Developing Robotic System for Harvesting Pineapples

This paper develops a robotic system to harvest pineapple autonomously. The system contains a machine vision unit, two robotic manipulators mounted on a platform, custom end-effectors, and an image-based harvesting control unit. The manipulators with Gantry 3DOF PPP configuration are geometrically optimized to move the end-effectors approaching pineapples. Each end-effector is actuated by pneumatic actuator and equipped with a cage-shaped gripper to fix the selective pineapple inside and a cutting device to cut its stalk. YOLOv3 approach is implemented for detecting and recognizing pineapple fruits that meet requirements for harvest. The experiment results demonstrate the success of pineapple recognition with 90.82% mAP. The 3D position of the recognized pineapples will be calculated and sent to the control system. The control system, including an industrial computer communicating with PLCs to conducts the manipulators and end-effectors to approach and d the recognized pineapples. The complete system has been tested on the experimental field-model. The success rate of pineapple harvesting is 95.55% and the average time is 12 seconds per one fruit. In the future, this system will be improved for automatic harvesting in real pineapple fields.