Tomato Harvesting Robot Using Yolov8, Midas and Inverse Kinematics for Smart Farming Automation

Abstract

This project presents a cutting-edge tomato harvesting robot that combines advanced image processing and robotic manipulation to optimize the harvesting process. Manual tomato harvesting is labour-intensive and requires a significant workforce, especially during peak seasons. Labor shortages and rising wages increase production costs, making automation a more viable alternative. Human pickers may vary in their selection criteria, leading to inconsistent quality and harvesting of unripe tomatoes. Improper handling can cause bruising or mechanical damage, reducing the market value of tomatoes. Farmers struggle to track and optimize harvesting schedules, often leading to wastage or missed ripe tomatoes. An automated system with deep learning-based classification ensures uniform selection based on ripeness. A robotic system with a customized soft end-effector minimizes damage by ensuring gentle gripping and precise detachment. This robot uses a YOLOv8 model to detect tomatoes and determine if they are ripe, unripe or partially ripe. MiDaS depth estimation helps measure the exact position of tomatoes in 3D space for precise picking. A 6-DOF robotic arm moves towards the detected tomato using inverse kinematics for accuracy. A soft-gripper gently holds the tomato, while a scissor detaches it from the vine without damage. The robot moves on a track-based system, ensuring stable and smooth navigation in farms and greenhouses. The results demonstrate that the autonomous tomato harvesting robot is a feasible and effective solution for modern agriculture. By combining deep learning, computer vision and robotic automation, this system enhances yield efficiency, reduces post-harvest losses and promotes sustainable farming.