Most aerial drones today are remote-controlled, but they may not have the precision required for certain mapping and inspection applications. Researchers at Aarhus University and the Technical University of Denmark said they have developed AI pilots to make measuring and documenting gravel and limestone quarries faster, cheaper, and easier.
“We’ve made the entire process completely automatic,” stated Erdal Kayacan, an associate professor and expert in artificial intelligence and drones at Aarhus University’s Department of Engineering. “We tell the drone where to start, and the width of the wall or rock face we want to photograph, and then it flies zig-zag all the way along and lands automatically.”
Measuring and documenting gravel and limestone quarries, cliff faces, and similar natural and man-made formations is often done using drones that photograph the area. The recordings are then uploaded to a computer that automatically converts everything into a 3D terrain model.
However, human pilots are costly, and the measurements are time-consuming because the drone has to be controlled manually to hold the same constant distance to the wall of an excavation, while simultaneously keeping the drone camera perpendicular to the wall.
Furthermore, there must be a specific overlap in the images taken, so that the computer can then “sew together” the images into a large 3D figure.
AI pilots offer precision
Researchers in the Department of Engineering at Aarhus University have now automated this process using AI pilots.
“Our algorithm ensures that the drone always keeps the same distance to the wall and that the camera constantly repositions itself perpendicular to the wall,” said Kayacan. “At the same time, our algorithm predicts the wind forces acting on the drone body.”
AI pilots could compensate for wind better than human pilots. Source: Erdal Kayacan, Aarhus University
The researchers have been able to compensate for one of the major challenges associated with autonomous drone flight: the wind.
“The designed Gaussian process model also predicts the wind to be encountered in the near future,” said Mohit Mehndiratta, a visiting Ph.D. student at Aarhus University. “This implies that the drone can get ready and take the corrective actions beforehand.”
Today, it takes little more than a light breeze to blow a drone off course, but with the help of Gaussian processes, the AI pilot team has taken into account gusts and the overall wind speed.
“The drone doesn’t actually measure the wind; it estimates the wind on the basis of input it receives as it moves,” Kayacan explained. “This means that the drone responds to the force of the wind, just like when we human beings correct our movements when we are exposed to a strong wind.”
The Aarhus research into AI pilots is being carried out in collaboration with the Danish Hydrocarbon Research and Technology Centre at the Technical University of Denmark. The team plans to present the results of the project at the European Control Conference in May 2020.
Tell Us What You Think!