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.