Artificial intelligence-based swarm robots for growing agricultural plants and usage in agricultural control

In this Study, an artificial intelligence-supported swarm drone system was developed and modeled in the Gazebo simulation environment for use in cultivating agricultural plants and their protection against pests. The study's main objective is to integrate the components of mission planning, object detection, and route optimization of swarm robots capable of performing autonomous tasks in agricultural production processes within a unified architecture. The developed system features a multi-drone structure capable of simultaneously performing data collection and spraying tasks. The explorer drone scans the agricultural field within the swarm using sensor data to collect images. These images are processed by a deep learning model based on YOLO (You Only Look Once), and the detected target locations are transmitted to the spraying drones. Thus, precise and target-oriented pesticide application is achieved in areas where plants are located. The system architecture integrates the Robot Operating System (ROS), ArduPilot, MAVLink, and MAVROS protocols to ensure multi-drone communication, data flow, and mission synchronization. For mission planning, Ant Colony Optimization (ACO) and Nearest Neighbor (NN) algorithms were employed together for route optimization, resulting in significant improvements in flight distance, battery SOC value, and mission duration. In the non-optimized scenario, the total flight distance was 960 m, which decreased by 10% to 865 m with the NN algorithm and by 13.4% to 831 m with the ACO algorithm. In parallel, the battery SOC values were measured at 63%, 65%, and 66%, respectively, and a 4.54% advantage was achieved with ACO. Similarly, flight times were recorded as 491.39 s, 470.34 s, and 447.28 s, respectively, with an improvement of approximately 9% achieved with ACO. In the object detection module, YOLOv3 and YOLOv5 models were comparatively evaluated, and the highest accuracy was achieved with the YOLOv5x model (99.5% mAP@0.5). However, to ensure real-time operation of the system, the computationally efficient YOLOv3 architecture was preferred. This way, hardware resource utilization was reduced while maintaining sufficient detection accuracy. Moreover, in emergency scenarios where one drone becomes inoperative, the automatic reassignment of tasks to the remaining drones enhances the system's reliability and continuity. The artificial intelligence-supported swarm drone architecture developed within this thesis's scope has significantly improved speed, accuracy, energy efficiency, and operational safety for agricultural spraying, crop monitoring, and data collection processes. The integrated execution of object detection, task allocation, and route optimization has produced an autonomous, scalable solution that minimizes human intervention in agricultural production. The results demonstrate that the coordinated motion capability of swarm robots optimizes resource utilization and enhances operational continuity. Furthermore, the modular design developed in an open-source simulation environment enables easy system adaptation to different hardware platforms. This study provides a robust engineering foundation for the widespread use of swarm robots in future autonomous agricultural systems.