A drone that lands on its own: SmartAgriHubs

SmartAgriHubs demo, 2021. Credit: Exobotic Technologies.

Context

Under the Horizon 2020 SmartAgriHubs programme, Exobotic, Robovision and Proximus joined ILVO and UGent around a single question: can a drone inspect a maize or potato field autonomously, week after week, and feed an AI-driven variable-rate sprayer? In June 2021 the consortium publicly demonstrated what 5G, AI-based weed detection and an autonomous drone can do together to produce real-time task maps for site-specific spraying. I was technical lead on Exobotic’s side of the demo.

Technical aspects

Three building blocks had to fit together:

Autonomous charging ground station. A compact outdoor enclosure that opens on request, guides the drone to a centred landing via a physical centering mechanism and a visual fiducial, and tops up the battery wirelessly between flights, so nobody has to be on the field to flip switches or swap batteries.
Closed-loop autonomous flight. A PX4 autopilot driven by a MAVLink mission-manager on the companion computer. Takeoff and landing are synchronised with the station door; waypoints follow terrain at fixed AGL; the hyperspectral camera shutter is triggered in the loop, and each frame is logged with its GPS fix, IMU state and image id for downstream stitching.
5G imagery pipeline. Hyperspectral frames are heavy; streaming them to the cloud as they are captured, rather than dumping to an SD card after the flight, collapses the inspect/decide turnaround to flight-time plus seconds. Robovision’s vision pipeline turns them into segmented crop tiles and an actionable spray map.

Results

The consortium flew a full closed loop (drone takeoff, hyperspectral capture, cloud inference, spray map) on real maize and potato plots. As a SmartAgriHubs representative put it: “the autonomous drone takeoff and landing solution of Exobotic is an important step towards truly autonomous drone flights, overcoming current limitations imposed by the balance between weight and battery size.”