UAM - Fremdeles ikke tegn til enhetlige sertifiseringsforskrifter - Avionics Interntional

Det eneste jeg vet er at utgangspunktet til EASA og FAA er det samme: UAM skal ha samme sikkerhetsnivå som for kommersiell flyging - Uber, som til dels har gjort seg til talsmann for industrien, mener imidlertid at det ikke trenger å være så strengt. (Red.)

Airbus Wayfinder Is Building Autonomy for Urban Air Mobility and Single-Pilot Operations

By Brian Garrett-Glaser | November 21, 2019
Send Feedback

artificial intelligence, Autonomy, Computer vision, detect and avoid, machine learning, single-pilot operations

Airbus Wayfinder grew out of efforts to build a sense-and-avoid system into the Vahana demonstrator eVTOL aircraft, pictured here. (Airbus)

Airbus is expanding its foray into new autonomy for aircraft, hoping to use artificial intelligence and machine learning to build certifiable systems for both urban air mobility and — eventually — single-pilot operations in commercial aircraft.

The project, now called ‘Wayfinder’ and housed under Airbus’ A³ unit in Silicon Valley, began as part of the unmanned Vahana eVTOL development project, according to Cedric Cocaud, chief engineer at Wayfinder. Initially, the goal was to leverage technologies not traditionally used in aerospace to bring awareness and decision-making capabilities to the Vahana demonstrator aircraft.

“We ended up working on machine learning algorithms using cameras, and the use case that we had was when the Vahana aircraft takes off and starts flying straight, it’s about detecting a drone in front of it and providing an avoidance path around that drone,” Cocaud told Avionics International. “The initial focus was really about how far can we see the drone, how robustly can we detect it, and then can we provide an avoidance solution in time and keep a safe distance with respect to the drone.”

In addition to building the beginnings of an airborne detect-and-avoid system, Cocaud’s team taught the aircraft to interpret the ground beneath it to recognize safe landing locations. However, according to Zach Lovering, vice president of urban air mobility systems at Airbus, these capabilities were not actively tested during the Vahana’s flight test regimen, which was successfully completely earlier this month.

“The vehicle carries a sense and avoid system which is actively collecting data to enable real-time generated trajectory deviations to avoid obstacles in flight,” Lovering told Avionics in an interview earlier this summer. “However, it will remain a passive data collection payload while the vehicle performance envelope is being evaluation.”

As the Vahana team presented their work on autonomy systems to the rest of the company, it became clear to Airbus executives that the technology was relevant for more than just the Vahana — it was actually applicable to the whole line of Airbus products.

“We decided to create our own group, about a year ago, that has this unique expertise in Airbus on machine learning for perception and decision-making,” Cocaud said. “We shifted our attention from purely focusing on UAM to focusing both on air taxis as well as single-pilot operations, meaning autonomous commercial aircraft.”

Since the rise of large-scale commercial aviation in the 1950s, requirements for the number of technical crew-members aboard an aircraft have fallen from five to two qualified pilots for short-haul flights. Spurred by both carrot and stick — the promise of cost reduction and the looming pilot shortage — the industry is exploring what systems would be necessary to enable single-pilot operations without impacting safety.

There is, of course, significant resistance to this effort. The Air Line Pilots Association (ALPA) maintains that the most important safety assets on a passenger or cargo airliner are “at least two adequately rested, fully qualified, and well-trained pilots.”

“While ALPA supports new technology, serious questions remain unanswered regarding remotely piloted flight operations, including the latency — or delay — that occurs during electronic communications,” ALPA told Avionics in an emailed statement. “Currently, technology cannot adequately replicate or report the sensory information — sounds, smells, and vibrations — a flight crew depends on to safely operate a plane in real-world conditions. In addition, during flight operations, it is important for the flight crew to have the ability to interact visually with one another — something a remote pilot cannot do.”

Funding for the FAA to study single pilot operations was initially floated as part of the FAA Reauthorization Act in 2018, but was stricken from the final bill passed by Congress and signed by the president, largely due to resistance from pilots' organizations.

"The FAA is aware that manufacturers are researching autonomous aircraft and single-pilot aircraft for commercial use," an FAA representative told Avionics in a statement. "The knowledge gained from the Integration Pilot Program (IPP) with small unmanned aircraft systems (UAS) for use in delivery, and the anticipated projects with Urban Air Mobility (UAM), will assist regulators with the challenges posed by aircraft designed for other than the current two-pilot crew. We look forward to the innovations and working with applicants to integrate these designs into the aerospace system."

ALPA also listed to Avionics many of the functions pilots perform — “They interact with air traffic control, communicate with dispatch, check weather patterns and trends, visually scan for other aircraft, and monitor the performance of multiple engines” — some of which could be replaced by autonomous systems, such as a detect-and-avoid system produced by Wayfinder that could forego the need for pilots to visually scan for other aircraft.

But other functions of a pilot require a human in the loop as a result of how the larger system is designed. How would an autonomous system interact with air traffic control or communicate with dispatch?

Industry Analysts

Garmin’s Autoland avionics system, currently undergoing certification for general aviation aircraft, demonstrates that it is possible to automate communication with ATC and other pilots, but Cocaud acknowledges that this may make introducing greater autonomy into the commercial system more difficult than for UAM, in which case there is no pre-existing system that might need to be redesigned.

“We have most of the autonomy to be able to make [commercial aircraft] autonomous,” Cocaud said. “There’s still some bricks, like perception and decision-making, that we’re working on. But the key thing is that because we have all of that legacy infrastructure — because of how ATC works and how the airports work — how are they going to change the way they perform those tasks to be able to operate these very different aircraft?”

“Some of the functions that the pilot used to do are going to be shifted to ground infrastructure, for instance,” he added, discussing an example system architecture with a “mission manager” on board the aircraft rather than a pilot.

Single-pilot commercial operations are many years — perhaps decades — away, of course. For now, Wayfinder is focused on perception and decision-making, using sensory inputs and machine learning to make an aircraft capable of detecting an airborne threat and taking the appropriate action if the rest of the air traffic system fails.

“In essence, this is the problem that Wayfinder is really tackling,” Cocaud said. “When all those layers of protection that minimize interaction with other aircraft and air activities — the preflight check, establishing a flight plan, tactical deconfliction and in-flight rerouting — when all of those systems fail and the aircraft is on its own, how does it understand its surroundings and make the right decision?”