NIFA: Collaborative Research: ARMOR-Harvest: Aerial Robotic Morphing for Occlusion Resilient Apple Harvesting

Agricultural harvesting is one of the most labor-intensive operations in food production, and labor shortages are creating significant challenges for growers. Although robotic harvesting technologies have advanced rapidly, robots still struggle to operate reliably in real orchards because fruits are frequently hidden by leaves and branches, preventing robots from accurately identifying and accessing them. This project will develop ARMOR-Harvest, an intelligent robotic system that enables robots to actively overcome these environmental obstacles rather than simply observe them. By combining artificial intelligence, aerial and ground robots, and adaptive physical interaction, the system will allow robots to selectively clear obstructed views, coordinate harvesting actions, and operate effectively in complex orchard environments. The outcomes of this research will advance autonomous agricultural systems, improve the sustainability and resilience of food production, and establish new foundations for intelligent robots capable of performing tasks in challenging real-world settings. This project will develop ARMOR-Harvest (Aerial Robotic Morphing for Occlusion Resilient Apple Harvesting), a multi-robot framework that integrates learning-based decision-making, coordinated ground-aerial robotic planning, and morphable unmanned aerial vehicle (UAV) control to enable active occlusion mitigation using rotor airflow to clear leaf occlusions and a UAV-mounted hook to reposition obstructing branches. The research addresses a fundamental limitation in autonomous harvesting: the inability of current robotic systems to adaptively resolve visual and physical obstructions in unstructured agricultural environments. The project will pursue four integrated research thrusts. First, an attention-guided learning framework will combine 3D vision foundation models and reinforcement learning to determine when and how occlusion mitigation actions should be performed. Second, a heterogeneous multi-agent planning framework will integrate graph neural networks with optimization-based trajectory generation to enable real-time coordination between multi-arm harvesting robots and UAVs. Third, a morphable UAV platform with adaptive thrust allocation and task-aware control will be developed to perform precise leaf displacement and branch repositioning while maintaining robust flight stability. Finally, the complete system will be evaluated through simulation, laboratory experiments, and field demonstrations in commercial orchard environments. This research will advance fundamental capabilities in perception-driven decision-making, multi-robot coordination, and aerial manipulation, providing a new paradigm for deploying intelligent robotic systems in dynamic and unstructured environments. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.