While scientists rush to pin down the cause of colony collapse disorder and race to find a cure, engineers have wondered whether we might one day supplement real bees with mechanical ones. Called DelFly, a miniature robot has been built by an engineering team of the Micro Air Vehicles Laboratory at Delft University of Technology in the Netherlands. The engineers there have various practical applications in mind. For example, when perfected, the bots could flit around greenhouses spotting diseases with their cameras. The robots could also be fitted with apparatuses to perform an even more vital and insect-like task—pollinating crops.
What is this flitting about? It flutters over a green patch of grass, beating its transparent wings so fast it’s hard to tell how many it has exactly. It hovers over the grass, darts back and forth, and then shoots upwards making an impressive 360-degree flip, worthy of an aerobatics show. Zipping around, it acts eerily similar to a pesky housefly dodging a swatter.
While the flyer is insect-size, weighing less than an ounce, it is entirely man-made.
Called DelFly, the miniature robot was built by an engineering team of the Micro Air Vehicles Laboratory at Delft University of Technology in the Netherlands. The engineers there have various practical applications in mind. For example, when perfected, the bots could flit around greenhouses spotting diseases with their cameras.
The robots could also be fitted with apparatuses to perform an even more vital and insect-like task-pollinating crops. That sounds like something out of the Black Mirror television series, but the idea of building robotic pollinators predates that. A mysterious plague dubbed colony collapse disorder has decimated entire beehives worldwide since the mid-2000s. Scientists pondered many possible causes of bees’ plight, including pesticides, habitat loss, poor diet due to the limited variety of agricultural crops, parasites called Varroa destructor mites, and other pests-or a combination of these factors.
No matter the cause, science provided no clear remedy for the little creatures that helped our crops bear harvests for millennia. In America, bees pollinate about 80 percent of flowering plants and about 75 percent of the nuts, fruits, and vegetables humans eat, according to the U.S. Department of Agriculture. Other countries are similarly dependent on healthy bees to assist in food production.
While scientists rush to pin down the cause of colony collapse disorder and race to find a cure, engineers have wondered whether we might one day supplement real bees with mechanical ones.
That’s no small task. Bees are multifaceted creatures, artful flyers, and tireless pollen-gatherers known for their complex behaviors and hive hierarchy. Bee species went through millions of years of evolution to perfect the ability to fly through wind gusts, spot flowers, and gather pollen using the fine hair on their bodies and legs.
Should humans even attempt something as complex and beautiful as Mother Nature’s bee? Or is that a fundamentally flawed idea?
The answers to all these questions are as complex as the bees themselves. Robotics groups are tackling them from various directions.
No Small Ambition
When Guido C.H.E. de Croon at Delft University of Technology and his colleagues began working on what would become DelFly, he wasn’t thinking about building a mechanical pollinator. Instead, the team was interested in creating a small flapping wing drone capable of carrying a camera.
The first DelFly, built in 2005 by 11 students as a capstone project, was insect-like, but on a scale of the giant dragonflies of the Carbonaceous era. Made from ultra-lightweight materials—thin sheets of Mylar stretched over a carbon fiber frame—it weighed 21 grams (slightly less than an AA battery) and had a wingspan of 50 cm. Over time, as the Delft team learned more, the drones miniaturized. DelFly II had about half the wingspan and could hover and fly forward and backward, staying airborne for 15 minutes. DelFly Micro shrunk to 3 grams and a 10 cm wingspan. The Guinness Book of Records named it the “smallest camera-equipped aircraft in the world.”
That’s still enormous as bees go. A worker honeybee weighs approximately 130 milligrams and has a wingspan of less than 1 cm.
As the drones lost mass, they added capability. The 2013 DelFly Explorer gained autonomous navigation, thanks to a stereovision camera and hardware for processing obstacle detection—no mean feat for a system that could only weigh a few grams. Self-driving cars make complete 3D maps of their environment to feed into their obstacle detection algorithms, but that requires sensors, processors, and memory, even the lightest of which would be too heavy for a gossamer robot to lift.
So the team built very different vision algorithms, inspired by hunting dragonflies. “If you ask a biologist how a dragonfly catches flies, the current hypothesis is that it keeps them in the same position [in the field of vision] while making them bigger and bigger” as they approach, de Croon said. “So you could make it very complicated and calculate the fly’s velocity and direction—or you can use the same kind of a simple intelligence they use.
“That’s what we did—we either programmed it into the robot, or tried to make the robot learn it by itself.”
The drones learned to avoid collisions, but they weren’t very stable in flight. Even when they flew inside a room, an air-conditioner’s draft could toss them around. They needed more agility to recover from such gusts. The team realized they had to abandon some of the traditional concepts of flying machines and once again look at how insects work.
The result was the DelFly Nimble. Unlike its predecessors, whose flight is controlled similar to conventional airplanes via deflections of specific surfaces on the tail or behind the wings, DelFly Nimble has neither. Instead, it is controlled by insect-inspired motion adjustments of its two pairs of flapping wings. The absence of the tail makes DelFly Nimble very agile and less vulnerable to wind. That enables it to dart as nimbly as a fly escaping a swatter—and then straighten itself out and return to its original heading.
“If you try to swat a fly, it will make a super sharp turn like a fighter jet, and then it will quickly turn its body into the flight direction again,” de Croon explained. What’s really interesting, he added, is that the fly doesn’t do it consciously. It does so “passively,” essentially relying on the aerodynamic forces specific to its flapping wings.
Even more surprisingly, so does the robot, even though it hasn’t been programmed to do it.
“The robot also corrects for the flight direction error, but we know for sure that we didn’t put this into the control code,” de Croon said. “It’s a new effect and we think it applies to all flapping wings.”
Bringing DelFy to greenhouses will take more time and work. But also, commercial pollinating will require many such bots operating in an organized fashion. You’d need a beehive. And that beehive would have to learn some swarming behavior.
Building a robotic hive was just what a team of roboticists at Harvard’s School of Engineering and Applied Sciences have been doing for the past decade. A brainchild of Robert Wood, Gu-Yeon Wei, and Radhika Nagpal, the Harvard hive of miniature robots was at least a partial response to colony collapse disorder.
Wood had built a small fly-like machine and then Wei, who had read about pollinators’ deaths, suggested in 2009 that they could build bees. The trio envisioned a box of 1,000 RoboBees, cumulatively weighing under a kilogram, which would be set free to find flowering plants to pollinate—or, with different programming, search for survivors at the scene of a disaster or the location of a gas leak along a pipeline.
The team, which drew on researchers from not only the engineering department, but also biologists, computer scientists, and materials experts, spent several years developing their idea with funding from the National Science Foundation. The biggest challenge was in forging the bees’ brain. The aim was to build the brain in layers—sensors to interpret the surroundings, an electronic nervous system for basic control tasks, and a programmable cortex for decision-making.
The team used a deep neural network (DNN) chip, which mimics the real brain’s architecture and neuronal connection, enabling the robot to do many things concurrently. DNNs can work with speech recognition, so in theory, RoboBee hives could be voice-controlled.
That ambition ran into the limits of electronic hardware. Technologically, it’s not yet possible to equip a 100-milligram flying robot with enough control and battery power to make complex computations, fly autonomously, and interact with other bots. Consequently, prototype RoboBees remained tethered—attached to a wire that supplied power and control, which in real crop-pollinating settings would likely amount to a lot of tangled wires.
So researchers built a slightly bigger bot inspired by a tobacco hawkmoth, which weighed 3 grams but had full autonomy.
“It took lots of principle from RoboBees but we put it on a slightly larger platform in order to put power and control onboard,” explained Michelle Rosen at the Harvard Microrobotics Lab who worked with this larger platform. The moth bot was completely autonomous, but the hive is still a work in progress.
Flying is not enough to make a bee. To fill the ecological niche, a robotic bee would need to gather pollen and transfer it from one flower to another. Biological bees hold pollen with fine hairs on their bodies and legs. Would robotic bees need hairs too? Eijiro Miyako, an engineer at the National Institute of Advanced Industrial Science and Technology in Tsukuba, Japan, wanted to see just how simple a pollinator could be. He affixed horsehair onto a small quadcopter drone, then covered the bristles with a specially designed ionic gel, sticky enough to transfer pollen between flowers. When the furry drones flew onto the pink Japanese lilies, they indeed gathered pollen, which they could then transfer to other flowers.
Unlike the real bees that approach florets gently, Miyako’s do more of a crash-landing—but they successfully collected pollen in 53 percent of attempts and pollinated flowers in 37 percent. And while the success rates could be improved, the drones proved they can do it.
“Such materially engineered artificial plant pollinators should lead to the development of high-performance robotics that can help counter the decline in honeybee populations,” the team wrote in the study published in the journal Chem.
Miyako believes that with further work, such drones can learn to navigate from flower to flower. “We believe that robotic pollinators could be trained to learn pollination paths using global positioning systems and artificial intelligence,” he said.
Even so, not everyone salutes the artificial bee’s flight. A 2018 article in the journal Science of the Total Environment presented multiple problems associated with robotic pollinators’ use. Building and deploying enough of them to be commercially significant would result in high energy, carbon, and material footprints. And old and broken ones littering the countryside could become a source of environmental pollution.
In addition to the engineering problems, replacing honeybees with bot bees would have ecological and sociological ramifications. There are about 350,000 species of flowering plants on the planet, many of which use specialized pollinators, so if man-made bees focus only on pollinating commercial crops, other plants will decline, leading to severe biodiversity loss across the ecosystem. And because manmade bees will not be cheap, the world’s two billion low income farmers will not be able to afford them, which could radically upend rural economies.
Even the engineers working on robotic bee projects agree that replacing living bees with mechanical ones isn’t the right way to solve the CCD problem.
“Humanity should try to preserve the natural pollinators,” de Croon said, but added that robotic bees still can be helpful in greenhouses or indoor hydroponic farms. “We’ve been working on greenhouse applications.”
More interestingly, man-made bees may prove their worth in other ways. For instance, some of the flapping-wing drones have taught their creators about the intricate behavior of real bees—as DelFly did with its passive aerodynamic wing-flapping effect.
Now de Croon is pondering another question. He noticed that as his bee drones land on a surface, they slightly oscillate up and down. When he looked closely, he noticed that the living bees do the same.
“I saw that the little bumps in their movements were similar to what my drones were doing,” he said. Remarkably, biologists don’t yet have an explanation for that behavior. So man-made bees can serve as a robotic proxy for learning more about their wild counterparts.
“You can use robots to find out things about biology,” de Croon said. And knowing more about bees’ inner workings may also help us protect them better.