Pekiştirmeli öğrenme kullanarak eşzamanli alan tarama ve bağlanirlik i̇çin çok iha'li rota planlamasi

This thesis presents a multi-objective optimization (MOO) framework in multi-Unmanned Aerial Vehicle (UAV) path planning, transitioning from classical optimization methods to advanced reinforcement learning techniques. We address two main challenges in multi-UAV systems: the need for real-time adaptability and the balance between competing mission objectives. First, we develop a new multi-objective optimization framework that simultaneously considers coverage time and network connectivity. This framework provides a diverse set of Pareto-optimal solutions, highlighting meaningful trade-offs between different mission goals. Building on these findings, we introduce a dynamic reinforcement learning system using Deep Q-Networks that allows UAVs to adjust their behavior according to changing mission requirements. This system demonstrates improvement in target detection times. Finally, we present an advanced Proximal Policy Optimization framework that effectively manages heterogeneous targets with varying information priorities. This framework includes a dynamic reward mechanism that integrates area exploration, UAV-Ground Control Station (GCS) connectivity, and target-specific requirements. Our experimental results show that the proposed framework remains computationally efficient while scaling to larger teams of up to 20 UAVs and adapting to different target configurations. The progression from static optimization through basic reinforcement learning to advanced adaptive systems not only enhances the technical capabilities of multi-UAV systems but also offers valuable implementation insights for researchers and practitioners in the field.