Autonomous systems for vineyard and orchard tasks: applications in navigation, harvesting, and pruning

Precision agriculture in vineyards and orchards faces increasing challenges due to labor shortages, high costs, and the need for sustainable practices. This thesis addresses these issues by developing an autonomous, electric robotic platform designed for navigation, harvesting, and pruning in agricultural environments. The platform is a tracked electric rover with a payload capacity of up to one ton, featuring a modular power system and advanced perception and control modules. Robust navigation in unstructured fields is achieved through sensor fusion of LiDAR, RGB-D cameras, RTK-GNSS, IMU, and wheel odometry, ensuring accurate localization even under canopy occlusions. Two main applications are presented. The first is a kiwi harvesting robot for T-bar orchard systems, integrating a custom robotic arm and end-effector designed through workspace analysis, motor sizing, and kinematic optimization. A vision-based detection pipeline combined with hybrid motion planning allows the system to reach and detach fruit while avoiding obstacles. The second application is an autonomous tree inspection framework for pruning. It reconstructs tree structures from LiDAR point clouds, performs skeletonization and branch classification, and integrates RGB-D bud detection using YOLOv11 with slicingaided hyper inference. An active vision strategy selects new viewpoints to enhance coverage of occluded regions and enable accurate bud-to-branch association. Experimental validation in laboratory and real orchard conditions demonstrated that the rover achieves stable navigation, the harvesting system effectively detaches fruit in T-bar kiwi orchards, and the inspection framework improves bud association accuracy, particularly with active vision. Remaining limitations concern perception robustness under diverse canopy geometries and model generalization. Overall, this work delivers an integrated framework combining perception, planning, and manipulation for orchard and vineyard automation, advancing sustainable agricultural robotics and outlining future directions for improved generalization and adaptability across different crops.