Design, construction, and evaluation of an autonomous robot for harvesting damask roses, equipped with a specialized end-effector
Year: 2025
Type: article
Abstract: Despite the high commercial value of Damask roses, harvesting remains a manual process, posing significant challenges for related industries. Given the uneven blooming of flowers and their scattered distribution on the plant, automating the harvesting process requires the development of an autonomous robot capable of selectively picking bloomed flowers. Therefore, this study designed and constructed a novel harvesting robot that detects the position of bloomed Damask roses using the YOLOv9t (You Only Look Once version 9 tiny) model, optimized through a knowledge distillation method based on offline responses. To develop YOLOv9t as an object detection model, all images were collected from the field under real-world conditions, achieving an accuracy of 96.2 %. However, the robot's performance metrics were evaluated in a laboratory setting. The robot consists of four main mechanisms: three independent rail-based linear movement mechanisms, an advanced end-effector, a TCS3200 color detection sensor module, and a rotary blade mounted at the tips of the fingers. Laboratory evaluations of the prototype robot for Damask rose harvesting indicated that the average detection rate, grabbing rate, cutting rate, picking rate, harvesting rate, damage rate, and false positive detection rate were 97.4 %, 93.2 %, 88.4 %, 82.4 %, 80.2 %, 8.0 %, and 10.5 %, respectively. Additionally, the average time required to harvest a single flower was 39.65 s, with 6 % of this time allocated to detecting bloomed flowers and approximately 65 % dedicated to the movement of the end-effector toward the flower. The promising results of this study offer valuable insights for the development of a robotic system capable of harvesting Damask roses under real-world conditions. • Design, construction, and evaluation of an autonomous robot for harvesting roses. • Development of a specialized end-effector for picking roses. • Real-time detection of fully bloomed roses using YOLOv9t (OFD). • Detection of the spatial coordinates of flower centers using an RGB camera and a TCS3200 sensor.
Authors: Farhad Fatehi, Hossein Bagherpour
Institution: Bu-Ali Sina University
Language: English
Topic: Smart Agriculture and AI
Subfield: Plant Science
Domain: Life Sciences
Open Access status: gold
Estimated APC paid: $2,530
Funder: Bu Ali Sina University
Award: G1314784140
OpenAlex ID: W4414979893