Physically Realistic Digital Replica of Strawberry Plants for Robotic Harvesting

In agricultural robotics, creating realistic simulation environments is essential for developing and testing computer vision and control algorithms used in tasks like fruit detection, localisation, and harvesting. However, most existing digital plant models are designed only for visual purposes and lack the physical properties needed to simulate realistic robot-plant interactions. This limitation is particularly significant for strawberry harvesting, as these plants have complex, flexible structures that are prone to damage, which directly affects the success of automated picking. To address this gap, we introduce a methodology for generating physically accurate, harvestable, randomised strawberry plant models compatible with the common robotic simulator Gazebo Classic and Open Dynamic Engine (ODE), the high-performance library for simulating rigid body dynamics, within the ROS2 framework. The proposed models simulate dynamic behaviour using a simplified Euler-Bernoulli beam formulation, which approximates plant bending and vibration. These models are discretised into a simulation-ready format, with each plant segment represented as a rigid body, where joint properties define stiffness and damping. Our pipeline enables users to generate randomised yet biologically plausible strawberry plants, incorporating variations in leaf count, stem branching, fruit positioning, and biomechanical characteristics. According to our findings, the fracture plug-in accurately reproduced fruit detachment once normal and shear stress thresholds were reached, while stiffness validation tests yielded mean deflection errors of 32-40%, confirming the model’s reliability for robotic harvesting simulations.