A Proposal of FPGA-Based Low Cost and Power Efficient Autonomous Fruit Harvester

In this paper, we present a power-efficient and low- cost prototype of a robotic harvester which employs multiple subsystems such as fruit detection, odometry, localization, profi- cient manipulation through computer vision, deep learning and a novel end-effector design. Fruit Plucking is performed using an end effector, and 3-degree of freedom (DOF) arm (made out of the integration of two linear actuators and a rotating platform) consolidated with a 4-wheeled differential drive mobile platform. Effective implementation of the visual processing is executed on the FPGA Fabric of the Xilinx PYNQ-Z2 Board, which accelerates Deep Neural Networks (DNNs) with improved Latency and Energy Efficiency as compared to a CPU or GPU based implementation.