Drones are everywhere: inspecting crops, delivering packages, and exploring industrial sites. Yet, despite rapid development, most drones still rely on heavy, power-hungry computers for navigation. Now, scientists led by Delft University of Technology have found inspiration in honeybees to solve this problem.
The “Bee-Nav” navigation strategy, inspired by honeybees, enables tiny robots to travel long distances and return home with only 42 kilobytes of neural memory. Like bees, these robots first perform a short learning flight to gather visual memories of their surroundings. The team was inspired by the way honeybees make short learning flights near their hive before exploring further, allowing them to recognize their environment and reliably return home. After that, they can venture far and still find their way back. This innovation opens the door to lightweight, safe, and energy-efficient drones, especially for tasks like greenhouse monitoring. The research also sheds light on how insects navigate in nature.
Professor Guido de Croon explains that bees use odometry for long-range navigation and visual memory when close to home. Using a neural network, the team mimicked this by having robots combine these strategies to process panoramic images and estimate direction and distance home. The robots first collect panoramic images, which a small neural network processes to estimate the way home, even if the destination is hidden or distant. Even when the robot can’t see home directly, odometry and visual cues are enough to navigate back with high accuracy, as shown in experiments by the first author, Dequan Ou.
Tests in both indoor and outdoor settings proved successful, with drones flying over 600 meters and reliably returning using Bee-Nav’s compact neural network. Success rates remained high indoors, while wind sometimes reduced outdoor reliability—a challenge the team aims to address. Despite minor errors, the robots successfully returned home from various starting points, adjusting their speed as they neared their destination. The neural network required minimal memory for these tasks.
“The experiments are very encouraging,” said Ou. “But they also show that our current system needs to become more robust in real-world conditions.” Greenhouse monitoring is a key application. Lightweight, safe drones could help growers detect crop issues early while being easy to use near people. Outdoor trajectory is the goal for the full Bee-Nav strategy, with the learning flight, outbound and inbound flight based on odometry, and then visual homing in the learned area. The research also deepens our understanding of how bees return to their hives using visual learning.
