In this paper, we introduce an informative path planning (IPP) framework foractive classification using unmanned aerial vehicles (UAVs). Our algorithm usesa combination of global viewpoint selection and evolutionary optimization torefine the planned trajectory in continuous 3D space while satisfying dynamicconstraints. Our approach is evaluated on the application of weed detection forprecision agriculture. We model the presence of weeds on farmland using anoccupancy grid and generate adaptive plans according to information-theoreticobjectives, enabling the UAV to gather data efficiently. We validate ourapproach in simulation by comparing against existing methods, and study theeffects of different planning strategies. Our results show that the proposedalgorithm builds maps with over 50% lower entropy compared to traditional"lawnmower" coverage in the same amount of time. We demonstrate the planningscheme on a multirotor platform with different artificial farmland set-ups.
展开▼
