The term for unmanned aerial drones in English originally means a male bee. It comes from the fact that the sound of propellers resembles the buzzing of bees in flight. Now drones are becoming more like bees not only in name but also in flight principles. A research team led by Guido de Croon, a professor in the Faculty of Aerospace Engineering at Delft University of Technology in the Netherlands, said on the 14th in the international journal Nature that it "developed an efficient drone navigation system using the principle by which honeybees find their way home."
Drones in flight have become an everyday sight. They deliver parcels, film broadcast footage, and check the growth status of crops. In the military, they have become essential equipment for detecting and attacking enemy forces. As drones are used for more purposes, the flight range has also increased. But mounting a high-performance Global Positioning System (GPS) makes drones heavier and increases energy consumption.
The Delft team turned to nature for a solution. They applied to drones the secret that lets honeybees venture far in search of nectar and pollen yet unfailingly find their way home. The bee-inspired drone navigation system, Bee-Nav, succeeded in return flights using just 42 kilobytes (KB) of memory. That capacity can hardly store even a single low-resolution thumbnail image.
◇Thanks to prelearned flights, return path is faster
Honeybees are an excellent model for an efficient navigation system. A bee's brain, about the size of a sesame seed, has only one one-hundred-thousandth as many neurons as a human brain, yet it performs higher cognitive functions such as remembering its location and measuring distance. The researchers began by analyzing the principles of bee navigation.
Bees first estimate how far and in what direction they have moved using visual cues. This is odometry—measuring travel distance. It is akin to a person counting steps. While flying, bees calculate distance by measuring the speed at which the surrounding scenery sweeps past their eyes. The faster nearby objects move toward the rear of their field of view, the more quickly they perceive themselves to be moving.
This alone has limits. It may work at close range, but over long distances it is hard to count steps accurately. Even small variations in stride add up to large errors. Bees compensate for odometry's limits with visual memory. They remember the scenery around important places such as home to check whether they are on the right path.
De Croon said he "was fascinated that bees can fly far along a winding path yet return almost in a straight line." That suggests they can immediately recognize the area around home even from afar. The team explained that the secret is a brief learning flight circling the surroundings when leaving home. It is like stepping outside and walking around the neighborhood. That way, on the way back, they can recognize the neighborhood with the home no matter which route they approach from.
A drone equipped with bee navigation also performs a brief learning flight near the point of departure. Through this, the drone collects panoramic images of the environment. It then uses a small neural network to process these images and learn to estimate the direction and distance back home.
Bee-Nav integrates odometry and visual memory to find the way. When a drone flies far, home may be hard to see or hidden by obstacles. It first uses odometry to head home even if accuracy is low. As it gets closer, it uses visual memory to refine the path more accurately.
◇Outdoor flights must also handle wind and lighting changes
The experimental results were successful. The drone first flew a learning area measuring 5 meters by 5 meters at an indoor test site. As a result, it achieved a 100% return rate in four test flights conducted in a test field measuring 50 meters by 30 meters. Using a small neural network of 3.4 KB, the drone interpreted the surrounding panoramic images and estimated its heading and the remaining distance to home.
The same method succeeded in larger-scale tests. Using a 42 KB neural network, the drone returned successfully even after flying more than 600 meters. However, while it succeeded in every test in large indoor spaces such as hangars, its success rate dropped to 80% in windy outdoor conditions. The researchers said, "Because wind tilted the drone, it became difficult to use images for navigation," adding, "Changes in sunlight or grasslands without distinctive landmarks can also be causes of failure."
The team said that even at the current level, Bee-Nav is sufficient for indoor conditions such as greenhouses. Drones are used in greenhouses to check crop growth and to detect disease or pests early. The researchers noted, "In greenhouses, drones need to be lightweight to keep coworkers safe, and Bee-Nav is suitable for this."
Barbara Webb, a professor in the School of Informatics at the University of Edinburgh in the United Kingdom, wrote in a commentary published in Nature the same day that "this study is meaningful in that it both broadens scientific understanding of insect navigation systems and presents efficient robotics technology."
Drones that cannot use GPS need complex 3D maps to fly. But on farms it is hard to use high-performance drones that require massive computation and memory, and at disaster sites, even if such drones are available, prior information is insufficient. Webb said, "Bee-Nav does not require large compute capacity, so low-cost drones can work sufficiently in such places," adding, "as seen in outdoor tests, technical improvements are needed to handle wind and lighting changes."
References
Nature (2026), DOI: https://doi.org/10.1038/s41586-026-10461-3
Nature (2026), DOI: https://doi.org/10.1038/d41586-026-01321-1