What we do

Robotics is expected to play a major role in the agricultural/farming domain. Swarm robotics, in particular, is considered extremely relevant for precision farming and large-scale agricultural applications. However, swarm robotics research is still confined into the lab, and no application in the field is currently available. SAGA will demonstrate for the first time the application of swarm robotics principles to the agricultural domain. Specifically, we target a decentralised monitoring/mapping scenario, and implement a use case for the detection and mapping of weeds in a field by a group of small unmanned aerial vehicles (UAVs).

Who we are

The SAGA experiment is founded by the ECHORD++ project. SAGA is a collaborative research project that involves: the Institute of Cognitive Sciences and Technologies (ISTC) of the Italian National Research Council (CNR), which provides expertise in swarm robotics applications and acts as the coordinator for SAGA's activities; the Wageningen University and Research Centre (WUR), which provides expertise in the agricultural robotics and precision farming domains; and Avular B.V., a company specialised in UAV solutions for industrial and agricultural applications


Advanced UAV for Swarm Operations

We start form the PrecisionScout UAV developed by Avular B.V., an industrial-grade quadcopter with four motors, triple redundant autopilot, five inertial measurement units (IMUs) and RTK-GPS. The standard PrecisionScout will be equipped with self-localisation and communication devices based on UltraWideBand (UWB) technology, and on a enhanced onboard processing based on Nvidia Jetson.

Onboard Weed Recognition

The on-board vision system will perform object detection or semantic segmentation onboard and online, to count the number of weeds above a specific size or otherwise measure their development. Several machine learning methods will be tested for this purpose (e.g, SURF features, bag-of-visual-words clustering, and SVMs), and the results will be exploited for mapping and for motion control.

Collective Field Mapping

We seek collective strategies with an optimal trade-off between exploration and timely weed recognition: UAVs will be recruited to monitor areas in the field that have been identified as potentially containing weeds, while weedless areas are quickly abandoned. In this way, resource allocation is adapted to the field heterogeneities, and error-prone individual inspection will be compensated for through collaborative re-sampling.

more info






End Users and Partners