It has been almost a decade since Jeff Bezos announced that Amazon Prime members would be able to receive orders in an hour or less with drone technology. Unfortunately, the idea of drone enabled deliveries was far more complex than Bezos anticipated. However, the Federal Aviation Administration (FAA) has been working tirelessly with drone manufacturers and developers to decrease the restrictions on drones in shared airspace. Since the outbreak of the global pandemic, COVID19, the feasibility of drone deliveries has become more apparent than ever.
So as to maintain logistics with social distancing and quarantine requirements, drones became critical tools for ensuring people in remote and populous regions had access to necessities like medications, PPEs, emergency supplies, and vaccines. Soon, drone deliveries of nonessential items like books, electronics, burgers, and coffee orders were being approved in select communities. These drone delivery programs are mostly still in trial stages, but we are getting closer to the reality of mass drone delivery operations. One caveat holding back drone deliveries is something uncontrollable, the weather.
The majority of drones, especially those that would be for parcel delivery, cannot fly unless the weather is calm and clear. Many drones are water-resistant, but add a bit of wind to the equation, and even a medium sized commercial drone could be unsafe in the air. Windy days happen all over the world, and in certain regions, drone deliveries would be impossible to schedule. Logistics companies cannot wait for windless days to plan on making deliveries. If they are to use drones, the drones need to be able to fly every day, regardless of wind conditions.
A team of researchers from the California Institute of Technology (CalTech) in Pasadena, CA, have developed an AI software program called Neural-Fly that should allow drones to fly through unpredictable windy conditions. Bren Professor of Aerospace and Control and Dynamical Systems at the CalTech Jet Propulsion Laboratory, Soon-Jo Chung, explained that the team needed to find a way to teach a drone to right itself, much like how a pilot would in a manned aircraft when traveling through wind variations. “The issue is that the direct and specific effect of various wind conditions on aircraft dynamics, performance, and stability cannot be accurately characterized as a simple mathematical model,” Chung says. “Rather than try to qualify and quantify each and every effect of turbulent and unpredictable wind conditions we often experience in air travel, we instead employ a combined approach of deep learning and adaptive control that allows the aircraft to learn from previous experiences and adapt to new conditions on the fly with stability and robustness guarantees.”
Neural-Fly separates parameters in real time to adjust for wind conditions without having to overprocess unquantifiable data scenarios. To test the system, the team applied Neural-Fly to drones in CalTech’s Center for Autonomous Systems and Technologies Real Weather Wind Tunnel. The Tunnel is custom built with more than 1,200 small fans capable of simulating a light breeze, gusts, and even gale force winds. With only 12 minutes worth of flight time, the drones with Neural-Fly were able to learn how to make linear adjustments to compensate for different wind effects. Neural-Fly enables drones to fly with an error rate of 2.5-4 times smaller than those with conventional aerodynamic response systems. It’s a start, one that could soon lead to broader drone delivery operations.