Unmanned Aerial Vehicle Path Planning Based on Improved DDQN Algorithm

Unmanned Aerial Vehicles(UAVs) is intelligent devices with flexibility and adaptability that have been widely applied in various fields.UAV path planning refers to the process of planning the flight routes of drones in three-dimensional space in order to accomplish specific tasks or follow specific rules. Traditional UAV path planning algorithms often fail to fully consider the dynamics and real-time nature of flight in complex environments and tasks. This paper conducts relevant research on UAV path planning tasks based on deep reinforcement learning algorithms. Firstly, a three-dimensional aerial environment is constructed based on real wind field data to bridge the gap between simulation and reality. Second, a composite action selection strategy is proposed to prevent UAVs from getting stuck in local optima during the path planning process, thus improving the convergence speed and average cumulative reward of the algorithm. Finally, a distributed value network is introduced to allow UAVs to fully utilize value information during path planning, reducing oscillation during the learning process and improving the accuracy and stability of path planning. Experimental results demonstrate that the VD-DDQN algorithm performs exceptionally well in UAV path planning tasks across various scenarios.