Tiny Robot Drones Learn To Navigate The World Like Honeybees

Tiny robot drones study to navigate nan world for illustration honeybees

Mapping their starting constituent for illustration bees do helps autonomous drones find their way

By Jacek Krywko edited by Sarah Lewin Frasier

A honeybee-based navigation strategy could thief miniature autonomous drones find their measurement home.

Jeff R Clow/Getty Images

Insect-size drones are excessively mini to lug astir analyzable navigation systems. To thief mini autonomous fliers find their measurement home, researchers are taking their cues from honeybees pinch nan Bee-Nav, described coming successful Nature.

A honeybee leaving nan hive first takes a short learning formation to memorize adjacent landmarks, explains nan study’s lead writer Guido de Croon, an artificial intelligence and robotics interrogator astatine nan Delft University of Technology successful nan Netherlands. As a bee flies away, “it keeps way of nan guidance and velocity of its movement,” de Croon says, successful a process called way integration. Because way integration is prone to accumulating mini measurement errors complete time, nan insect relies connected nan memorized landmarks to correct its people arsenic it gets backmost home. De Croon and his colleagues copied this workflow.

First, a drone performs a beelike learning formation astir its starting constituent utilizing a minuscule omnidirectional camera to seizure nan surrounding scenery. Midflight, it trains a mini onboard neural web to representation these images to location vectors, fundamentally invisible arrows pointing backmost to nan launchpad. This establishes a safe area called nan Learned Homing Area. Once trained, nan drone tin beryllium sent acold distant and travel backmost utilizing way integration first, backtracking based connected measured velocity and direction. If nan drone winds up anyplace wrong its starting safe zone, nan ocular neural web guides it nan remainder of nan measurement home.

On supporting subject journalism

If you're enjoying this article, see supporting our award-winning publicity by subscribing. By purchasing a subscription you are helping to guarantee nan early of impactful stories astir nan discoveries and ideas shaping our world today.

The Bee-Nav does this utilizing an off-the-shelf Raspberry Pi 4 machine nan size of a in installments paper that runs neural nets pinch betwixt 3.4 and 42.3 kilobytes of memory—thousands of times little than accepted mapping setups use. The team’s trial bots homed successful from a maximum of 600 meters (1,970 feet) distant outdoors contempt upwind gusts and camera-blinding sun glares.

“What I find particularly breathtaking is really small computation is needed,” says Sarah Bergbreiter, a mechanical technologist astatine Carnegie Mellon University, who was not progressive successful nan study. “For nan small-scale robots that my group and others activity on, this is nan benignant of attack that makes superior outdoor deployments plausible.”

De Croon’s squad is still moving retired a fewer challenges for nan platform, specified arsenic navigating betwixt aggregate memorized places and dealing pinch landmark-free starting points. “Platforms moving Bee-Nav will besides request section obstacle avoidance and readying capacity if nan situation is cluttered aliases dynamic,” says Sean Humbert, a mechanical technologist astatine nan University of Colorado Boulder, who was not progressive successful nan study.

But moreover now, de Croon says, nan Bee-Nav tin make autonomous, outdoor drones smaller and much power-efficient. “We could easy put it connected a 50-gram, moreover 30-gram drone,” de Croon claims. Scaling autonomous drones further down to nan size of existent bees, he notes, would require solving different basal problems for illustration miniaturizing batteries. “But we dream erstwhile these problems are solved successful nan agelong term, we will person nan intelligence fresh to lucifer that,” de Croon says.