OSAKA Japanese researchers have developed a real-time drone-locating system that uses an onboard laser sensing device and artificial intelligence that enable it to operate even where GPS fails.
The technology developed by Kyoto University researchers led by Masatoshi Hamanaka allows a drone in flight to determine its position and altitude based on geographical features beneath it, detected by a laser-based sensing device and analyzed using AI technology.
Under the system, data on the geographical features from the onboard sensor is compared with existing data in a geographical database to determine the drone's position.
The technology has a margin of error of a few dozen meters, but this is a practical level for drones that fly at 30-40kph, the researchers said.
They aim to raise the accuracy of the system and make it viable for practical use within three years.
The accuracy of GPS, the current go-to location technology for drones, can drop considerably when the vehicle is flying between high-rise buildings and in mountains where signals from satellites tend to be randomly deflected or blocked. The new system was developed to address this problem so a drone can remain on the right path in such areas.
The onboard sensing device uses a laser to ascertain the geographical features beneath it, and then uses AI technology called deep learning to match the data to an existing geographical database from satellites and aircraft. The system divides the surrounding space into imaginary cubes 30 meters on a side and identifies the one that currently contains the drone.
This all happens in an instant, allowing real-time applications in, for example, self-flying drones carrying cargo and those used in land surveys, the researchers said.
Before use, the deep learning software is made to "learn" to match given sets of geographical data with existing data.
As the technology requires considerable computing power, the system uses a high-performance central processing unit, the heart of a computer, enabling it to name the location in real time.
The researchers have yet to test the system on an actual drone in flight, but in a computer simulation, it identified locations with more than 90% accuracy instantaneously, they said.