In the Blink of a Fly: Flies’ Upside-Down Landings Inspire Robot Designs
By Asher Jones
Flies are some of nature’s best aviators. It may come as no surprise then, that the aptly named insects have inspired the design of past robotic fliers. Now, researchers have described how flies land upside down on ceilings—with the goal of teaching drones to perform similar inverted landings.
Drones have countless uses for surveillance, search and rescue, agriculture, and conservation. The problem is that current commercially available models only have enough battery life to power 30 minutes or less of airtime. Robots with perching abilities could rest on a surface to conserve power, recharge, or do their work.
“They [drones] have both civil and military applications,” said Bo Cheng, Ph.D., assistant professor of mechanical engineering at Penn State. “To have these robots land upside down on the ceiling or on a vertical surface will greatly expand their functionality because they don’t have to fly all the time. That will make the operation time much longer.”
The team used high-speed video to capture the way Calliphora vomitoria flies (sometimes known as blue bottle flies) land on a ceiling. After putting the insects into a flight chamber box, the researchers shook the box to startle the flies and recorded the landings.
The flies that successfully accomplished inverted touchdowns—or touch-ups, perhaps?—displayed a coordinated sequence of behaviors, the researchers report in their article published in October 2019 in Science Advances. The insects accelerated, rotated their bodies upside down, and extended their front legs to grip the ceiling. Finally, they used this grasp to cartwheel their remaining legs onto the surface, completing the landing. Flies that botched any part of this sequence—by not rotating their body enough, poorly timing their maneuvers, or failing to extend their legs—headbutted the ceiling and failed to land.
To invert their bodies for a successful landing, the study shows flies must rotate at the exact right moment. The team found that C. vomitoria use visual information to decide when to start these maneuvers. The researchers inferred the fly’s-eye view by measuring the each fly’s body orientation in relation to a grid on the ceiling. As flies moved closer to the ceiling, the image of the grid increased in their eye like a tossed ball that appears to increase in size as it approaches you. Just as a ball player uses the rate of size expansion of the ball to judge when to extend a hand to make the catch, flies appear to use the same visual cue to trigger their rotations.
According to Cheng, flies are expert aviators because they flap their wings constantly during flight. By changing the angle of their wing strokes, the insects can rotate themselves in any of three possible directions. To think about this, try imagining a canoe. It can capsize by flipping end-over-end (pitch) or by rolling sideways (roll). It can also turn left and right (yaw), although this won’t lead to an inversion. Similarly, the team found that flies used either pitch, roll, or a combination of the two to capsize their bodies as they soared toward the ceiling.
The degree of each fly’s pitch and roll was associated with visual cues and flight speed, suggesting that the insects measure speed using their antennae or wind-sensitive hairs. By perceiving this sensory information and adjusting their rotation strategy “on the fly,” C. vomitoria flies achieve the most efficient touchdown.
“We didn’t expect that flies could use such a rich repertoire of behaviors to land,” says Cheng. “Previously people thought that landing upside down is much simpler than what we have revealed.”
To build flying robots that can land upside down, the team will use a machine-learning approach to teach the flies’ movements to drones.
“You can make the analogy of the fly being the parent and the robot being the kid,” says Cheng. “The kid has seen what the parent has done—say, to pick up a cup or some complicated motor task. They will observe that [behavior] and start to mimic it. The first step is trying to imitate the parent. But they need to practice afterwards to get better and better. That is the reinforcement learning part. So, we are doing very similar things for these robotic fliers.”
Although some drones have completed inverted landings previously, these models used cameras linked to external computers to achieve the stunt. According to Cheng, this approach is computationally demanding. The team hopes to design drones that can land upside down using onboard sensors alone. These fly-inspired robots will use simple visual information to coordinate a successful landing.
“There is a lot of work going on in the realm of unmanned aerial vehicles that explore the behaviour and dynamics of insects as a model for drone flight,” said Pauline Pounds, associate professor in mechatronics in the School of Information Technology and Electrical Engineering at the University of Queensland and who was not involved with this study. “This work on visual control of flies perching on surfaces is well-situated within the field [of aerial robots] and has clear value in the context of existing work. It is also potentially very useful as the utility of small unmanned hovering robots as observation or inspection platforms is increasingly recognised.”