Artificial intelligence (AI) is disrupting every facet of the aviation industry, and the technology is poised to completely change the way aircraft fly. While pilotless passenger flights on self-flying, AI-powered airliners won’t become a reality anytime soon, AI technology is already creeping into airplane flight decks and transforming how pilots do their jobs—and how they learn to fly.
A handful of companies developing AI-driven flight control systems are already flying autonomous aircraft experimentally, but AI isn’t piloting any aircraft yet. Rather, autonomous airplanes being flown today by companies such as Xwing, Reliable Robotics, and Merlin rely on enhancements to existing autopilot technologies and remote-piloting capabilities. Using Cessna Grand Caravans as testbeds, those companies aim to convert existing legacy aircraft into autonomous aircraft under supplemental type certificates (STCs).
“Autonomy is not necessarily equal to artificial intelligence, or machine learning, or deep neural networks, or non-deterministic systems or anything like that. Autonomous machines can be perfectly deterministic,” explained Honeywell Aerospace’s Stephane Fymat at the Vertical Flight Society’s 10th biennial autonomous VTOL technical meeting. “That's good, because it means that the certifiability of autonomous aircraft can be a more tractable problem, especially considering some of the deeper challenges that we're addressing right now around the explainability of AI, trust of AI, the generalizability of AI.”
The Road to Self-piloting Airplanes
Air safety regulators such as the FAA and the European Union Aviation Safety Agency (EASA) currently have no set path to certification for aircraft that fly themselves with AI. However, that’s something both agencies are looking to change in the not-too-distant future.
EASA and the FAA have both published their own AI “roadmaps” in which the agencies spell out similar approaches to certifying AI. In EASA’s AI roadmap, a “living document” last updated in 2023, the regulator said it expects fully autonomous AI aircraft to enter service in the 2035 to 2050 timeframe—yet it expects to certify its first AI avionics system as early as next year.
According to EASA’s roadmap, initial applications for AI in the cockpit will center around pilot assistance, helping to reduce pilot workload and improve safety. Over the next decade, progressively more automated solutions will enable reduced-crew or single-pilot operations—maintaining a human in the loop—and by 2050, AI could be flying aircraft without any human supervision.
The first AI-powered avionics system appears to be on track for EASA certification by the end of this year, according to Swiss AI start-up Daedalean, which is working on an STC for an AI-powered pilot assistance system called PilotEye.
Developed in partnership with U.S. avionics manufacturer Avidyne, PilotEye uses cameras, sensors, and AI software developed with machine-learning algorithms to provide pilots with complete situational awareness plus detect-and-avoid capabilities. The company refers to its computer vision technology as “situational intelligence” because it “encompasses situational awareness as well as the ability to anticipate and react to a threatening situation.”
PilotEye uses AI and neural networks to analyze images and data from various cameras and sensors to accurately perceive its surroundings. “You don't need AI for decision-making; you need the AI part for seeing that it's a runway and not a road and that it's actually going the way you wanted to land on,” Daedalean founder and CEO Luuk van Dijk told AIN.
The AI-enabled computer vision software can accurately identify objects and obstacles to determine, for example, whether a camera has spotted “a cloud that’s not actually moving, or there's an airplane that's coming right at you,” he explained. “These are a couple of the harder problems.”
Avidyne intends to offer the PilotEye as a feature in the Quantum avionics suite it launched last year specifically for new advanced air mobility aircraft, such as eVTOL air taxis and hybrid-electric regional airplanes. Daedalean is also developing an AI flight planning system that integrates with an aircraft’s autopilot and enables autonomous navigation. This would ensure an aircraft can safely travel through designated corridors, such as eVTOL air taxis flying at low altitudes in urban areas.
“Things such as the ability to perceive or localize the aircraft in the world, the ability to interact perhaps with ATC using natural language (as much as we would all like it to go to data), the ability to help an aircraft plan complex courses of action on the fly in novel situations, and the ability to learn from the world also might make the system less brittle, and less expensive potentially to build or maintain,” Fymat explained.
Beyond Flight Controls
Although totally AI-piloted aircraft may not enter service for several decades, some AI technology is already flying on aircraft today—just not with the power to control the aircraft. So far, the focus of AI on aircraft has revolved around improving operational efficiencies.
For example, flight management systems use AI algorithms to optimize routes and reduce fuel burn. Generative AI can also help to automate routine tasks. For example, in October, Turkish low-cost carrier Pegasus Airlines introduced AI-generated pilot announcements that provide passengers with flight information in several languages, allowing pilots to stay focused on flying.
Earlier this year, Swedish company Web Manuals launched a new AI-powered search tool for pilots’ flight manuals. Called Amelia AI, the platform functions as a chatbot that pilots can use to quickly retrieve information from within their flight manuals, allowing them to make swift and well-informed decisions.
Web Manuals’ AI search function “makes navigating through large amounts of operational and company manuals much more manageable and time efficient, with data retrieval within seconds,” said the company’s chief technology officer, Richard Sandström.
Researchers are also experimenting with AI-driven face and voice recognition software that could potentially monitor pilots’ health and mental well-being. Blueskeye AI, a U.K.-based software company, uses facial recognition and voice analysis software “to look at medically and biophysically relevant behavior, so we can use it to help assess, diagnose, monitor, and treat medical conditions that actually change your expressive behavior,” such as fatigue, pain, and anxiety, Blueskeye AI founder and CEO Michel Valstar told AIN.
When it comes to AI, the aviation industry (along with the rest of the world) is only beginning to understand the wealth of possibilities the technology can offer. While AI is changing the way aircraft and pilots fly, it’s also changing just about every aspect of ground operations, ranging from flight booking platforms to supply chain management.
AI Improves Pilot Training
Flight schools already have initiatives underway to use AI technology to make pilot training more efficient, effective, and accessible. At the forefront of those efforts is CAE, a global company that manufactures flight simulators and other training devices for airlines and flight schools.
CAE is developing AI-driven tools that can provide personalized training experiences by analyzing a student pilot’s performance and tailoring training plans to every individual’s specific needs.
“For future learning ecosystems to be successful, training providers must find ways to smartly blend the incredible advances we’re seeing in education technology with the latest understanding in learning science,” said Regan Patrick, CAE’s chief learning officer. “It’s important for CAE to understand what cognitive, behavioral, and neuroscience tells us regarding how humans learn so we can carry that forward into our training systems designs.”
According to CAE, its patented adaptive learning technology makes training more efficient and cost-effective. It implements systematic and objective rule-based measures of performance and effectiveness, and it provides instructors and students with real-time feedback. With an integrated biometric suite, it allows instructors to gauge students’ stress, engagement, and cognitive workload.
“One of the biggest challenges we see is the need to adjust mindsets, specifically migrating from traditional instructor-based training approaches to more student-centric experiential learning,” Patrick said.
CAE has also been using AI for years to enhance its flight simulators. According to the company, AI enables a broader range of realistic and dynamic environments and scenarios, including unanticipated events that could catch a pilot by surprise.
In addition to AI and machine-learning capabilities, CAE and other flight training companies are integrating several other cutting-edge technologies into their training products, including virtual reality, mixed reality, and data analytics. CAE’s virtual-reality training hardware includes headsets and hand-tracking technology. According to CAE, combining those types of tools with artificial intelligence will further expand the horizons for extended reality training.
Another leader in AI-enhanced flight training solutions is FlightSafety International, which is also developing AI-powered adaptive learning capabilities. In 2019, FlightSafety and its partner IBM announced plans to introduce a platform called FlightSmart that uses AI to improve training processes.
“The product, through the collaboration with IBM, is utilizing advanced algorithms, machine learning, artificial intelligence—all of those cognitive technologies—to provide the objective evidence or objective evaluation of the student's performance,” said Matt Littrell, FlightSafety product director of AI and adaptive learning.
FlightSafety is working to integrate FlightSmart with flight simulators as well as other training processes, including task training on avionics and operating flight management systems, the use of automation, standard operating procedures, crew resource management, and maintenance.
During simulator training, the AI software monitors more than 4,000 variables in real time, objectively evaluating a student’s performance in specific maneuvers. It also distinguishes between different flying styles, which FlightSafety’s team has classified as either shy, reactive, risk-taking, or aggressive. The technology is capable of picking up nuances that instructors may not be able to directly observe, like how much force a student pilot is applying to the stick controls, for example.
So far, use of the FlightSmart system has been limited to military operations. For its first use case, FlightSafety implemented the technology on 16 T-6A training devices at a U.S. Air Force base in Columbus, Mississippi. It completed acceptance testing in 2020. Now FlightSafety continues development work to refine the system and optimize the design of the FlightSmart interface.
FlightSafety aims to make FlightSmart available for civilian pilot training next. Eventually, the technology could expand to help train other types of aviation professionals, including maintenance technicians and drone operators, according to FlightSafety. The technology could potentially transfer over to other industries, improving training processes for locomotive conductors, maritime operators, truck drivers, and even surgeons.