A drone developed by researchers at the California Institute of Technology (Caltech) can withstand strong gusts that would render most existing versions inoperable.
Most drones can only function effectively in favorable weather. However, drones must be able to adapt to wind conditions in real time in order to carry out important tasks.
An answer was discovered by a group of Caltech engineers. They developed “Neural-Fly,” a deep-learning technique that allows drones to adapt to novel and unexpected wind conditions and keep flying even during catastrophic events like hurricanes.
Hurricane Resistant Drone
According to a study published in Science Robotics, the five-pound drone can recalculate meteorological conditions around it five times a second and smoothly adjust its course.
The drone was put through its paces at Caltech’s Center for Autonomous Systems and Technologies. This facility features a special 10-by-10-foot array of more than 1,200 miniature computer-controlled fans, allowing researchers to mimic anything from a mild wind to a storm.
E&T reported that during the presentation, despite being battered by winds of up to 12.1 meters per second, the drone performed a series of figure eight maneuvers between small objects without losing its balance.
The AI-powered drone has the potential to innovate various sectors from medical emergency services to air travel, according to the team.
The engineers are also developing many types of drones. One looks more like an airplane than the other, but the one using the same software to reach medical situations is an autonomous flying ambulance. Within a year, the team hopes to be able to develop this design in a year.
Also read: “Flying hotel that never lands”: AI Sky Cruise Ship will use nuclear power to fly in luxury
All about the neural fly
The neural network can be pre-trained by scientists using a meta-learning technique, so only a few key parameters need to be updated to accurately adapt to the changing environment.
The drones equipped with Neural-Fly learned to react to strong winds so successfully that their performance could be significantly increased after only 12 minutes of flight data.
Compared to drones currently on the market that do not use neural networks, studies so far have shown that the error rate is 2.5 to 4 times lower, according to E&T.
The group has also demonstrated how flight data collected by one drone can be shared while another drone is responsible for creating a database of information for autonomous vehicles.
Caltech engineers claim that since the drone is just an improved version of a standard drone and not an entirely new design, the system can be easily marketed and expanded to various robots and aircraft.
The research, published in the international journal Science Robotics, heralds a new technological era that could make drones more practical and widely accessible.
Related article: New AI that predicts heart attacks could save patients from cardiac death for 10 years
This article is owned by Tech Times
Written by Joaquin Victor Tacla
ⓒ 2022 TECHTIMES.com All rights reserved. Do not reproduce without permission.