SCARLETT: Towards scalable, structured and resource-efficient harvesting of lettuce with collaborative biomimetic robots

In recent years, there has been a significant shortage of labour in agriculture, particularly in manual and repetitive tasks that require a high level of dexterity, task-specific adaptivity and cognition. Robotic automation in such environments can have a multidimensional impact- from mitigating labour shortage, increasing productivity and efficiency in the harvesting workflows. This thesis focuses on scalable robotic harvesting of Lettuce in state-of-the-art deepwater pool hydroponic farm. The primary contributions of this thesis are 1) development of deep learning based 2D/3D perception system for precise identification/localization of plants in varying growing stages, picking/placing points in floats of different dimensions in noisy/unstructured conditions; 2) development of an ANN based adaptive motion controller to coordinate the CR5 and CR10 Cobot arms to automate three critical harvesting tasks- seedling transplantation, float handing and harvesting; 3) Embedding of the robotic framework into the farm operation taking into account food safety, scalability and human-robot collaboration. The SCARLETT framework has been evaluated in the state-of-the-art 1.1Ha hydroponic farm of JEPCO, producing 1.95 million plants annually. While this thesis focuses specifically on Lettuce, the framework itself is adaptable to other ‘crop types, tasks and growing environments’, hence opening further opportunities for robotics in Smart Farming.