Jungin Kim,  Dongho Shin, Inhoon Jang, "Simulation-Based Apple Harvesting Perception Framework Using Synthetic Data from a Digital Farm", in Proc. 2025Eur. Kor. Conf. on Science and Technology (EKC 2025), Wien, Austria, Jun. 2025.

This study presents a simulation-based framework for vision-centric robotic apple harvesting, focusing on synthetic

data generation and object recognition in a virtual orchard environment. Rather than physically executing

harvesting tasks, our work emphasizes building a virtual apple farm environment (Apple Digital Farm) using Unity,

generating diverse synthetic datasets, and integrating with ROS 2 via TCP communication for mobile robot

simulation. By simulating a robot-mounted camera within the digital farm, synthetic images of occluded apples are

captured under varied perspectives and lighting conditions. These images are used to train and validate a YOLObased

apple recognition model, which performs real-time pose estimation (position and orientation) of fruits in the

simulation. This framework addresses challenges such as limited real-world data collection periods and the risk of

fruit damage in physical trials, offering a safe and scalable test environment for perception-based tasks in

agricultural robotics. The proposed approach demonstrates the potential of synthetic vision systems in agricultural

automation and provides a practical foundation for developing robust object detection methods without relying on

time-consuming field experiments.