UAV trajectory planning method based on dynamic window approach and adaptive deep Q-networks

To address the limitations of deep Q-networks (DQNs) in effectively navigating dynamic threats in challenging environments, we propose a novel UAV trajectory planning method that integrates the dynamic window approach (DWA) with adaptive deep Q-networks (ADQN-PDWA). The UAV operating environment is discretised and modelled as a grid map. To overcome the low success rate of trajectory planning in scenarios where both environmental states and actions are discrete, we introduce a global planning method based on adaptive DQN. This method improves the efficiency of global trajectory planning by leveraging adaptive mechanisms to optimise the reward function and action selection strategies. Additionally, we design a locally focused dynamic window trajectory planning method with environment-awareness, ensuring real-time performance and trajectory smoothness. By seamlessly integrating this local planning approach with the global adaptive DQN framework, the method enables UAVs to effectively avoid dynamic threats. Experimental results show that the proposed method achieves shorter trajectories and significantly enhances trajectory planning success rates in complex environments with varying threat densities.