It’s hard to imagine a drone doing what a flying squirrel does best: gliding, braking midair, and darting through forests with acrobatic grace. Yet that’s precisely the inspiration behind a new breed of aerial robot, equipped with flexible, foldable wings and a ‘brain’ powered by machine learning.
Engineers in South Korea have developed a drone that mimics this airborne behavior — a quadrotor equipped with foldable wing membranes that can suddenly slow down, execute sharp turns, and avoid obstacles in ways that traditional drones cannot.
The team behind the new design, a collaboration between Pohang University of Science and Technology and the Agency for Defense Development (ADD)’s AI Autonomy Technology Center, hopes this new design will help drones better navigate tight or unpredictable environments — whether in forests, disaster zones, or urban canyons.
From Forests to Flight Labs
Flying squirrels don’t actually fly — they glide. They stretch flaps of skin between their wrists and ankles, creating a kind of natural wingsuit. It’s an evolutionary design that allows them to steer through complex terrain and, crucially, to decelerate rapidly just before landing.
These features make flying squirrels extremely agile, which can’t be said about your typical quadcopter. The same features that make them stable — fixed rotors and rigid frames — also limit how sharply they can turn or respond to sudden obstacles.
To make quadcopters more squirrel-like, the South Korean researchers designed feather-light silicone wings — just 24 grams in weight — that can fold and unfold with servo motors.
But wings alone aren’t enough. The real magic lies in coordinating them.
That’s where the Thrust-Wing Coordination Control system, or TWCC, comes in. This framework constantly assesses whether deploying the wings would help or hinder the drone’s movement. When the onboard controller predicts that a maneuver would exceed the drone’s pitch or roll limits, it signals the wings to deploy, enhancing the available force without pushing the drone into instability.
<!– Tag ID: zmescience_300x250_InContent_3
–>
It uses an array of sensors—GNSS, barometers, inertial measurement units—to track its position and orientation. These feed into the TWCC algorithm, which decides in real time whether to fold or unfold the wings, helping the drone dart, dip, or brake as needed.
“The wings are spread, and thrust is adjusted… allowing [the drone] to generate a stronger force in the desired direction,” the team explained in their study, for now, published in the pre-print server arXiv.
The drone doesn’t rely on a supercomputer or even a remote server. Instead, it operates autonomously on a simple microcontroller unit (MCU), the kind you’d find in hobbyist electronics like Arduino boards. Being able to operate such a sophisticated AI on a cheap, off-the-shelf chip is one of the most impressive things about this project.
Avoiding Obstacles at Speed
To see if the drone could handle real-world chaos, the team tested it outdoors on a course with virtual obstacles. When a traditional wingless drone approached an obstacle, it struggled — either by veering off path or by losing altitude. The drone’s motors simply couldn’t deliver enough vertical force during sharp maneuvers.
But the flying squirrel drone, using both propeller thrust and wing-generated resistance, maintained its path and altitude across the complex obstacle course. It could climb and brake with minimal drift. In one test, it zipped through turns at 7.3 meters per second (roughly 26 km/h) and still improved its trajectory tracking by nearly a meter compared to the wingless version.
“The wing membranes’ impact in the real-world experiment appears to exceed what was observed in the simulation,” the team noted.
This extra lift and drag also alleviated battery strain. With less demand on the motors during sudden maneuvers, the drone avoided brownouts and instability that plagued the traditional quadcopter.
This isn’t the first attempt at adding passive surfaces to drones, but it may be the most complete integration of soft structures, physics-based learning, and real-time control. And it opens the door for new applications.
In disaster zones, drones like these could navigate tight debris fields. In forests, they might follow wildlife. And with better obstacle avoidance, they could even fly safely in urban canyons or dense warehouse aisles.
The authors are already looking ahead. Future versions could include even more advanced trajectory planning, allowing the drone to anticipate not just the next move — but the smartest one.