Robots Flexible Drone Can Withstand Collisions

ETH researchers have developed FlexiQuad, a soft drone whose frame deforms in a controlled manner upon collision. They have filed a patent for the design. Despite its soft frame, the drone can fly quickly and nimbly.

Quadcopters have been built according to the same basic principle for more than a hundred years: Four arms, each with a propeller, extend from a rigid central body. This design is ideal for precise and agile flight, but it has one weakness: the drones are very sensitive to unexpected contact with their surroundings. If they collide with a wall or another obstacle, they can lose stability or even break apart.Researchers at ETH Zurich have now developed a new design. Their FlexiQuad drone consists of a soft, ring-shaped frame without a central body. The rotors, batteries, and control module are evenly distributed along the frame. Thanks to this novel architecture, the flying robot can deform upon collision, thereby absorbing some of the collision energy. FlexiQuad was developed by Luca Girardi and then-ETH professor Stefano Mintchev at the Environmental Robotics Lab. Girardi is currently conducting research at the Mobile Robotics Lab at ETH Zurich, while Mintchev has since taken up a professorship in Agricultural Robotics at the University of Bonn. The two have filed a patent for FlexiQuad. The invention has been nominated for this year’s Spark Award.

Our drone is very agile and can fly at speeds of up to 80 km/h. It achieves linear and angular accelerations comparable to those of rigid drones. Flexible doesn't automatically mean slow.

Luca Girardi

Soft Upon Impact, Stiff in Flight

The frame is not equally flexible in all directions. Girardi explains: “We drew inspiration from birds and insects. We were interested in the stiffness of their wings and how they bend them—or don’t bend them—in different directions.” Based on this principle, the researchers designed a frame that yields where deformation is desired: laterally. In the vertical direction, however, it is stiffer so that the rotors can transmit the forces necessary for flight. Despite its flexible frame, FlexiQuad is fast and agile. “Our drone is very agile and can fly at speeds of up to 80 km/h. It achieves linear and angular accelerations comparable to those of rigid drones. Soft does not automatically mean slow,” says Girardi. 
In collision tests, FlexiQuad struck obstacles at speeds of up to about 25 kilometers per hour. The flexible frame temporarily deforms during the impact, absorbing part of the force. This allows the drone to continue flying after the collision. According to Girardi, the main reason the researchers did not test higher collision speeds was a lack of space for a longer deceleration distance. The flexibility can also be used to grasp and transport objects.

FlexiQuad also has a modular design and is relatively inexpensive to manufacture. The materials for the frame cost less than 15 francs. According to the researchers, costs could be reduced even further in the future. When the ring is opened, the flat frame elements can be stowed away compactly. Girardi has already transported a prototype this way in a medium-sized suitcase. With quick-release fasteners and slide-in batteries, the drone could be assembled on site in a short amount of time in the future, the researcher explains.

Use in Environments with Limited Visibility and Difficult Access

This flexibility could be particularly advantageous during inspections: A larger drone could carry more sensors and stay airborne longer, yet still need to fit through narrow openings. The concept could also be useful for transporting small objects. In tests, FlexiQuad has already picked up, transported, and set down a soft first-aid kit. The drone was also able to carry a load heavier than itself. “If a drone needs to cover long distances quickly during a mission, its sensors might detect an obstacle—such as a tree—too late to avoid it,” explains Girardi. For a rigid drone, such a collision could mean the end of the mission. Because FlexiQuad can better withstand collisions, the drone could in the future fly faster even in areas where it would be too risky to rely solely on detecting and flying around obstacles. 
The patent envisions applications in complex and hard-to-reach environments, such as inspecting ballast tanks on ships, conducting search-and-rescue operations in collapsed buildings, or taking environmental measurements in dense forests. Girardi also sees other potential applications: For example, a net could be attached to the open center of the FlexiQuad to capture unauthorized or unwanted drones.

At first, I was skeptical about whether the drone would even fly without a central amplifier. But I really liked the approach. Then we gave it a try—and it worked.

Luca Girardi

What Flexible Drones Still Need to Learn

Before FlexiQuad can be used outside the lab, the drone must be able to reliably map its surroundings and navigate autonomously. “We continue to challenge the centuries-old tradition of designing drones with rigid structures—a tradition that stems from the fact that a rigid structure facilitates precise control,” says Girardi. 
After a collision, vibrations in the flexible frame can blur the camera images, making navigation more difficult. In a recent study, Girardi and his colleagues therefore drew inspiration from natural reflexes: Using reinforcement learning, they taught FlexiQuad to first stabilize itself using its motion sensors after a collision and then resume autonomous flight. At the same time, the team is investigating whether the flexibility itself can be used as a source of information. Inspired by how insects perceive deformations in their wings, the scientists are investigating whether bends and vibrations in the frame could provide additional information for flight control in the future. 
Girardi has been refining the drone for about five years. He particularly remembers the moment when FlexiQuad took off for the first time without central reinforcement: “At first, I was skeptical about whether the drone would even fly without central reinforcement. But I really liked the approach. Then we tried it—and it worked.”