A new job, ground-based pilot, is likely to appear in aviation over the coming two or three decades, according to experts at the French Air and Space Academy (AAE). During a conference in Toulouse on June 1 and 2, they asserted safety trends demonstrate that more automation on board translates into better safety. There is still room for debate, however, as progress in neuroergonomics might reveal ways to compensate for human weaknesses. This would strengthen the case for safety specialists who believe aviation must retain a high level of human presence. They argue that the brain has long proved itself to be effective as master of the cockpit. While research in the man-machine interface sometimes struggles, military experience with the latest-generation unmanned and single-pilot fighters might help automate civil aircraft, including business jets.
The AAE presented the findings from research it started in 2012. It was assumed that automation is good for safety. “A decisive stage was passed in the 1970s and 1980s, when greater automation was made possible by the electronic revolution from analog to digital,” said Alain Garcia, an AAE v-p and a former Airbus engineering v-p. Thanks to automated systems, crew tasks have been simplified. In turn, this has reduced the number of crewmembers and “has led to a significant drop in the accident rate,” according to Garcia.
Michael Feary, a research scientist in the human-systems integration division at NASA Ames Research Center, concurred. He noted that introduction of four generations of automation since 1958 has had a visible, favorable effect on safety statistics. However, one unwanted phenomenon consistently recurreded during the first couple of years after introducing a particular technology, Feary said. A brief but noticeable spike could be seen in the number of accidents per million flight hours. Nevertheless, the downward trend was overwhelming, according to Feary.
Garcia expressed his expectation that researchers and industry will pursue this path. “An extension of automation, aimed at countering persistent failures such as loss of control in flight, collision with mountains, or runway excursions, will improve flight safety,” he said. He warned that designers should build in precautions based on human capacity.
Six scenarios were devised for a gradual increase in automation, with various distributions of responsibilities between ground and air. The AAE determined that, for commercial passenger transport, human presence on board will remain until at least 2050. It studied safety but stopped short of investigating economic implications.
The role distribution between humans and systems will have to be devised with care. “Will tomorrow’s aircraft be a robot? Will the pilot be a robot of that robot?” asked Catherine Tessier, head of the “piloting and decision” research unit at France’s aerospace research center, Onera.
Pilots on Board, Pilots on Ground
A desirable and attainable goal for 2050, according to Garcia, is to have one pilot on board (PB), whatever the flight duration. The pilot on the ground (PG) would need to be able to react within 10 minutes. This calculated time assumes that the PG will have to analyze the situation before acting. On a long-haul flight, the PG would be in control while the PB is in a rest period. In that situation, the PG would have to react in two minutes, in case of a problem.
The PG should contact the PB regularly, to ensure the latter is physically able to fly. How often they should be in touch “depends on the flight phase, as threats vary in density and type,” said Jean Broquet, an AAE member and former designer of automated satellite control systems. Contact should be continuous during takeoff, initial climb, final approach and landing. In cruise, a frequency of one contact every 30 minutes seems adequate, according to Broquet.
PGs, also known as ground operators, would be qualified as pilots. Each would hold a type rating. “Airlines might manage PGs and PBs together in human resources,” Broquet suggested. The AAE has estimated one PG can take care of five flights, on average, in short- to medium-haul operations. This could expand to seven flights after a few years of experience, efficiency improvement and workload reduction, Broquet said.
For long-haul operations, a pair of PGs could deal with six to eight flights, according to the AAE study. A rest period could be interrupted for threat management. This would happen one to three times every 10 flights, in the AAE’s assessment.
Denying the PB his or her authority would require more than one PG. A team of PGs would confer before making such a last-resort decision. During the discussions, the aircraft could be in autonomous mode.
“We recommend evaluating the weak and strong points in PB-PG cooperation,” Broquet added. Airbus has been working on single-PB operations for medium-haul, said Bernard Rontani, head of the company’s center of competence for systems. Under such an arrangement, communications will play an essential role, AAE’s Garcia noted, explaining that it will place more demands on coverage and bandwidth. Real-time video could be key.
The communication system to support the automated aircraft concept is considered achievable, according to Luc Deneufchâtel, an expert formerly with French air navigation service provider DSNA. However, the spectrum required for such operations is a critical issue, he said, citing the current congestion. The Ka band is the only potential additional spectrum for aviation needs, he noted. But it is not without challenges, including range. Separately, the need for two independent radio links–to ensure redundancy–will affect cost at aircraft and ground levels, Deneufchâtel cautioned.
“Making flights without a human [pilot on board] would be a serious mistake,” said Jean Pinet, an AAE member, doctor in psychology-ergonomics and former Concorde test pilot. Like automation, Pinet said, the human brain combines the skills for calculating derivatives (trends) and integrals (short-term memory) thanks to its perception of the present time.
“No study has ever counted the instances when the crew avoided a danger,” he pointed out, and a computer has difficulty defining a critical situation.
NASA’s Feary described big data as “associative intelligence,” while the human brain’s skills are rather about “generative intelligence.” Not all problems are associative in nature, he noted. “Humans are particularly good at adaptive problem-solving and discovery, areas where there has been little machine intelligence progress,” Feary added.
Bruno Nouzille, technical v-p of avionics for Thales Group, agreed. Although automation improves safety, humans are better at managing the unexpected, he emphasized. An airplane system malfunction occurs on 20 percent of flights, according to NASA data.
As an example of the strengths human beings provide over automation, he cited Toyota’s 2014 decision to replace some robots with humans on the production floor. The change reduced scrap by 10 percent, Feary pointed out. Moreover, “when a robot is doing the work, the process stops improving,” he said. This assertion was partly contradicted by Michael France, of UTC. He showed images of a robot designed to fold clothes. It could fold clothes it had never seen before, thus proving that it could deal with a high level of uncertainty.
Frédéric Dehais is the holder of the Axa chair for flight safety. His research laboratory in Toulouse, part of the ISAE SupAéro engineering school, has been making strides in neuroergonomics for pilots. As the word suggests, the goal is to adapt the flight deck to the way the human brain works best, using the latest developments in neuroscience. Future, safer cockpits could thus help the pilot if some situation degrades his or her performance.
One focus has been alarm deafness. For most human beings, measurements show that the sense of sight is dominant, according to Dehais. A stimulus coming from the eye can outweigh, in 100 milliseconds, a simultaneous stimulus coming from the ear. This happens before the person is even conscious of the stimuli, which takes 300 milliseconds. Before the brain knows what’s happening, it can shut down its own aural channel. Although these occurrences are rare, some high-profile accidents have focused scrutiny on such alarm deafness.
Nine pilots in a simulator were subjected to a stressing scenario Dehais devised. The sample group missed a number of aural alarms; in fact, one subject missed more than half of them. The simulation posed the tests before the formal simulation session began. One test sought insight into the brain’s working memory, which usually is key in pilot proficiency; the other was about visual dominance.
“We saw a good correlation between the latter test and how the pilot performed in the simulator,” Dehais said. A person with strong visual dominance was found to be more prone to miss an aural alarm. More trials are under way in flight, in a TB20 piston single.
Mind wandering is another topic Dehais’ team is working on. “The human brain needs mind wandering but it can be a safety issue when driving a car or flying an aircraft,” Dehais said. The issue is the delayed response to an unexpected, urgent situation. Research is still at an early stage but has already produced a useful interim result. Mind wandering can be detected when one area of the brain activates. In a test, subjects had to give clearances to aircraft flying in one direction. The clearance had to be denied to an aircraft flying in the opposite direction, a rare instance. As the task was predictable most of the time, the subject’s mind was often found in wandering mode. “Detecting mind wandering predicted an error 10 seconds in advance,” Dehais said.
Humans Not So Good at Monitoring
Dehais’ work is part of a broader research endeavor to help pilots in today’s cockpits. Robert Sumwalt, a member of the NTSB and a former airline pilot, emphasized the issue of monitoring. The Asiana crash in 2013 epitomized the problem, as it was found the crew was not properly monitoring airspeed.
Sumwalt quoted the late Raja Parasuraman, a neuroergonomics specialist who expressed a fundamental concern in 2002: “[Automation] is highly but not perfectly reliable in executing decision choices, then the operator may…fail to detect the occasional times when the automation fails.”
Sumwalt also mentioned change blindness. The human brain is poor at detecting even a gross visual change if it occurs in an object that is not the focus of attention. An example shown to the audience was the disappearance of the a/p (autopilot) indication on a display. As a potential advancement to help the pilot, Sumwalt suggested that the installation of a “low energy alert” system should become widespread.
Onera’s Tessier emphasized the need for transparency. The designer must ensure that an automated system will not make a decision unbeknownst to the pilot, she said. The risk, otherwise, is to disrupt the pilot’s situational awareness.
Training could be reworked, said Philippe Crebassa, deputy director at engineering and pilot school Enac. “Training is still oriented too much toward using the system rather than understanding it,” he said. Pursuing the latter goal yields a higher level of safety, Crebassa suggested, as critical situations dwindle but grow in complexity.
To help the crew with situational awareness, Thales is offering a “fly by trajectory” feature in its Avionics 2020 flight deck demonstrator. The upcoming flightpath–a blue line equivalent to two or three minutes of flight–is displayed in synthetic vision. The source of information is the flight director, programmed by the crew. The proposed trajectory can take into account weather information, so the system may offer to detour around a thunderstorm. If the pilot makes an input on the stick, the blue line will bend accordingly.
What about a comparison with the flightpath vector some head-up display users have become familiar with? “The flightpath vector is tangent to the trajectory in our design,” Thales’ Nouzille answered.
Research into the man-machine interface sometimes struggles. Dutch research center NLR has devised risk level displays that pilots deemed interesting, but tests showed they made little use of the new displays. The idea is to use colors to provide a comprehensive aircraft status overview. A combination of failures may color the takeoff performance indicator in red and the landing performance indicator in orange. The tested displays were called the Risk Information System. While pilots emphasized that the system was useful, they said it was difficult to incorporate it in their workflow and used it only when time was available.
Military experience with automation and autonomous aircraft might show the way for commercial and business aviation. A “recovery” button has been added in the Rafale cockpit, a pilot explained. If the pilot realizes he or she is spatially disoriented and pushes the button, the system puts the aircraft in a gentle climb with sufficient airspeed. The addition, made about 18 months ago, is intended to avoid a repeat of the spatial disorientation that caused the fatal crash of a Rafale in 2007.
Further automation is being considered. “We are studying how to make in-flight refueling automated for long missions, so the pilot can concentrate on the attack phase,” the Rafale pilot said. He is a member of a French air force working group on “the future of combat aviation.”
In partnership with other European firms, Dassault has tested an unmanned combat aircraft demonstrator, the Neuron. The aircraft has a wingspan greater than that of a Mirage fighter and uses Falcon 7X flight control sensors. It has already flown at Mach 0.70. Gradually, the Neuron has been allowed to fly beyond initially limited airspace. The level of confidence is now such that flying by populated areas and chemical factories near the Istres, southeast France flight-test center is no longer seen as an issue.
The Neuron relies on both pilots on the ground and autonomy. On the ground, no direct piloting means (like a stick) is available. Instead, ground control gives high-level orders such as “taxi” or “go around.” Low-level orders are available, too, like “capture and hold heading.” In the event of a datalink failure, the aircraft automatically flies to a so-called “safety area” in the flight-test center’s airspace.
Dassault’s system engineering director, Frédéric Falchetti, noted that the company’s pilots do not like the idea of moving the pilot out. However, they could not imagine anyone else being at the controls (albeit on the ground) for the test flights, he said. Although the program does not include the problems of air traffic management, “Neuron experience has shown the path to follow for future Falcons,” Falchetti said, stopping short of suggesting a business jet without a pilot on board. A design driver for future Falcons should thus be “extending the autopilot and autothrottle operation domain to cover all flight phases,” he said, including taxiing. The authority of automation should be raised to perform in all external conditions, including heavy turbulence.
The resilience of automation should be improved in all known degraded modes. “Today, a loss of automation induces a high pilot workload, often in a complex context which, if combined with a lack of training, may lead to an accident,” Falchetti emphasized.
Pilot understanding should be simplified. The man-machine interface should become “service oriented,” as opposed to system architecture-oriented. o
Could the Recovery Button Transfer to a Business Jet?
Dassault has added a “recovery” button to the Dassault Rafale cockpit. In the event the pilot realizes he or she is spatially disoriented, pushing the button puts the aircraft in a gentle climb at sufficient airspeed. What about a recovery button on a Falcon to prevent loss of control in flight? Dassault’s system engineering director, Frédéric Falchetti, said it would be tricky to implement. He referred to the Air France 447 crash in 2009. “What is disturbing us in the community is, how do I deal with incorrect airspeed data? How do I detect that and how do I gather accurate data?”
Insurer: Carrier Liability Will Remain
Sophie Moysan, chief claims and legal officer at insurance firm La Réunion Aérienne, endorsed the view that introducing four generations of automation has had a visible, favorable effect on safety statistics. But she predicted that moving to more automation will not affect liability in an accident. The trend toward the carrier being seen as fully liable will remain, she said. She noted that the Chicago convention mentioned pilotless commercial aircraft in 1944.