AI and Autonomous Flight

Autonomous flight is one of the most popular and controversial topics in aviation. While the industry is working toward fully autonomous aircraft, many near-term efforts are focused not on replacing pilots on the flight deck, but on improving the overall safety of flight operations.

Artificial intelligence (AI), specifically machine learning, could be part of this process, though perhaps not how some may envision. Rather than AI making on-aircraft, flight-critical decisions, these systems are used to inform the development of proven algorithms governing those responses.

“AI is what you call an algorithm that doesn’t work yet; once it works, you name it,” said Robert Rose, CEO and co-founder of Reliable Robotics. “We don’t use AI on the airplane. We use working algorithms that will meet the FAA’s robust safety standards and existing regulations.”

Reliable Robotics is developing what Rose terms “continuous autopilot engagement” which handles all necessary flight tasks from engine start through shutdown. “We’re approaching this as an incremental process,” he said, “and what we believe will benefit our industry most quickly from a safety perspective. Making air transportation safer, more efficient and universally accessible is the ultimate goal.”

Zipline, which utilizes fixed-wing unmanned aircraft systems (UAS) to deliver medical supplies and other products as far as 50 miles in locations as varied as Arkansas, Japan and Africa, employs a similar philosophy.

“Whether you call it autonomy or automated, I fully believe you can have a static set of code, leveraging machine learning, operating an aircraft at a level of safety that meets or will likely exceed human performance in similar circumstances,” said Jon Hanlon, Zipline’s director of UAS maintenance & airworthiness. “That’s where we’re going.”

Safety is also at the forefront of advanced air mobility (AAM) provider Wisk’s plan to utilize autonomous, electric vertical takeoff and landing (eVTOL) air taxi vehicles over urban areas.

“AAM will be transporting people above densely populated urban environments, and any accident will be front page news,” said David Oord, Wisk’s policy manager. “We are committed to preventing that from happening. Safety is our North Star.”

One Pilot, Many Aircraft

While many AAM developers look to piloted vehicles as an interim step toward full autonomy, Wisk has conducted more than 1,600 test flights of five generations of eVTOLs at several locations, including a non-towered California airport and in New Zealand. The company announced its sixth generation, certification-conforming vehicle in October 2022.

“We have really committed to self-flying first,” Oord said. “Our approach is to offload some of the routine responsibilities that are traditionally held by a pilot to an autonomous system, so that a ground operator who might be supervising up to three aircraft is free to focus on high-level aeronautical decision-making.”

Reliable Robotics plans to initially utilize a remote pilot to supervise the flight with an onboard test pilot to monitor autonomous aircraft functions designed to handle all necessary flight tasks while also preventing common accident scenarios such as loss of control inflight (LOC-I) and controlled flight into terrain (CFIT).

“With a continuously engaged autopilot, there’s no envelope protection because the system keeps the aircraft within the flight envelope; the system is the envelope,” Rose explained. “Similarly, the system checks the terrain for you and refuses any flight plan that would cause you to intersect terrain.”

In May 2023, the company demonstrated these capabilities to U.S. Air Force officials during a readiness exercise in California. Reliable Robotics also operates a regional air cargo subsidiary flying conventional Cessna 208B Caravans in New Mexico, an environment Rose acknowledged is well-suited to future testing.

“These aircraft are single engine, and flown by a single pilot, in a lot of nighttime and instrument operations,” he said. “The FAA-certifiable automation system we are developing is going to make those operations much safer.” The company will apply these initial operational experiences to the FAA certification of a full remote piloting capability.

Zipline also uses human pilots to monitor aircraft operating with a high level of automation. In the U.S., one pilot may monitor up to six aircraft performing commercial operations, while overseas a single monitor may oversee more than 20 aircraft.
“The aircraft handles all exceptions by itself, with defined contingencies and responses,” Hanlon said. “Even the fault handling is highly automated. The monitoring pilot steps in when factors beyond the individual aircraft require action, for example, changes in weather.”

Certification Challenges

Autonomy also factors into many upcoming transportation modes that hinge on greater availability and access. “Future air travel will look very different from today,” Rose said. “Rather than large airliners carrying passengers from a few major hubs, I think it will become economically viable to have fleets of smaller airplanes and on-demand service from more convenient, regional airports. That isn’t scalable or economically viable without autonomy.

“I can appreciate those who resist this kind of thing, but it is happening regardless,” he added. “If it doesn’t happen in the United States, it is guaranteed to happen elsewhere.”

The most pressing challenge may not be in implementing such systems but in receiving approval from regulators to operate commercially. “The U.S. has always been the global leader in aviation, but the industry has outpaced the FAA for some time,” Oord said. “The industry must work with the FAA and provide them with the data so we can solve these challenges together and maintain leadership in autonomous flight.”

Rose drew parallels to his past work with commercial launch provider SpaceX and autonomous flight development.

“Our co-founder – Reliable CTO and former SpaceX engineer Juerg Frefel – once joked that if you can autonomously rendezvous and dock with an orbital body, clearly we could find the runway centerline,” Rose said. “There is no research problem that needs to be solved here. We possess the capability and the technology to do this.

“The challenge is moving from a prototype phase where you may accept a failure rate of one out of 100 missions, to something that is certifiable at one failure in 100 million hours,” he continued.

“People often complain about the regulatory environment, but technical and operational maturity are also factors,” Hanlon said. “We’re not changing the safety bar [but] we do need to prove these systems perform at the level where existing structural mitigations – low population density, separate airspace, etc. – aren’t required.

“Meeting most regulatory requirements [without a pilot onboard] are eminently solvable by mapping existing requirements to the system capability and remote pilot function,” he added. “We must demonstrate the ability to still meet the intent of the rule.”

“Autonomy is challenging, obviously,” Oord agreed. “The question becomes, how do we get the FAA to be comfortable in that space where it’s a little unknown, a little unproven, but it’s on us and industry to prove it, and we’re willing and able to do so.”

That said, “most people think of autonomous flight as something that will happen sometime around 2040, but we’re building off technologies that are already available and being used every day,” he continued.

“Aircraft fly CAT III approaches and landings completely hands-off; even some smaller general aviation aircraft can now land themselves in an emergency.”

Despite the challenges, autonomy remains the clear goal. “We’re only moving the needle further to evolve to the next level, with aircraft able to aviate, navigate and communicate entirely on their own,” Oord said.

Review NBAA’s advanced air mobility resources at nbaa.org/aam.