On Sunday September 6, a Boeing 767 Amazon Prime Air flight operated by “21 Air” landed at Miami International Airport, overran Runway 30 and struck two ground vehicles. It killed five airport workers sitting in a van on a perimeter road. The two pilots aboard the aircraft survived with minor injuries.
The tragedy, which deserves the usual investigation, accountability and respect for the victims, remains ongoing. But this accident raises an interesting question for those who believe that it is time that we start thinking about removing pilots from crewed aviation in aircraft that are increasingly capable of flying, navigating and landing with extraordinary levels of automation. It’s not about whether pilots are good or bad, but rather a question about whether the traditional human-centered architecture of aviation still makes sense for every category of flight, particularly emerging aviation like electric vertical takeoff and landing (eVTOL) aircraft.
What Went Horribly Wrong…and Why?
Even though the NTSB has not yet determined the probable cause, its preliminary report (NTSB), which contains an analysis of the cockpit voice recorder (CVR), indicates a serious break of procedures and incorrect use of Crew Resource Management (CRM). Preliminary recorder data show the aircraft touched down at 158 knots groundspeed. The maximum authorized by the operating manual is 145 mph. Braking began at 146 knots. The pilots initiated a go-around after touchdown but aborted without an explanation. The data also shows no deployment of thrust reversers.
What happened on September 6 at MIA is a disgrace to good pilots around the world doing their job efficiently and safely. The cockpit voice recorder is unequivocal that one pilot repeatedly expressed concern that the aircraft was coming in too fast. The crew had also been operating in strong winds associated with thunderstorms in the area and that required cabin coordination and precision in the approach. In pilot circles, we informally refer to a pilot’s dangerous determination to reach a destination despite worsening weather, mechanical problems or other clear safety issues as “get-there-itis.”
How could this have happened? As a pilot, I can tell you that today all my colleagues around the world flying under instrument flight rules (IFR) are increasingly becoming monitors of systems, not handlers of controls. From the moment a plane takes off on an IFR flight plan and as soon as the gear and flaps are retracted, we engage the autopilot and become button pushers and knob twisters. The air traffic control (ATC) system tells us what to do, when to climb and to what altitude, what heading to follow and what waypoints to go to. We retake control of the aircraft just a few hundred feet above our destination airport, generally just a mile away or less.
Automation has progressed to the point that today a number of manufacturers of general aviation aircraft, including Cirrus, Piper, Daher, Embraer and Honda Aircraft, even offer a feature that allows passengers to push a large red button if the pilot becomes incapacitated and the plane finds the closest airport, changes transponder code to emergency (7700), communicates with ATC and lands the plane. It’s called Autoland.
Even so, none of this establishes that automation would have prevented the accident. It does, though, illustrate the problem that aviation automation has been steadily designed to solve: humans operating complex systems under time pressure, changing conditions and erratic information.
Cargo Is Where Pilotless Aviation Makes The Most Sense
The strongest case for autonomous aviation begins with cargo. A cargo aircraft doesn’t need a flight attendant, doesn’t have passengers that complain about turbulence or even require a human being to physically sit at the controls of the aircraft throughout the flight. As such, cargo aviation is an obvious environment in which to separate the aircraft from the human operator.
Instead of two pilots physically occupying the cockpit, an autonomous cargo aircraft could potentially be supervised by personnel on the ground, with humans intervening, when necessary, rather than manually controlling every phase of the flight. That distinction is fundamental. Today’s aviation system largely assumes that humans should be inside the aircraft and that automation should assist these humans. The emerging autonomous model turns that architecture around.
One of the largest and most prolific manufacturers of helicopters in the world, Robinson Helicopters has launched a version of its most popular model, the R44 as a non-piloted aircraft called the R44 Airtruck designed to carry over 1,100 lbs of cargo. Joby Aviation recently completed a transcontinental flight using a Cessna Caravan without input controls from a pilot, in a completely autonomous 3,199 mile flight across the continental United States.
The technology is increasingly here and is designed to finally replace the increasingly sophisticated decades-in-the-making systems specifically to reduce the number of decisions that pilots have to make manually. The difference between a pilot flying a DC-3 in 1940 and the equivalent today piloting an Airbus 380 is astronomical. Yet when something goes wrong, the final responsibility still comes down to the two people sitting in the cockpit. Miami illustrates the vulnerability of that arrangement.
The NTSB’s preliminary evidence shows a cockpit in which speed was a concern. Yet a landing continued, a late go-around was attempted and the aircraft ultimately departed the runway. As investigators are still determining how those factors interacted, it would be irresponsible to declare that a particular human action caused the accident. My point is that autonomous systems approach the problem differently.
An autonomous aircraft does not get tired. It does not become distracted. It does not have to reconcile two humans’ interpretations of an unstable approach. It can continuously compare position, airspeed, wind, runway remaining, descent rate and aircraft performance against predetermined safety envelopes. And perhaps most importantly, an autonomous system can be programmed to make the decision that humans sometimes find difficult: don’t land! Go around.
Autonomous aviation may not be in a hurry to get home, but it does introduce its own risks. Among these software failures, sensor failures, cybersecurity, communications loss, unexpected edge cases and the possibility that an automated system will behave incorrectly in circumstances its designers did not anticipate rank pretty high. Those risks still have to be engineered, tested and regulated.
Ultimately, there is a critical difference between saying autonomous aviation is risk-free and asking whether it can ultimately be safer than putting a human crew inside every cargo aircraft. The aviation industry has already demonstrated that automation can make flying dramatically safer. The next step is not necessarily eliminating humans from aviation. It may be moving humans to where they can do the most good.
Miami Should Accelerate The Conversation
An airport is an extraordinarily complex environment in which aircraft, fuel trucks, baggage equipment, maintenance crews, cleaning crews, ground vehicles and thousands of people operate in close proximity to aircraft weighing hundreds of thousands of pounds. Every time an aircraft enters that environment, the consequences of a mistake extend beyond the cockpit, regardless of the presence of human pilots on board. The fact that the five people killed at Miami were not part of the flight crew and were simply doing routine work on the ground should haunt the aviation industry. Autonomous cargo operations offer the possibility of changing that equation.
Imagine a future in which cargo aircraft arrive, navigate and land autonomously. Ground personnel can be kept clear through integrated traffic-management systems. Automated aircraft will continuously monitor their own performance. Human operators can supervise multiple aircraft from a secure control center.
To be clear, this type of technology ecosystem will not eliminate every accident or every human decision. But it can reduce the risks of an accident. We know this based on the preliminary Miami evidence. An autonomous aircraft may not have avoided the September 6 accident, but at least it would have flown the approach would have been flown according to specifications and procedures.