Co-RoboPV: cooperative aerial robot system for photovoltaic power plant inspection based on reinforcement learning

Abstract Efficient inspection of large-scale photovoltaic (PV) plants can be significantly improved by employing a multi-aerial robot system. This article aims to develop an algorithm tailored to such a system for coverage path planning. Our approach involves determining the optimal number and types of aerial robots, designing a detailed flight plan, preparing the robots and their equipment, and analyzing the data collected from inspections. Autonomous operation is critical for enhancing efficiency and accuracy by eliminating direct human control of aerial robot flight. To this end, we utilize Deep Reinforcement Learning (DRL) to train aerial robots to navigate and perform inspection tasks effectively. The proposed algorithm focuses on optimizing path coverage to ensure comprehensive inspections of PV plants. Our methodology includes training aerial robots with extensive data to navigate safely and efficiently across the PV field. Results indicate that the DRL-based algorithm significantly improves coverage and operational efficiency in the inspection process, demonstrating the potential of advanced learning techniques for enhancing multi-aerial robot systems.