The development and application of increasingly autonomous (IA) systems for civil aviation (see Boxes 1.1 and 1.2) are proceeding at an accelerating pace, driven by the expectation that such systems will return significant benefits in terms of safety, reliability, efficiency, affordability, and/or previously unattainable mission capabilities. IA systems, characterized by their ability to perform more complex mission-related tasks with substantially less human intervention for more extended periods of time, sometimes at remote distances, are being envisioned for aircraft and for air traffic management (ATM) and other ground-based elements of the national airspace system (NAS) (see Box 1.3). This vision and the associated technological developments have been spurred in large part by the convergence of the substantial investments by the federal government in advanced unmanned aircraft systems (UAS) (see Box 1.4) for military and, to a lesser extent, civil applications and by advances in low-cost, high-capability computing systems; sensor technologies; high-throughput digital communications systems; precise position, navigation, and timing information (e.g., from the Global Positioning System (GPS); open-source hardware and software; and the emergence of an active and prolific hobbyist community that has provided and continues to provide fertile ground for innovation and entrepreneurship that did not exist a decade ago. The burgeoning industrial sector devoted to the design, manufacture, and sales of IA systems is indicative of the perceived economic opportunities that will arise. In short, civil aviation is on the threshold of potentially revolutionary improvements in aviation capabilities and operations associated with IA systems. These systems, however, pose serious unanswered questions about how to safely integrate these revolutionary technological advances into a well-established, safe, and efficiently functioning NAS governed by operating rules that can only be changed after extensive deliberation and consensus.
This report identifies key barriers and suggests major elements of a national research agenda to address those barriers and help realize the benefits that IA systems can make to crewed aircraft, UAS, and ground-based elements of the NAS. This agenda is in large part motivated by the demonstrated capabilities of advanced aerospace vehicles—such as crewed and unmanned military and commercial aircraft, spacecraft, and planetary rovers—that rely on systems that operate with varying levels of autonomy and by the growing desire to have both aircraft and ground-based systems with increased autonomy. Such systems, generally operating within well-defined limits on their ability to act without the direct control of operators, have demonstrated the ability to enable new types of missions, improve safety, and optimize the workload of, for example, pilots in crewed aircraft and remote operators for unmanned aircraft (see Box 1.5). On the other hand, autonomous systems can introduce uncertainties if
In this report, “civil aviation” is used to refer to all nonmilitary aircraft operations in U.S. civil airspace. This includes operations of civil aircraft as well as nonmilitary public use aircraft (that is, aircraft owned or operated by federal, state, and local government agencies other than the Department of Defense). In addition, many of the IA technologies that would be developed by the recommended research projects would generally be applicable to military crewed and/or unmanned aircraft for military operations and/or other operations in the NAS.
Increasingly Autonomous Systems
A fully autonomous aircraft would not require a pilot; it would be able to operate independently within civil airspace, interacting with air traffic controllers and other pilots just as if a human pilot were on board and in command. Similarly, a fully autonomous ATM system would not require human air traffic controllers. This study is not focused on these extremes (although it does sometimes address the needs or qualities of fully autonomous unmanned aircraft). Rather, the report primarily addresses what the committee calls “increasingly autonomous” (IA) systems, which lie along the spectrum of system capabilities that begin with the abilities of current automatic systems, such as autopilots and remotely piloted (nonautonomous) unmanned aircraft, and progress toward the highly sophisticated systems that would be needed to enable the extreme cases. Some IA systems, particularly adaptive/nondeterministic IA systems, lie farther along this spectrum than others, and in this report such systems are typically described as “advanced IA systems.”
National Airspace System
The NAS is “the common network of U.S. airspace; air navigation facilities, equipment, and services; airports or landing areas; aeronautical charts, information and services; rules, regulations, and procedures; technical information; and manpower and material” (Integration of Civil Unmanned Aircraft Systems [UAS] in the National Airspace System [NAS] Roadmap, FAA, 2013). Some NAS facilities are jointly operated by the FAA and the Department of Defense. IA systems could be incorporated into airport ground systems such as snow plows. However, the greatest technological, social, and legal challenges to the use of IA systems in civil aviation are associated with their use in aircraft and air traffic management systems, and the report does not specifically address the use of IA systems in airport ground systems.
Unmanned Aircraft/Crewed Aircraft
An unmanned aircraft is “a device used or intended to be used for flight in the air that has no onboard pilot. This device excludes missiles, weapons, or exploding warheads, but includes all classes of airplanes, helicopters, airships, and powered-lift aircraft without an onboard pilot. Unmanned aircraft do not include traditional balloons (see 14 CFR Part 101), rockets, tethered aircraft and un-powered gliders.” A UAS is “an unmanned aircraft and its associated elements related to safe operations, which may include control stations (ground-, ship-, or air-based), control links, support equipment, payloads, flight termination systems, and launch/recovery equipment” (Integration of Civil Unmanned Aircraft Systems [UAS] in the National Airspace System [NAS] Roadmap, FAA, 2013). UAS include the data links and other communications systems used to connect the UAS control station, unmanned aircraft, and other elements of the NAS, such as ATM systems and human operators. Unless otherwise specified, UAS are assumed to have no humans on board either as flight crew or as passengers. “Crewed aircraft” is used to denote manned aircraft; unless specifically noted otherwise, manned aircraft are considered to have a pilot on board.
In this report, the term “operator” generally refers to pilots, air traffic controllers, airline flight operations staff, and other personnel who interact directly with IA civil aviation systems. “Pilot” is used when referring specifically to the operator of a crewed aircraft. With regard to unmanned aircraft, the FAA says that “in addition to the crewmembers identified in 14 CFR Part 1 [pilots, flight engineers, and flight navigators], a UAS flight crew includes pilots, sensor/payload operators, and visual observers, but may include other persons as appropriate or required to ensure safe operation of the aircraft” (Integration of Civil Unmanned Aircraft Systems [UAS] in the National Airspace System [NAS] Roadmap, FAA, 2013). Given that the makeup, certification requirements, and roles of UAS flight crews are likely to evolve as UAS acquire advanced IA capabilities, this report refers generally to UAS operators as the flight crew rather than specifically as pilots.
they are not thoroughly assessed over a broad range of normal and abnormal operating conditions. Autonomous systems may reduce operational costs by reducing the need for highly trained human operators, but these savings may be partially or wholly offset by the need for substantive advances in system design, testing, and certification processes. As a result, it remains to be seen what benefits IA systems in various civil aviation applications will be able to provide without degrading safety, reliability, mission performance, and/or net costs. This is especially important in commercial aviation, where extremely high levels of safety and a very competitive business climate are the norm.
At the highest level, autonomy implies the ability of the system (often a machine) to perform tasks that involve dynamically executing a “decision cycle” in much the same fashion as a human. That decision cycle can
FIGURE 1.1 The classic OODA loop.
- Decide, and
The concept was originally developed during the Vietnam era to describe and advocate effective air combat tactics. An autonomous system, be it human or machine, first observes by sensing or acquiring information about the environment from other relevant sources. It then orients itself toward the task at hand. This second step infers a number of functions that can encompass information fusion, contextual interpretation, the integration of learned behaviors, and even inferences about future events. In the robotics community, a number of the capabilities associated with this step are often referred to in aggregate as perception. The third step then involves making a decision based on the task objectives and the results of the prior steps. This requires that the system be capable of implementing an appropriate action that accomplishes the task. Once the action is complete, the cycle repeats as the system observes the consequences of the action as well as changes in the environment caused by other factors. The OODA loop has found application in many diverse areas, including business and law.2,3
Within the aviation context, this decision cycle may be implemented one time in isolation (e.g., a simple aircraft heading change) or repeated with regularity (e.g., a digital flight control system operating at 30 decision cycles per second), depending on the task. The OODA concept of autonomy applies to tasks that range from lower-level functions, such as stabilization and basic maneuvering of an aircraft (e.g., a fly-by-wire control system), to high-level mission decisions and even to the accomplishment of a complete mission. Some of these capabilities exist today, while others will be possible only as IA technologies mature over time.
There is a continuum that describes the roles of humans and machines when they interact. This continuum ranges from systems at one end of the spectrum, where the human is in total control, to systems at the other end, where they operate without the need for human interaction. Some systems operate at the extremes. For example, antilock braking systems and airbag systems on cars are fully automatic, deciding on their own when to act. However, fully automatic systems as well as fully autonomous systems depend on humans to define and limit the scope of their authority and the range of possible actions. Movement along the continuum typically does not eliminate or
1 Frans P.B. Osinga, 2007, Science, Strategy and War: The Strategic Theory of John Boyd (Strategy and History), London and New York: Routledge, Taylor and Francis Group.
2 Chet Richards, 2004, Certain to Win: The Strategy of John Boyd, Applied to Business, Bloomington, Ind.: Xlibris Corporation.
3 A.S. Dreier, 2012, Strategy, Planning & Litigating to Win: Orchestrating Trial Outcomes with Systems Theory, Psychology, Military Science and Utility Theory, Boston, Mass.: Conatus Press.
TABLE 1.1 Characteristics of Advanced Automation and Autonomy
|Characteristics||Advanced Automation||Advanced Autonomy|
|Augments human decision makers||Usually||Usually|
|Proxy for human actions or decisions||Usually||Usually|
|Reacts at cyber speed||Usually||Usually|
|Reacts to the environment||Usually||Usually|
|Reduces tedious tasks||Usually||Usually|
|Robust to incomplete or missing data||Usually||Usually|
|Adapts behavior to feedback (learns)||Sometimes||Usually|
|Exhibits emergent behavior||Sometimes||Usually|
|Reduces cognitive workload for humans||Sometimes||Usually|
|Responds differently to identical inputs||Sometimes||Usually|
|Addresses situations beyond the routine||Rarely||Usually|
|Replaces human decision makers||Rarely||Potentially|
|Robust to unanticipated situations||Limited||Usually|
|Adapts behavior to unforeseen environmental changes||Rarely||Potentially|
|Behavior is determined by experience rather than by design||Never||Usually|
|Makes value judgments (weighted decisions)||Never||Usually|
|Makes mistakes in perception and judgment||N/A||Potentially|
diminish the importance of the humans to the operations of the system, but it does change their role. In fact, the human’s role becomes more rather than less important when moving toward the autonomous end of the spectrum because it is so important to assure that the systems are properly designed, tested, deployed, and monitored to ensure the aircraft’s continued airworthiness.
The International Civil Aviation Organization (ICAO) is an agency of the United Nations that develops aviation standards that member states may use when developing national aviation regulations. ICAO has defined an autonomous aircraft as “an unmanned aircraft that does not allow pilot intervention in the management of the flight.”4 This is an unfortunate definition since it is unlikely that any unmanned aircraft would be designed to “not allow” human intervention. On the contrary, both civil and military unmanned aircraft are designed to allow operators to intervene whenever they are in contact with their aircraft. In addition, autonomous aircraft would not necessarily be unmanned, in that properly equipped aircraft would be able to operate autonomously even with a crew and/or passengers on board. Accordingly, it would be more accurate and more useful to define an autonomous aircraft as “an aircraft that does not require pilot intervention in the management of the flight.” This slightly altered definition suggests a policy that focuses on assuring the safety of unmanned aircraft operations without the presumption that safety will necessarily require continuous oversight by an onboard or remote pilot. However, even this definition fails to describe the full range of capabilities that make systems autonomous, nor does it provide an understanding of how autonomous or IA systems can be distinguished from automated systems, which are quite common in the NAS. The differences between automated and autonomous systems—and the characteristics shared by both—are addressed below.
Automated systems can be very complex and behave in a way that some might say reflects a form of artificial intelligence and, thus, embodies a degree of autonomy. The difference may be in the eye of the beholder. In fact, complex automatic or IA systems share many of the same operational, technical, and policy challenges.
Table 1.1 describes the characteristics of systems with advanced automation or autonomy. There are no absolutes and no specific characteristics that label a system as being automatic or autonomous. Thus, this report focuses on barriers and required research necessary to make relevant technological advances and less on the definitions. Also, because there are no absolutes associated with a formal definition of an autonomous system, this report uses the term increasingly autonomous as a label to discuss the operational, technical, and policy barriers to implementation in civil aviation and in presenting a research agenda.
4 International Civil Aviation Organization, 2011, Circular 328 AN/190—Unmanned Aircraft Systems (UAS).
Automation and autonomy have been used in aviation almost from the beginning. The Sperry Autopilot developed in 1912 allowed straight and level flight without the pilot operating the stick. This system was demonstrated in 1914 at the Paris Concours de la Sécurité en Aéroplane show, with an aircraft that performed a stable fly-by with both pilots standing on the wings. This autopilot system was incorporated into the first unmanned aircraft, the Kettering Bug, in 1918.
The Second World War saw the expansion in the use of flight control systems. Navigation systems were coupled with autopilots to facilitate flying in adverse weather conditions and at night. In 1947 a C-53 conducted the first transatlantic flight with the aircraft completely under the control of the autopilot, including takeoff and landing. The beginnings of fly-by-wire technology were seen in the late 1950s and 1960s. These systems used an analog computer as an electrical control system, with a mechanical backup control system. The Avro Canada CF-105 Arrow was the first aircraft to implement such an approach, in 1958, and it was also used on the Concorde (1969) and the Apollo Lunar Landing Research Vehicle (1964), although the latter did not include a mechanical backup. The first production aircraft to use an analog fly-by-wire control system without a mechanical backup was the F-16A/B (1976). The F-16 was also designed for relaxed stability, which required an analog computer to maintain stability and keep it in the air; a quadruply redundant configuration was used for reliability.
Early radio-controlled unmanned aircraft (e.g., the Radioplane RP-4/OQ-1) appeared in the 1930s and 1940s. This type of aircraft, which was controlled by pilots on the ground, soon gave way to target drones (e.g., Ryan Firebee), which had greater endurance and range and were no longer remotely piloted. Instead, they used simple dead-reckoning navigation.
The Vietnam War saw the evolution of target drones into UAS performing intelligence, surveillance, and reconnaissance missions as well as electronic warfare missions (e.g., Ryan AQM-34 and AQM-91A). These systems had long-range navigation capability, autonomous pre-programmed missions, and the ability to react to internal monitoring. As the Vietnam era ended, the United States placed less emphasis on UAS development, while Israel began to ramp up its efforts (e.g., Tadiran Mastiff).
The replacement of analog computers by digital computers provided the necessary foundation for significant autonomy enhancements in crewed and unmanned aircraft. After developing a digital fly-by-wire control system for the Apollo command module and lunar lander, the National Aeronautics and Space Administration (NASA) validated the concept of an all-digital, fly-by-wire flight control system in 1972 using an F-8C Crusader with no mechanical backup. This would become the forerunner to the all-digital fly-by-wire control system used by the Space Shuttle Orbiter Enterprise (1977) for unpowered approach and landing tests and by the upgraded F-16C/D (1984). The Airbus A320 (1987) was the first commercial transport to include a full-authority digital fly-by-wire flight control system, which allowed it to be equipped with flight envelope protection. Virtually all large commercial transports, including business jets, now incorporate digital fly-by-wire flight control systems.
U.S. interest in military UAS ramped up in the mid-1980s with several programs conducted by the Defense Advanced Research Projects Agency that were focused on long-endurance operations, as well as the Navy’s early experiments with Pioneer UAS. In the early 1990s, during the first Gulf War, UAS such as the Predator A were acquired in large numbers. These systems rapidly gained the respect of the operational community for their ability to provide essentially uninterrupted, continuous surveillance of targets and other points of interest using another new technology, streaming video. During that period a number of enhancements were introduced to the largest of these systems, including satellite communications connectivity and basic mission-level automation. Waypoint navigation enabled unmanned aircraft to fly complex flight paths without the aid of real-time pilot interaction. The combination of these technologies permitted ground operators to be stationed at long distances from the aircraft in flight. Communications latency, however, precluded remote operators from piloting aircraft in real time,
in terms of controlling aircraft state and dynamics. In addition, automatic takeoff and landing improved safety and reliability by eliminating the need for real-time interactions between ground operators and unmanned aircraft during these time-critical events. The collective increase in capability provided by automated systems, along with other mission risk mitigation software, led to the elimination of many traditional cockpit controls. For example, stick-and-rudder pedal controls were not included in the design of the first ground station for Global Hawk UAS (1998) and subsequent advanced UAS.
Precision GPS for both military and commercial applications was introduced in the late 1990s. This capability greatly enhanced autonomous navigation. The merger between precision digital control and precision navigation allowed for an expansion of the mission envelope by UAS. Intelligence, surveillance, and reconnaissance capabilities were improved to the point that UAS were providing more information than ground personnel and systems could process in a timely fashion. Beginning in 2001, UAS, notably the Predator MQ-1A, were equipped with weapons, and after completion of testing they began to perform integrated surveillance and strike missions.
After 9/11 the number of military UAS skyrocketed. Fixed and rotary-wing unmanned aircraft as small as a few inches (micro air vehicles) and as large as a commercial transport have found their way into a variety of applications. The operational envelope has also been expanded to include high subsonic speeds, as exhibited by the X-47B unmanned combat air system. The array of missions has expanded to include air-to-ground combat and rapid resupply transport. Advanced features comprise increasingly sophisticated autonomous capabilities, including contingency management so that the aircraft can continue operations even if, for example, communications are lost or aircraft flight characteristics are degraded by damage. Because of the urgent demand for UAS in theaters of conflict and the military’s less stringent requirements for assuring operational safety of UAS prior to use in military airspace, UAS have served as pioneers in the development and test of IA capabilities.
Industry and government are using UAS in some other countries. For example, the Yamaha RMAX small unmanned helicopter is widely used in Japan for precision agriculture applications, including crop dusting. In the United States, the government is using UAS for civil missions such as border patrol, weather monitoring, and disaster response. The use of UAS by local law enforcement is still in its infancy. The deployment of UAS for commercial applications in the United States has been extremely limited but is likely to expand rapidly as technical and regulatory barriers are addressed. The first routine commercial operations in the United States are likely to involve small unmanned aircraft operating in accordance with the standards currently established for the operation of model aircraft, which are as follows:5
a. Select an operating site that is of sufficient distance from populated areas. The selected site should be away from noise-sensitive areas such as parks, schools, hospitals, churches, etc.
b. Do not operate model aircraft in the presence of spectators until the aircraft is successfully flight tested and proven airworthy.
c. Do not fly model aircraft higher than 400 feet above the surface. When flying aircraft within 3 miles of an airport, notify the airport operator, or when an air traffic facility is located at the airport, notify the control tower, or flight service station.
d. Give right of way to, and avoid flying in the proximity of, full-scale aircraft. Use observers to help if possible.
e. Do not hesitate to ask for assistance from any airport traffic control tower or flight service station concerning compliance with these standards.
The committee’s overarching vision for increased autonomy in civil aviation is that all aircraft and ground systems will be imbued with new or improved capabilities that enable them to function more safely, reliably,
5 FAA, 1981, Advisory Circular 91-57, Model Aircraft Operating Standards, June 9, http://www.faa.gov/documentLibrary/media/Advisory_Circular/91-57.pdf.
and efficiently over an expanded array of missions constrained only by technological limitations and acceptable margins of risk and cost. A mix of crewed and unmanned aircraft will operate in shared airspace, guided by ATM systems with distributed responsibilities and authorities to assure safe separation and maximize traffic flow during normal and abnormal operating conditions (e.g., in adverse weather). In addition, both air and ground systems are designed to minimize the susceptibility of individual systems and the NAS as a whole to various failure modes.
Detailed information on the potential benefits of achieving the visions for crewed aircraft, unmanned aircraft, and ATM appears in Chapter 2.
The committee’s vision for increased autonomy in crewed aircraft is that (1) flight crews will be better able to perform existing, proposed, and potential tasks because onboard IA systems will provide guidance and aid for the requisite tasks, monitor the performance of both the pilot and the system for predictors of failure, and—within the limits of authority granted to particular IA systems—intervene when necessary to assure flight safety; (2) single-pilot operations will be a viable option for air carriers because the onboard IA systems will perform functions that are currently being performed by humans; and (3) the need to have a pilot on board passenger aircraft may be eliminated.
The committee’s vision for increased autonomy in unmanned aircraft is that this class of vehicle takes on a growing array of missions as advanced IA systems improve the ability of unmanned aircraft to self-separate, to autonomously coordinate their missions, and to operate unattended for extended periods of time in segregated airspace and, ultimately, in airspace shared by crewed aircraft, while also reducing the cost and increasing the flexibility of UAS operations.
The committee’s vision for increased autonomy in ATM is that air traffic controllers, traffic flow managers, and other personnel will be better able to perform existing, proposed, and potential tasks because the IA systems that they use will provide guidance and aid for the requisite tasks, monitor their performance and that of the entire system for predictors of failure, and provide functionalities that heretofore could only be provided by humans, thereby increasing the efficiency of operations in the NAS.