Is there a future for autonomous flight?
Believe it or not, flight automation technology has been around since the 1930s in the form of aircraft autopilot, allowing pilots to maintain a specified airspeed, altitude, and heading. Yet developing safe, scalable, fully autonomous flight technology for large-scale aircraft is a major challenge that will be difficult to solve.
Even so, at JetBlue Technology Ventures (JTV), we think artificial intelligence (AI) and specific use cases for autonomous flight technology have a real future in supporting the aviation industry’s growth.
The safety hurdle
True autonomy must manage the entire flight, including; taxi, takeoff, and landing; navigating from origin to destination (often through multiple waypoints); communicating with air traffic control, dispatch, and other aircraft; and safely navigating unexpected circumstances such as bad weather, equipment malfunctions, or onboard emergencies. Clearly, this is a complex problem!
Later-stage startups like Skydio or Zipline have leveraged advances in low-cost, powerful sensors, cameras, and computers to develop small drones with autonomous capabilities. These systems have near superhuman capabilities and are finding traction in applications like wind turbine inspection, package delivery, and military reconnaissance. But the technology is still a long way from being suitable for commercial aviation applications. This is why regulators still impose significant restrictions on most uses of autonomous drones, such as size and altitude limits, “no-fly” zones, and, in most cases, a requirement for the aircraft to remain within a human operator or chase plane’s line of sight.
Autopilots have some minimally-autonomous functionality, but they are more like the cruise control on modern cars. Automated voice warning systems exist on most commercial and military aircraft today too, but these are ultimately just alarms, not AI systems. Truly autonomous flight is a long way off.
Money and miles
So, why is it so difficult to build an AI system to fly safely? The short answer: money and miles.
Before aviation applications, consider the case of autonomous vehicles (AV). The key to developing self-driving cars is accumulating real-world road miles to train and verify the AI algorithms. According to the California Department of Motor Vehicles, AV companies drove more than 4 million miles in the state in 2021. Even with all that effort, AVs are not ready to roll solo on public roads, and autonomous flying is even trickier.
As one can imagine, training an AI system to fly an airplane is more difficult than training one to drive a car. Flying has a much smaller margin for error than driving, and the consequences of errors can be much more severe.
For instance, a self-driving car can pull over on the side of the road if it encounters a problem, or just stop and wait when it doesn’t know what to do, but aircraft don’t have these options. Testing autonomous flight technology in the real world is a tough sell because safety must come first. Paradoxically, it costs much more per mile to fly than to drive and there are fewer opportunities to learn per mile, making a dedicated AI flight training program prohibitively expensive.
Another challenge is that AI systems are probabilistic, based on giant statistical models, not deterministic, where a specific combination of inputs has only one possible result, and they sometimes produce very different outputs for slightly different inputs. This makes it hard to prove to a regulator like the FAA that your system is going to behave in predictable ways in otherwise unpredictable circumstances. This is why some autonomous and teleoperation-focused flight tech companies are focusing on military drones and crewless cargo flights, where they are attempting to persuade the relevant authorities that there can be a bit more tolerance for risk.
A place for autonomous flight technology
We’ve discussed some of the challenges that hamper progress to using artificial intelligence for fully autonomous flight, but there are more limited applications for the technology that could be hugely helpful to pilots and the entire aviation industry.
Pilots do much more than just fly the plane: they have full responsibility for the aircraft, the crew, and the passengers. Pilots have so much work to do that most commercial flights require two people in the cockpit, and an ultra-long distance flight such as from New York City to Singapore requires two full crews (four total pilots) so that everyone can rest and stay alert.
AI assistants that advise and support human pilots would be a smart path forward for the commercial airline industry. This could take a number of forms. For example, a contextually-aware AI assistant could remind the pilot to shut down one engine for taxiing to reduce fuel use on the ground, or help run through a pre-flight checklist. A pilot could use AI-based natural language processing to communicate with the plane in an intuitive way–like a cockpit version of Siri or Alexa. Or an airline could use an AI system to suggest an optimal altitude and route in real time for weather conditions, fuel efficiency, and other key factors without adding to the pilots’ already heavy workload.
Autonomous flight technology could also support continued growth during a workforce pinch as airlines work to train and hire new pilots as they strive to keep up with growing demand.
Fully autonomous flight in the commercial airline industry today is a long way off at best. Instead, we think the hybrid model is the way to go: human pilots with an AI assistant that can serve as a coach and safety net. Think Luke Skywalker and R2-D2 from Star Wars. The U.S. military is already experimenting with this concept, and there are a number of commercially-focused startups in the field as well, including JetBlue Technology Ventures’ portfolio company Beacon AI. So the message to tech companies is to focus on helping pilots, not replacing them. Even without exact numbers about pilots and flights and how AI copilots could spur growth, the market opportunity is clearly massive.
