An Adaptive Row Crops Path Generator with Deep Learning Synergy

The autonomous navigation of agricultural field machines strongly depends on the global path generation system. Indeed, a correct and effective path construction heavily influences the overall navigation stack compromising the successfulness of the robot mission. However, the most commonly used search algorithms struggle to adapt to environments where a significant prior knowledge of the domain is not negligible. Despite this crucial factor, path generation for row-based crops has received little attention from the research community so far. The proposed research introduces a novel global path planning system that works in synergy with a deep learning model to provide an accurate and centered path with respect to the rows of the analyzed crop. It guarantees the full coverage of the given occupancy grid with less processing time compared to other available literature solutions. Moreover, the presented methodology can detect an anomaly in the path generation and provide the hypothetical user feedback of the missing full coverage of the given crop. Indeed, especially in a practical application, the correct coverage and centrality of the path are essential for effective autonomous navigation. Experimentation with synthetic and real-world satellite occupancy grid maps clearly show the advantages of the proposed methodology and its intrinsic robustness.