A Heterogeneous Robotic Framework for Autonomous Solar Panel Inspection using Computer Vision and Path Optimization

This work presents a comprehensive and modular framework enabling effective collaboration between Unmanned Aerial Vehicles (UAVs) and Unmanned Ground Vehicles (UGVs) for autonomous inspection of photovoltaic (PV) farms. The framework integrates real-time visual perception, intelligent planning, and adaptive coordination through a hierarchical, multi-layered architecture. The UAV is equipped with a camera to perform the inspection. The framework uses the You Only Look Once Version 8 (YOLOv8) Convolutional Neural Network (CNN) for high-precision detection and localization of solar panels. A Genetic Algorithm (GA) is responsible for computing energy-efficient and time-optimal inspection paths based on detected targets. The UGV acts as a mobile support and recharging station, coordinated via a centralized Ground Control Station (GCS), enabling dynamic relocation in response to UAV battery status and mission demands. Communication and task allocation among agents are managed through the MAVLink protocol within the Robot Operating System (ROS) environment. The framework was validated in a realistic simulation using Gazebo, achieving an F1-score of 0.91 and a Precision-Recall (PR) area under the curve (AUC) of 0.9460 for solar panel detection. A large-scale analysis demonstrates that the mobile UGV strategy enables complete inspection coverage, while reducing total mission time by up to 71% compared to a fixed charging station baseline. Additionally, the GA reduces route length by approximately 30% relative to classical baselines in targeted re-inspection scenarios. These results highlight the efficiency and scalability of the proposed cooperative UAV–UGV framework for large-scale PV inspection.