FROM HARVEST TO PACKAGE: AN AUTONOMOUS ROBOT FOR INTEGRATED TOMATO PICKING AND BAGGING

Abstract

In addressing the high labor costs and low operational efficiency of greenhouse tomato harvesting and separate packaging workflows, this study develops an integrated tomato harvesting robotic system embedded with RGB-D machine vision, 5-degree-of-freedom manipulator and vertical heat-sealing net bag packaging mechanism. The YOLOv8s lightweight detection model trained on self-built multi-light greenhouse tomato dataset (1260 annotated images covering unobstructed, semi-occluded and heavily occluded fruits) is adopted to identify ripe tomatoes with a recognition accuracy of 95.2%, and ImageJ software is introduced to conduct secondary maturity screening via RGB chromatographic analysis. A* global path planning combined with TEB local trajectory optimization realizes autonomous obstacle avoidance navigation of the wheeled mobile platform, while RRT-Connect bidirectional random tree algorithm is applied for obstacle-free grasping trajectory planning inside dense tomato canopies. A total of 120 valid cyclic tests are carried out in simulated greenhouse environment to verify the full-chain automation including fruit detection, in-situ picking and instant bagging. Experimental results show that the average single-fruit processing cycle is 12.1 s, with a picking success rate of 90.8% and bagging success rate of 98.3%. Compared with skilled manual picking and packaging, the overall working efficiency is improved by approximately 30%. This system firstly realizes continuous integrated harvesting and commercial packaging operation for greenhouse tomatoes, providing a feasible technical solution for full-process intelligent protected agriculture.