Why are researchers teaching this AI to fly a plane

In February, for the first time, a The Black Hawk helicopter flew by itself With no humans on board. The military self-flying helicopter project involved DARPA and Sikorsky, making the UH-60 helicopters.

Meanwhile, in some places, such as companies Zipline And the wing The goods are delivered by drones. Other companies working on electricity air taxi to me Transportation of people or goodsAnd, of course, regular air traffic — commercial flights from big airports, general aviation planes flying out from other airports — fly around, too. Factor in helicopters, hot air balloons, and more, and there can be a lot going on out there.

With all this busy in the sky, researchers at Carnegie Mellon University working on it An experimental AI system that can perform tasks such as predicting what another aircraft might do, or monitoring nearby aircraft using onboard cameras. The idea is that such AI could help fly drones, assist a human pilot, or even someday fly a plane alone.

Right now, in a flight simulator, AI can tell what another plane is He does or may do, then figure out how to safely land the plane accordingly. Think about the way the driver behind the wheel notices another car approaching an intersection, and begins proactively planning what to do if the other car is going to run at a stop sign, for example.

In this case, the AI ​​is looking for another plane, not a car of course. “He basically looks at their behavior for 10 seconds,” he says. Jay Patrikar, a doctoral student at the Institute of Robotics at Carnegie Mellon University. He’s trying to judge: ‘They’re here. What are they likely to do? ”

In that sense, it’s like artificial intelligence can play chesssays Patrikar, thinks about what step he would have in advance if his opponent took a certain action.

[Related: The Air Force plans to test an AI copilot on its cargo planes]

artificial intelligence systems need data to learn from. In this case, the team is collecting data from two real airports, both in Pennsylvania. One has an air traffic control tower, the other does not. At those airports, Patrikar says, the data they’re hiding includes visual information from cameras on a hanger or near the taxi aisle, spoken communications from radios, weather data, and more. “We record it completely,” he says. The idea is that the AI ​​can learn cause and effect by paying attention to all this information.

“She knows the causality of things,” he adds. This means that the AI ​​can learn, for example, that “it was caused by the weather in which they live.” [a pilot] She decided to do this very thing.” Patrikar says the training the AI ​​received in these scenarios helped her learn how to navigate the landing process in the simulator.

In addition, an AI that brings an aircraft to land at a small, unsupervised airport must follow FAA rules as well as other rules when interacting with other aircraft, Patrikar points out. “One of the ways humans trust each other is through our shared understanding of rules – our social norms,” ​​he says. People on a crowded sidewalk might decide how to pass each other by going right, for example, and rules like this apply in aviation that an AI pilot must follow.

[Related: This company is retrofitting airplanes to fly on missions with no pilots]

Relevant work in the real world, rather than in simulation, had the team placing cameras on an aircraft such as a Cessna 172 or a Hexacopter drone. These cameras and artificial intelligence are able to monitor, identify and know how far away other aircraft are in the area with an accuracy rate of greater than 90 percent at a distance of 700 meters (about 2,300 feet). This type of technology could help a human pilot in a small aircraft visually detect other traffic in the area. “I’d like to have this system on my plane,” says Patrikar, who holds a private pilot’s license. After all, artificial intelligence does not blink.

To be sure, Carnegie Mellon researchers aren’t the only ones exploring the new frontiers of AI that can fly or help fly. Zipline drone company works on Use microphones on its drones to listen to other aircraft in the area and then take evasive actions to avoid any potential collisions. In particular, the company is called Merlin Labs It has also developed a digital pilot that can replace the human co-pilot. As one example, it works with Air Force to equip C-130J cargo aircraft With their system, so instead of a human crew of two pilots, the plane could be piloted by one person with an artificial co-pilot.

Leave a Comment