Intelligent tech takes on drone safety

“The year 2022 will be a landmark year for drones in Japan,” says Takeshi Tsuchiya, a professor of aerospace engineering at The University of Tokyo. Japan, long known for strict regulations of drones, is shifting gears.

By December 2022, the government plans to end its ban on 'Level 4' flight, which is when drones fly above populated areas without visual observers who physically watch for potential hazards.

In anticipation, Tsuchiya and colleagues have been developing artificial intelligence (AI) that will recognize the presence of people under a drone’s planned path and take measures to avoid flying over them.

A much-anticipated development with Level 4 flight is the transport of goods with drones, including package delivery in cities and carrying supplies to remote mountains ranges or disaster sites, explains Tsuchiya. Present regulations do not make the transport of goods realistic, as it requires visual observers to be placed under a drone’s flight route if there is the possibility of it flying over people, or if the pilot is operating the drone out of their field of vision. But the changes mean now is a transformative time, says Tsuchiya.

AI on the lookout

There are still safety concerns, however. “Delivery drones have nearly crashed on people, and emergency measures like parachute deployment have also failed in other parts of the world,” says Akiya Kamimura, a robotics researcher at the National Institute of Advanced Industrial Science and Technology (AIST).

Tsuchiya, Kamimura and their colleagues have been developing AI as part of a three-phase project aimed at developing next-generation drones that fly safely above cities with minimal supervision. Using their current AI, drones start a flight with a pre-determined route, but when people or vehicles appear underneath the planned route, drones recognize them using advanced image recognition and plan alternative routes in real time, taking into account potential points at which the drone could fall.

Ensuring drones with payloads don’t crash demands aircraft stability for the whole flight. “It’s challenging because drones need to auto-calibrate balance based on factors like the weight of the payloads, the change in the centre of gravity as items shift during flight, and unexpected gushes of wind,” explains Kamimura. The team devised new controls for drones so that their profile remains stable in the face of changing parameters.

The future of the skies

Three demonstration flights have tested the drones’ capacity to avoid people in the contexts of disaster relief, safety inspections on bridges, and parcel delivery. “Now, drones are loaded with 4G’s Long-Term Evolution (LTE) communication systems so that their condition, and the videos they take, can be monitored in real time. We monitored our first demo in Okinawa from nearly 1,600 kilometres away in Tsukuba,” says Kamimura. This was aided by a policy change in 2020, which simplified requirements for obtaining permission to use LTE networks while in flight.

The current research concludes the first part of the three-phase project. The next two milestones are for drones to auto-detect early signs of mechanical failure, and to auto-select safe emergency landing areas.

“We envision a future where swarms of drones fly in the same airspace,” adds Kamimura. “More changes to regulations are coming.”