Modern agricultural practices heavily depend on active compounds like nutrients. For tablegrape cultivation, there has been a growing utilization of plant hormones as nutrients to enhance berry size and overall yield. However, the current approach of manually delivering these hormones poses significant challenges. The process is time-consuming, laborintensive, prone to inaccuracies, and frequently results in overdoses, presenting significant obstacles to effective grape cultivation.

This thesis is a part of a funded research focused on developing a robotic system for accurately and selectively delivering the nutrients to grape clusters in vineyards. The suggested system consists of a manipulator based on a mobile platform, equipped with multiple sensors and a spraying nozzle .

The objective of this study is to identify the optimal manipulator design to perform the selective spray. To address this optimization challenge, a novel methodology was developed. This methodology integrates various components, including a robotic simulator, the Particle Swarm Optimization (PSO) algorithm, and machine learning models based on the XGBoost architecture. The simulator was developed using a robotic software to assess the performance of each potential robotic configuration in a simulated vineyard. The optimization process utilized the PSO, a bio-inspired algorithm, within a constrained solution space. Complementing these tools, machine learning models, XGBoost architecture, were designed to predict the performance of robotic configurations in order to reduce the use of the simulator. These models improved the optimization process and decreased runtime.

The presented methodology resulted in several reasonable optimal configurations of manipulators for this task. This approach can be applied to other agricultural tasks, with potential for use in other industries. This study demonstrates the effectiveness of using AI techniques in robotics optimization to address current agricultural challenges.