On The Radar
Autonomous flight start-up Daedalean has published a roadmap outlining its path to developing self-flying aircraft by 2028.
The Swiss company is working on a vehicle-agnostic, fully autonomous flight system, starting with situational- awareness software and sensors that will be used as an aid for pilots. The company aims to eventually be able to convert any type of aircraft to fly fully autonomously—without the aid of an onboard safety pilot or remote pilot on the ground—using artificial intelligence (AI) and machine learning.
“If you want to actually replace the pilot, either remote or onboard, you have to make systems that can do that," Daedalean CEO Luuk van Dijk told FutureFlight. "So, for that, you need to have this complete situational awareness, and after that, you need sufficient sensors, and then you need these AI parts to make sense of these sensors.”
To start, Daedalean is working on getting its situational-awareness system certified for general aviation purposes, under design assurance level (DAL) C, with the aim of eventually certifying a fully autonomous flight system up to DAL A, the highest level of design assurance that can be applied to airborne software. The company plans to launch its PilotEye machine learning-based onboard pilot aid for general aviation in 2023. It is developing the system in collaboration with the Florida-based avionics company Avidyne.
“Our partner Avidyne is making the boxes, and we're going through the certification exercise and when that's done, we'll have a supplemental type certificate,” van Dijk said. “That will be a big moment for the industry as a whole because it will be the first time a machine-learning component will have been certified to design assurance level C, which means appropriate for safety cases that have major impacts.”
The first implementation of Daedalean’s technologies is based on computer vision. “The system—we call it VXS for ‘visual everything system’—relies only on the input from visual cameras, which makes it suitable for visual meteorological conditions,” Daedalean wrote in the white paper describing its roadmap through 2028.
“Working on that, we solved the two tasks that were the hardest: first, to create the machine-learning-based technology able to recognize, categorize, and interpret the sensor input at the distances, velocities, and uncertainties of the real flight, working in a computing box of reasonable size, weight, and power consumption. Second, to establish the principles on which such technology can be certified, working side by side with the regulators on developing design assurance methods for it,” the paper adds.
The roadmap then describes what Daedalean calls the IXS, or “instrument everything system,” which will use data from cameras and other traditional instruments—such as GPS, altimeters, lidars, radars, and traffic information sources—to be able to operate in instrument meteorological conditions. “Daedalean’s IXS will provide fully certified terrain and traffic collision avoidance, as well as landing guidance and flight-plan following capabilities, in VMC and IMC, certified to appropriate levels to function with minimal crew oversight only (EASA Level 3),” the paper states.
Daedalean plans to further enhance the IXS system with vehicle-to-vehicle communications capabilities, allowing aircraft to share their map data and flight plans to coordinate safe guidance. This feature, called XXS, could be particularly useful for air traffic management in urban air mobility, according to the paper.
To get these new AI and machine-learning-based technologies certified, Daedalean is working closely with the European Union Aviation Safety Agency (EASA). In February 2020, that agency published its AI roadmap, which formalized the stages of certification for machine-learning applications.
EASA’s AI roadmap is divided into three phases. The first phase, which extends through 2024, is an exploratory period. The next phase, from 2024 to 2028, is labeled as “framework consolidation,” which is when the agency plans to issue its first approvals of AI and machine-learning components. Phase three, labeled as “pushing barriers,” will see further innovation in AI technology and the first fully autonomous commercial air transport operations.
Daedalean has also participated in a joint research project with the U.S. Federal Aviation Administration (FAA) on neural network-based runway landings, which could help shape the agency's policies on AI and machine learning.
To get its fully autonomous flight technology certified, Daedalean expects regulators will require the company to complete several years and several thousand hours of flight and service hours.
“Technically, we're not flight testing yet because we don't have the qualification units from Avidyne,” van Dijk told FutureFlight. “But for the software purposes, we're gathering data and we're testing that our algorithms work in flight without connecting them to the aircraft. So that's an advantage of situational awareness. You can test it out on recordings.”
You can read Daedalean's full white paper, as well as several flyers describing its autonomous systems in detail, on the company's website.