Leonardo and the artificial intelligence (AI) startup Daedalean have completed a flight test campaign demonstrating practical uses of AI for traffic detection and advanced navigation.
The companies performed the 40-hour flight test campaign from July to September 2023, but revealed their collaboration this week on the eve of HAI Heli-Expo 2024. Supported by a Eureka Eurostars grant, the testing took place at Leonardo’s PZL-Swidnick facility in Lublin, Poland, where SW4 and uncrewed/optionally piloted SW4 Solo helicopters were equipped with Daedalean’s cameras, computer and interface display.
The campaign demonstrated three broad functionalities, explained Emanuele Bezzecchi, Leonardo’s AI roadmap manager. The first was visual detection of air traffic, encompassing not only so-called “cooperative” traffic — aircraft equipped with transponders — but also non-cooperative traffic such as birds and drones. Daedalean’s system identifies and estimates the distance to both kinds of traffic, essentially serving as “a second pair of eyes that will help the pilot to detect what’s going on around the helicopter,” Bezzecchi said.
The second functionality was visual positioning, or precisely determining the aircraft’s location in space using visual landmarks rather than GPS. According to Bezzecchi, this capability is particularly useful as a backup to GPS in uncrewed aircraft, where alternative navigation systems may result in too much drift. It is also an aid to flight crews in situations in which GPS signals are blocked by terrain or deliberately jammed or spoofed, which has occurred in both civil and military contexts.
Third was landing guidance, enabling very precise set-downs. Bezzecchi likened this functionality to the sensor-enabled parking guidance in a modern car: drivers are still able to park without it, “but if you have it, it’s better.” Leonardo has already demonstrated its interest in making landings safer and easier for pilots through the development of an auto-landing capability for the AW169 helicopter.
Bezzecchi said that Daedalean’s system demonstrated impressive performance within its inherent limitations. “We are talking about a vision system, so we have to understand that vision is just one part of the spectrum for perception, [meaning] what the machine can understand from . . . the environment,” he cautioned. But “it performed really good in the environment that it’s built for,” he said, adding that Leonardo’s flight test engineers were especially impressed by the visual positioning system.
According to Mattia Cavanna, head of technology and innovation at Leonardo Helicopters, the goal of all three functionalities is to improve the situational awareness of the pilot while progressively enabling higher degrees of autonomy. That will ideally reduce accidents in the near term while preparing Leonardo for a future in which autonomous and highly augmented flight are more widespread.
Headquartered in Switzerland, Daedalean has become a leader in developing certifiable AI for aviation, with a specific focus on the type of AI called machine learning (ML). It’s not a straightforward task. Much of the existing guidance for certifying aviation software cannot be applied to machine-learned algorithms, which operate very differently than classical software programs. In the subset of ML called deep learning, algorithms independently create models through repeated exposure to training data instead of through explicit programming.
Moreover, there are separate hurdles to certifying hardware with the computational performance needed to support Daedalean’s computer vision system. This is complicated by the fact that manufacturers of computer processors are often unwilling to provide aerospace suppliers with the proprietary data needed to support certification.
As a pioneer in the space, Daedalean has consequently been forced to do more than simply develop a functional product. It has worked extensively with the European Union Aviation Safety Agency and Federal Aviation Administration to develop frameworks for ensuring that its ML algorithms work as intended. It has also partnered with Intel to develop certifiable hardware for its system. The company is currently in the process of certifying its first AI-based product for general aviation, a visual traffic awareness system called PilotEye that it is developing jointly with Avidyne Corporation.
For Leonardo, the partnership with Daedalean has allowed it to leverage the startup’s foundational work in visual awareness systems, rather than having to replicate it in-house. This approach frees up Leonardo to focus on other technology areas where solutions do not already exist.
“We’re very happy with the collaboration so far,” said Cavanna, noting that the helicopter manufacturer is striving to stay “humble” as it pursues its technology roadmap. AI is one core piece of that roadmap, he said, and “the elements offered by brilliant [subject matter experts], brilliant startups and other technology partners around us are even more essential today than in the past to achieve success.”
