Toward Aerial and Ground Robots Collaboration for 3D Map Building in Precision Agriculture

Farming robots offer a great potential for minimizing the amount of chemical inputs released in the fields through targeted interventions. In multi-robots applications, where both Unmanned Aerial Vehicles (UAVs) and Unmanned Ground Vehicles (UGVs) operate in a coordinate way, the ability to obtain a unified environment representation of the target field is a must-have capability. However, a global registration of heterogeneous ground and aerial maps is a challenging task. This turns out to be especially difficult in agricultural fields, where the visual appearance is rather homogeneous and no meaningful 3D geometrical structures can be exploited. In this paper, we tackle the problem of fuse data from heterogeneous robots by proposing a novel registration pipeline that leverages semantic information extracted from image-based 3D reconstructions of the target environment. The proposed approach performs the alignment exploiting only meaningful parts of the recorded data, by clustering 3D points that belong to vegetation and filtering out the less significative and redundant points belonging to the soil. We evaluated our system on data acquired on a real field in Eschikon, Switzerland, showing good alignment properties in two challenging scenarios.