A Robotic Tomato Classification System Using Computer Vision and a Soft Gripper on an xArm6 Collaborative Robot

The development of intelligent technologies for analyzing and classifying agricultural products plays a key role in Agriculture 4.0. This is particularly relevant for tomato crops, where substantial morphological variability and a high susceptibility to mechanical damage call for precise and consistent evaluation methodologies. This study proposes a robotic classification system that integrates computer vision, an XArm6 collaborative robot, a SoftGripper, and an RS D435i camera to enable reliable detection, classification, and damage-free manipulation of tomatoes. The perception module uses an RGB-D camera and a YOLO model specifically trained for tomato detection, enabling precise estimation of the fruit's three-dimensional position. Ripeness is determined through HSV-space segmentation into four categories, while size is estimated from the apparent diameter combined with depth information. The end-effector, manufactured from Smooth-SIL® 940 food-grade silicone and validated through structural simulations in ANSYS, ensures soft and adaptive manipulation. The system achieved over 95 % accuracy in 8 of the 12 evaluated classes, demonstrating stable detection, classification, and grasping performance. Overall, these results highlight the potential of intelligent systems as an effective and scalable solution for tomato classification in modern agricultural settings.