Flipping Awesome Drones Take Flight

Today’s drones may be pretty nimble in the sky, especially in the hands of an experienced pilot, but they still pale in comparison to the skills of birds. When flying autonomously, without the aid of a human pilot, the performance gap between drones and birds grows much larger. This is a big problem if drones are going to live up to our expectations in the future. They will not be making package deliveries, locating victims of natural disasters, or much else beyond aerial photography if we cannot improve their capabilities.

A big reason for this difference in abilities is birds’ instinctive understanding of the mechanics of flight. At times they do things that just seem wrong to us — they may momentarily enter unstable modes of flight, for instance, to swoop after prey or evade a predator. Flipping around the sky seemingly out of control is something we seek to avoid at all costs. But if you know what you are doing, it turns out it can actually be a good idea under the right circumstances.

Drone developers are awakening to this realization, which is resulting in the creation of some unconventional control mechanisms. Some of these mechanisms are hardware-based (like this flying squirrel drone we recently covered), while others rely on algorithmic advances. Researchers at Zhejiang University have just announced a new type of control system that falls into the latter category. Their control algorithm was developed specifically to carry out complex aerobatic maneuvers that might make us nervous, but that would make a falcon proud.

The algorithm is centered around a novel flight representation system. The team breaks down flight into a series of aerobatic intentions, which are discrete movements that reflect changes in both the drone’s position and orientation. By combining these intentions in various sequences, the system enables the drone to adapt to complex environments, such as obstacle-dense forests or urban spaces, by twisting, rolling, and diving their way to their goal.

To ensure these maneuvers are both safe and efficient, the researchers introduced a spatial-temporal joint optimization planner. This software helps the drone calculate the smoothest, most dynamic paths while avoiding collisions with fixed objects. A unique challenge in this area was yaw sensitivity — a rotational movement that can become unstable during fast turns. To address this, they incorporated a compensation mechanism to prevent disruptions during aggressive flight.

Using preloaded maps, onboard computing, and real-time visual data, the system continually adjusts the drone's path mid-flight. It even allows the craft to momentarily assume unstable flight attitudes — a trait shared with birds and bats when performing quick maneuvers.

The system has been tested extensively in both simulation and real-world settings. It performed well indoors and outdoors, successfully navigating obstacle courses and executing complex routines with minimal error. With further enhancements, this work may ultimately prove to be a significant step toward fully autonomous drones capable of handling missions once thought impossible without human control.