Real-time Driving Assistance System for Autonomous Robots in Orchard Environments

In recent years, autonomous mobile robots have been applied to various industries, and the development of agricultural automation technology utilizing autonomous robots has been actively researched in the agricultural sector. This study proposes a depth camera-based driving assistance system for robot autonomous navigation in a rain-shielded citrus orchard environment. The proposed system consists of a segmentation-based obstacle filtering method and a tree trunk-based way points generation method for generating driving paths. A trained YOLOv8-Seg model is used to determine whether an obstacle needs to be avoided, and the center points of the tree trunks on both sides of the orchard are detected to generate way points for robot navigation. The accuracy of the YOLOv8-Seg model was measured through comparative experiments using a dataset acquired directly from a citrus orchard, and real-time operability was verified on the Jetson AGX Orin board. Furthermore, the efficiency of the proposed driving assistance system was demonstrated by measuring the way points generation error for autonomous robot navigation in a model orchard.