Particle.news
Download on the App Store

Bee-Inspired 'Bee-Nav' Guides Tiny Drones Home With Kilobyte-Scale AI

The Nature study shows tiny neural nets can guide small drones home using bee-like learning flights.

Overview

  • Researchers report in Nature that Bee-Nav lets a drone map its home area during a short learning flight that captures surrounding landmarks.
  • Once deployed, the drone first relies on path integration to estimate its route back and then switches to the learned visual cues near home.
  • The system ran on an off-the-shelf Raspberry Pi 4 using neural networks sized between about 3.4 and 42.3 kilobytes.
  • Trials returned drones from up to 600 meters with near-perfect indoor results but about 70% success outdoors in wind because tilt warped the panoramic images.
  • The team sees use on 30–50 gram craft for greenhouse monitoring and inspections, though multi-place memory and local obstacle avoidance still need to be added.