Multi-UAV active sensing for precision agriculture via Bayesian fusion

Unmanned aerial vehicles (UAVs) represent a well established technology for the monitoring and mapping of crop fields in precision agriculture applications. In order to fully exploit all the possibilities they offer, a key challenge is devising optimal flight plans that maximize both data accuracy and field coverage. Active sensing techniques have shown promising results in this direction, as they permit online adaptation of plans based on the monitoring objective at hand. Here, we propose an active sensing strategy for multiple autonomous UAVs based on the concept of information gain, and a factor graph field representation for mapping in a Bayesian framework by taking into account spatial correlation effects. We validate our approach on a simulated feature detection scenario, and study the effects of different feature distributions and varying levels of information sharing. Simulation results show large performance improvements over common non adaptive methods based on uniform or random-walk monitoring, highlighting the advantages of active monitoring in reducing uncertainty and maximizing coverage.