In the early hours of January 11, 2000, US Coast Guard helicopter pilot Mark Ward responded to a distress call from a ship taking on water, caught in a Nor’easter off the North Carolina coast. Battling 70-mph winds and 30-foot seas, Ward struggled to keep the chopper steady as he and his crew pulled all five fishermen to safety.

Ward recalls the mission as one of the most harrowing is the 22 years he spent as a search-and-rescue pilot. And now, he’s got a gig ensuring his successors won’t face the same dangers: He’s the chief test pilot in Sikorsky’s autonomous helicopter program. “Even a modest degree of autonomy, your workload goes way down and your stress and apprehension disappears,” he says. “The system sees things you can’t, and it processes information and reacts in a way you may not be able to.”

Even in a world where planes spend most of their time on autopilot and robo-cars are roaming cities all over the world, teaching a helicopter to fly itself is a gnarly problem. These workhorses must be able to hover over ships bobbing up and down on rough seas, and descend onto oil rigs in gusting winds. They have to dodge power lines and cell towers that may not show up on navigation charts, and balance single skids on sheer cliffs in order to rescue injured climbers.

“Helicopters have very high crew workloads and obstacle-rich environments,” says Chris Van Buiten, vice president of Sikorsky Innovations, the division of the Lockheed Martin-owned company that’s pursuing autonomous flight. A robo-chopper needs a lot more computation than a self-flying plane, he says, especially since the flights they take on don’t involve cruising between well-regulated airports. “You’re usually not called out to a sinking ship on a sunny day, but rather off the coast of Alaska at night in the rain,”

LEARN MORE

The WIRED Guide to Drones

The aviation industry is already deep into the challenge. In May, Boeing-owned Aurora Flight Sciences’s unmanned cargo delivery system, installed in an old Bell UH-1H helicopter, completed the first autonomous mission, bringing gas, water, and medical supplies to Marines in California. Lockheed Martin has been developing its K-MAX unmanned helicopter since 2007, beginning with remote-controlled and semi-autonomous versions that made supply deliveries in Afghanistan between 2011 and 2014.

Sikorsky’s version is the Matrix Technology system, which it’s been testing since 2013 aboard the Sikorsky Autonomy Research Aircraft (SARA) testbed that Ward pilots, an adapted version of the company’s S-76 commercial helicopter. Its most basic functionality includes flying traffic patterns around airports and tracking moving objects on the water for approaches and landings.

More impressively, SARA has completed a 30-mile autonomous flight with takeoff, cruise, and landing—including landing-site evaluation and selection—all done by computer. That was enough to get it to the final phase of Darpa’s Aircrew Labor In-Cockpit Automation System (ALIAS) program, which seeks a system that will reduce crew requirements for existing aircraft. The company is also in the process of modifying two UH-60 Black Hawk helicopters with Matrix, to offer the Army “optionally-piloted” options for the aircraft. It will demonstrate these over the coming year.

Edging Toward Autonomy

The end-game for most of the companies pursuing helicopter autonomy is fully hands-off flight with human passengers, not just cargo. This will be key to the nascent air-taxi industry, and for military and commercial operators who may be facing pilot shortages. But it’s also the most demanding possibility, given the challenges of validating and certifying such systems to actually carry people on board.

“When we decided to go after this, the problem became reliability and safety,” says Igor Cherepinsky, Sikorsky’s director of autonomy programs. “We decided that if we’re going to do this, it needs to be just as safe as our other aircraft. That’s our guiding principle.”

That led to some counterintuitive strategies, like minimizing the role of artificial intelligence. “High-end artificial intelligence and deep learning are higher-order functions,” says Van Buiten. “Higher-order functions are difficult to certify. Until we know how to do so, we want to use more deterministic methods.”

That means using systems that don’t rely on interpretation or guesswork—which AI is essentially an advanced form of—but on defined and predictable behaviors. Cherepinsky adds that this is true across the board, from developing responses for when things go “off plan,” to applying computer vision data from the optical sensors. “Even our pattern recognition is done in a different algorithmic way. It’s very reliable and very flyable,” he says.

And where self-driving cars rely on high-definition maps of any environment they’ll explore, Sikorsky skipped the cartography and trained its aircraft to fly using only their real-time sensors.“There have been quite a few accidents where aircraft hit things that weren’t on their maps,” Cherepinsky says. “Companies are notorious for throwing up cell towers without notifying anyone, for example.”

Evaluating how all these elements and algorithms function in the air falls to Ward. On test flights, he evaluates the system’s decision-making to help fine-tune the flying, and streamlining the user experience (and stays ready to take control if needed). “We want a few taps on the tablet to replace ten minutes of playing around with the flight management system of a conventional helicopter,” he says.

On that rescue mission in 2000, once Ward got the helicopter in position to hover over the endangered boat, he had to keep working the stick, levers, and pedals to hold its position against the wind. With the Matrix’s level of automation, it’s a matter monitoring the system, making alterations such as position adjustments via slight nudges on a virtual joystick on the tablet. Sikorsky’s decision to develop the tech in-house speeds up the development process, allowing Ward to make recommendations about things like the placement or prominence of the tablet controls—and see the changes a few days, or even minutes, later.

The Human Touch

Full autonomy—the kind where no human pilot is required—will take longer to achieve, but these interim stages could pay huge dividends by simplifying a pilot’s work. “Just tracking alongside a vessel in a storm at sea is intensely challenging, but an autonomous system locks on, managing your airspeed, altitude and position even in the worst conditions,” Ward says, adding that many accidents result from pilots being overloaded during such scenarios and losing situational awareness. “When your stress level goes down, your situational awareness goes up, and you’re better able to focus on your crew and the mission.”

Indeed, full autonomy may not be appropriate for many of the missions helicopters fly. “There are lots of discussions about autonomy versus automation,” Cherepinsky notes. Humans can always use help, but it may not be wise to replace them altogether. “A machine cannot find its own mission. Creative humans do that—they plan them, decide what the machines do, choose who gets priority in rescues, and so on. Think of the Starship Enterprise. Five or six people on the bridge are making the decisions, but the machine actually takes the ship from point A to point B.”

And if you ever find yourself stranded at sea, you’ll probably be far happier to see the fully focused equivalent of Captain Kirk managing the situation when the helo arrives, with Scotty cheerfully beaming you up. Let the computer deal with the wind.


More Great WIRED Stories