” Alexander Walan, a TTO program manager, said at the time in an agency release. “Our collaboration with ONR has brought closer to reality a future fleet in which both manned warships and capable large unmanned vessels complement each other to accomplish diverse, evolving missions.”
“ACTUV represents a new vision of naval surface warfare that trades small numbers of very capable, high-value assets for large numbers of commoditized, simpler platforms that are more capable in the aggregate,” added TTO Director Fred Kennedy. “The U.S. military has talked about the strategic importance of replacing ‘king’ and ‘queen’ pieces on the maritime chessboard with lots of ‘pawns,’ and ACTUV is a first step toward doing exactly that.”
“Sea Hunter serves as a promising test bed for a new kind of mentality in the Navy about what it would take to have an autonomous, long-duration vehicle. But we think commercial fleets also can benefit from those same capabilities,” Chung said. “We see the law enforcement community adopt bomb disposal robots that came out of DARPA for the military well over a decade ago.”
Most likely far more consequential will be the self-driving car technologies that evolved from DARPA’s Grand and Urban Challenges, Chung said, adding that “cellphone assistants, such as Siri on the iPhone, evolved from PAL [Personal Assistant that Learns], a language assistant program that came out of DARPA roughly four years before Siri was launched.”
Autonomy manifests itself in many ways and holds promise as an enabling framework for multiple DARPA projects. For example, robotic arms developed to service and repair orbital platforms in space also can be used for such earthbound roles as inspecting oil and gas pipelines.
In airborne settings, having a broad understanding of how autonomous UAVs can expand surveillance, communications, and other facets of the battlespace opens the way to better interactions between human pilots and UAVs, perhaps leading to UAVs acting as a combat pilot’s “wingman.” That would reduce the number of manned aircraft needed on some missions and thereby remove at least some human pilots from harm’s way.
Ushering autonomous UAVs into the battlespace is “not just a matter of self-reporting to the human pilots what it is doing, like a sophisticated transponder, but being able to make decision recommendations, alleviating the [pilot’s or military air traffic controller’s] responsibilities, leaving him to be a tactician or battlespace commander more effectively. That touches on a lot of what DARPA is doing,” Chung said. “On the ground, you see … Ground X-Vehicle Technologies, which are more about crew augmentation, such as instrumenting a windowless ground vehicle so the crew inside will have a better situational awareness of their environment.
At sea, being able to maneuver, especially underwater, remains an active area for the development of autonomy options. The Cross-Domain Maritime Surveillance and Targeting [CDMaST]) program, for example, is trying to better monitor and protect the waters, and part of the vision includes autonomous systems. These could provide cueing to other systems – manned or unmanned – to defend and protect those areas. So a distributed network of sensors and autonomous vehicles could help shape the waterscape for maritime warfare.
DARPA’s OFFSET program is all about leveraging tactical opportunities that could come with coordinating swarms of air and ground robots – up to 250 of them – in urban areas. Missions could include aiding a small unit of warfighters charged with securing a bridge, or more crucially, aiding in the historically challenging and costly task of clearing a multi-floor building of the enemy. Important to this technology vision is understanding where such swarms excel and where they don’t at all.
“The complexity, command, and design space of 250 physically disjointed robots, some air, some ground, with different speeds and capabilities, requires having to break people’s familiar and ingrained mentality of how we interact with smaller numbers of vehicles,” Chung said. “Think of the number of different tactics we could employ with larger numbers of vehicles. Those 250-unit swarms are not that far away.”
The subterranean domain, perhaps the most unsung among all of the warfighting domains, was important in both Vietnam and Iraq. In the spirit of previous DARPA challenges, the agency has issued a Subterranean (SubT) Challenge to both accelerate an entire technology category and catalyze the formation of a research community to do just that. The capabilities the SubT Challenge calls on participants to deliver include rapidly navigating, searching, and mapping underground environments. In addition to its military applications, the potential payoffs of such technology range from finding trapped miners in a mine collapse to diagnosing specific problems in damaged storm drain systems and utility tunnels to mapping out natural caves.
“Caves are unpredictable and unstructured, but provide a natural test bed for assessing how well the systems we hope the participants deliver meet the various capabilities at the heart of the Challenge,” Chung said. “We find, time and again, that the geology and configuration of rock formations, and the location of water tables, can change only a short distance away as the environment changes. We want to inspire competitors to address all three domains [human-made tunnels, underground urban spaces, and natural cave networks] and come up with system solutions – a federated solution of platforms – that can work across unpredictable environments.”
Organizers of the SubT Challenge hope the effort will deliver new and powerful technology options in a wide range of arenas, among them mine safety, infrastructure inspections, GPS-free navigation, and communications – often with jamming and out of line of sight. “Many of our systems, even UAVs, tend to be thought of in a two-dimensional environment, where subterranean contexts may have an inclined shaft or a deep, dark hole,” said Chung. “Competitors in this Challenge will have to think through how to deal with such encounters.”
It will be years before any seeds of technology that emerge from the SubT Challenge can grow into actual new features of the technosphere. Perhaps the autonomous technology most frequently in the news these days is self-driving cars – both their amazing feats on the road and their accidents, some of them tragic. But while this mix of success with some failure has raised questions about whether the technology is sufficiently sophisticated and reliable to be unleashed on millions of public streets and roads, the technology’s track record has convinced Department of Defense (DOD) decision-makers that it’s ready for primetime in military applications.
“The questions raised are relevant; these are issues DOD has and will continue to take seriously,” Chung said, adding that the DOD has issued directive 3000.09, providing guidance regarding what decisions autonomous vehicles can and cannot make, including ones regarding the use of potentially lethal onboard weapons.
“We have to hold our systems to a high level of scrutiny because just as with driverless cars, lives will be on the line. The same type of reliability we care about in the military is of interest to the commercial market,” he added. There also is a confluence of interest between the military and commercial sectors when it comes to interoperability and compatibility. “Both industry and government can step up to help that in the future,” Chung said. “The commercial sector has to rely on market demand, but the stars are aligning between the interests of DOD and the needs for greater autonomy and robotic systems on the commercial side, which is a great development.”
At the heart of DARPA’s autonomy portfolio are various levels of artificial intelligence (AI) – from the level of basic machine learning to that of adaptive systems to a level that is “cognizant” and can continuously learn from experience and from data it acquires on the fly. A current theme of the overall DARPA research portfolio centers on moving up the AI levels and applying these ever more capable AI systems to many technologies, including autonomy.
“Machine learning is still relevant,” Chung noted, offering as an example the need of a threat detection system to tell the difference between a person holding a broom and an AK-47. “Adaptive AI is closing the loop around decisions associated with basic machine learning to outputs where autonomous systems will thrive,” he explained. “Moving from adaptive to cognitive AI, you get to a framework of not only understanding the environment, but being able to explain and think through scenarios. We’re accelerating our efforts toward that.”
Chung says he is excited to do his part in accelerating the rate of technology development and maturation in robotics, AI, and autonomy. Driverless cars might be one of the first ubiquitous technologies to arrive, but as he sees it, we are going to witness many more such arrivals of AI-powered robotics and autonomy.
“Human beings are intelligent and autonomous, and we know very well about how many different ways things go right and wrong with people,” Chung said. “So we can learn from ourselves what we need to keep an eye on as we usher in artificial forms of intelligence and autonomy.”
This article originally appeared in DARPA: Defense Advanced Research Projects Agency 1958-2018 published by Faircount, LLC.