A Machine Learning Matheuristic for Agricultural Drone Routing

Advancing smart systems through Artificial Intelligence is crucial for developing more intelligent futures across various domains. In precision agriculture, autonomous systems such as drones are transforming operations like irrigation and fertilization, demanding sophisticated and efficient path planning strategies. Efficiently covering agricultural areas for targeted operations requires optimizing the path, a task complicated by geometric constraints, coverage requirements, and the combinatorial nature of visit sequences. This paper proposes a hybrid approach combining Machine Learning with mathematical optimization to generate optimal drone paths. An ML agent is employed to reduce the number of possible links between nodes, effectively handling the combinatorial complexity and potentially incorporating learned environmental patterns or human preferences. Subsequently, a mathematical model is formulated to optimize the precise trajectory. This model accurately captures geometric constraints, such as maintaining coverage within designated areas and potentially vehicle dynamics, finding the shortest and most feasible path for the provided sequence. This hybrid methodology explores the strengths of both approach: the first in handling complex sequential decisions, while the second focuses on precise geometric optimization. Results demonstrate that this integrated approach effectively generates efficient and robust drone paths for agricultural operations, making a significant advancement toward intelligent, autonomous smart-farming systems that enhance productivity and support human decision-making.