Honeybees as a model for efficient indoor drone navigation

Scientists at Delft University of Technology have developed Bee-Nav, an indoor navigation system for drones inspired by the navigation techniques of honeybees. Bee-Nav requires only a few technical resources for drone swarms to navigate independently indoors, such as in greenhouses or industrial facilities.

In indoor environments where GPS is unavailable, drones and robots often rely on detailed maps of their surroundings for navigation. However, such map-based systems require high processing power and significant storage space on the drones and robots. This makes them unsuitable for micro-drones, as they become expensive and energy-intensive to operate.

Therefore, the researchers at Delft University of Technology have devised a system that requires less processing power and minimal storage, drawing inspiration from the navigation of honeybees. With their tiny brains, bees have only a limited memory capacity. The scientists have summarized their research findings on Bee-Nav in the study “Efficient robot navigation inspired by honeybee learning flights“, published in the journal Nature.

Navigation with Odometry and Visual Memory

Honeybees navigate using methods such as odometry. Based on visual motion information, they estimate how far and in which direction they have moved. The principle is similar to a combination of a pedometer and a compass. This allows bees to find their way back to their hive. The problem is that odometry is subject to drift, meaning the recorded distance and direction become less accurate as the distance increases. Therefore, bees also use their visual memory by memorizing important locations, such as the area around their hive, in advance.

This typically happens through short learning flights in the immediate vicinity of the hive. The more familiar bees are with their surroundings, the further they can venture from the hive and still find their way back.

“We were fascinated by the fact that honeybees can fly far away from home along winding paths, yet return almost straight back,” says Guido de Croon, Professor of Bio-inspired AI for Drones at Delft University.

The researchers discovered that as bees get closer to their hive, they rely more heavily on their visual memory, while largely depending on odometry otherwise.

Bee-Nav System

The scientists have artificially replicated this navigation system for drones. The Bee-Nav drone undertakes a short learning flight around its “home.” During this flight, it collects panoramic images of the surroundings. A neural network learns to process these images to calculate the direction and distance to the base.

Similar to a bee, the drone does not always know where its base is located. For instance, it might be perceived as too small from a distance or be obscured by vegetation. The researchers compensated for this by training the neural network with odometry estimates of the distance and direction to the base. In combination with visual navigation, this was sufficient to allow the drones to return to their base despite the drift as the distance increased. In four flights starting from different points within the known area, the drone consistently returned successfully to its starting point. The neural network used for evaluating the panoramic images comprised only 3.4 KB. The drone flew faster at greater distances and slowed down as it approached the starting point.

The scientists tested the drones in further unknown indoor and outdoor environments. The drone completed the indoor tests without any problems. In an outdoor test at the Dutch drone field laboratory “Unmanned Valley” in Valkenburg, the drone covered a distance of more than 600 m and still returned successfully to its base. However, the neural network used in this case was slightly larger, at 42 KB.

In windy environments, however, Bee-Nav reaches its limits. The wind forces the drone into a tilted position, limiting the usefulness of the drone's images for navigation. Consequently, the success rate for flights in windy conditions dropped to 70 percent.

The researchers believe that the technology could be used, for example, in greenhouses. The small drones could inspect plants there and detect diseases and pest infestations early on. This could help reduce crop spoilage.

(olb)