A gazebo-simulated approach for strawberry ripeness detection toward sustainable agriculture

Abstract Recent advancements in simulation technologies are supporting the efficient development of more sustainable agricultural practices by reducing resource-intensive physical testing and promoting environmentally conscious research practices. In this study, a simulation-based approach was used to develop and assess a perception pipeline for strawberries using a virtual plant environment, designed with 3D design software and implemented in the Gazebo simulator. Synthetic images depicting strawberries at different maturity stages were generated from the simulation. YOLO11n, YOLO11s, YOLO11m, and YOLO11l object detection models were used on this synthetic dataset to detect individual strawberries and classify their ripeness level. The models accurately identified and classified strawberries by ripeness, with the best performers of YOLO11m and YOLO11l achieving peak mAP scores of 0.933 and 0.932, respectively. Notably, this improved accuracy was associated with increased model size and reduced frame rates. These perception results are a foundation for developing motion planning algorithm of robotic arm in harvesting robot in the simulation.