Search This Blog

Friday, June 9, 2017

Boeing Studies Planes Without Human Pilots, Plans Experiments Next Year

Boeing is researching the possibility of commercial-passenger jets that will rely on artificial intelligence rather than pilots. Initial experimental flights, without passengers, are planned next year, with such systems taking over some of the pilot decisions.
Boeing has begun researching the possibility of commercial-passenger jets that will fly without pilots, using artificial intelligence guiding automated controls to make decisions in flight.

“The basic building blocks of the technology are clearly available,” said Mike Sinnett, former chief systems engineer on the 787 Dreamliner and now vice president at Boeing responsible for innovative future technologies, at a briefing before the Paris Air Show.

“There’s going to be a transition from the requirement to have a skilled aviator operate the airplane to having a system that operates the vehicle autonomously, if we can do that with the same level of safety,” Sinnett said.

“That’s a really big if,” he added.

It sure is. Think about a machine that could do what US Airways Capt. Chesley Sullenberger did in New York City in 2009.

When a flock of geese took out both engines on an Airbus A320 with 155 people on board as it took off out of La Guardia, Sullenberger communicated with ground controllers, rapidly sized up his limited options within two minutes and guided the plane to a safe ditching in the Hudson River.

Sinnett, who plans a June 21 presentation on the subject at the Paris Air Show, agreed that the Sullenberger scenario is the standard that has to be achieved. It also underscores the challenge Boeing faces in attempting to take the human out of the flight deck.

“We are not smart enough to preprogram all those things. The machine has to be capable of making the same set of decisions,” Sinnett said. “If it can’t, we cannot go there.”

Sinnett said his team will fly a simulator this year with an artificial-intelligence system making some of the piloting decisions.

Next year, he said, they’ll fly the system on a real plane. Those would be experimental flights, with engineers and pilots on board, but no passengers.


Image result for Artificial intelligence cockpit

Go for zero

Wild as it sounds to consider a commercial jet flying without a pilot, the times are ripe for such thinking.

Sinnett said Boeing’s research is driven by the pilot shortage worldwide that is only going to become more acute.

In the next two decades, Boeing forecasts a demand for about 40,000 new commercial jets, roughly doubling the world fleet.

“Where will the experienced pilots come from?” Sinnett asked.

Meanwhile, small autonomous drones are flown by the military and are being tested by Amazon for package delivery. And the public increasingly accepts the notion of driverless cars navigating the public roads.

Yet Sinnett understands why it seems more radical to think of the same for a passenger jet. Last year 40,000 people died in road accidents in the U.S. — leaving lots of room for potential improvement by autonomous, driverless cars.

By contrast, Sinnett said, there were zero deaths in the U.S. last year on scheduled jet aircraft. To make autonomous aircraft as safe as flying commercial is today, “We’ve got to be as good as zero,” he said.


Image result for Artificial intelligence cockpit crash landing

Autopilot systems

Some of the technological building blocks of autonomous flight are already embedded in today’s aircraft.

On long flights, airline pilots will switch to autopilot as they cruise for hours.

What’s less well known is that commercial jets often auto-land, which is what makes landing possible in conditions of very low visibility due to weather.

The auto-land is the closest thing today to autonomous flight because the system reacts to changes in the environment as it comes in, adjusting for small changes in the winds.

Sinnett said that when he was developing the 787, the eighth landing the aircraft made was an auto-land without pilot input.

Auto-takeoff is not allowed, but today’s airplanes can do that too.

“If you want to end your career, you could take a 777 out and do an automatic takeoff,” Sinnett said. “The airplane is capable of doing it, but not capable at the same levels of integrity we have today. So we have pilots in the loop.”

The pilots always are expected to monitor the functions of the automated systems.

Because of the multiple redundant systems on aircraft, airplane accidents are almost always the result of a series of mishaps, any one of which would not alone have caused the accident.

So, if say, an autopilot does something unexpected, a crucial function of the pilot is to step in and catch that first piece of unintended behavior before the next step in any chain that could lead to disaster.



Artificial intelligence

Could a machine do the same?

If a passenger has a heart attack, will it divert?

If one engine goes out, will it know the best response given its position?

What about both engines?

Sinnett points out that a primary requirement for certification of commercial jets today is that the systems operate deterministically: given a set of inputs you must always get the same result.

But he said because no one is likely to be able to predict all the potential things that could happen during all phases of a flight anywhere in the world, an autonomous flying machine has to be able to respond non-deterministically — to react to a situation that has not been preprogrammed into the software.

“So we are doing early exploration with machine learning and artificial intelligence,” Sinnett said.

When safety regulators tell him that they cannot certify non-deterministic systems — as they have — he responds that yes they can, because they certify pilots.

Humans inevitably react differently to one another. An individual may even react differently to the same circumstances at different times.

Still, don’t expect pilotless passenger jets anytime soon.

Sinnett sees this problem of building a system — a machine — capable of intelligent, non-deterministic behavior as the toughest challenge.

“I have no idea how we’ll do that,” he said, with a laugh. “But we are studying it right now and developing those algorithms.”