Research on UAV data acquisition trajectory based on deep reinforcement learning

Drones offer advantages such as low costs, flexible operations, and high safety, leading to their widespread application in various fields including search and rescue, military operations, and agriculture. Drones can achieve data collection and transmission through communication systems, sending data to ground stations for processing and analysis. Research on drone communication mainly includes aspects such as path planning, trajectory optimization, resource allocation, and resource optimization. However, it is challenging to obtain the end-to-end channel models and parameters required for optimization in practice. Moreover, even with accurate channel models and all relevant parameter information, most offline optimization problems are highly non-convex and difficult to solve effectively. To overcome this challenge, this paper employs Deep Reinforcement Learning (DRL) techniques to address the optimal decision-making problem of drones in complex dynamic environments. This study considers the possibility of no-fly zones in areas where drones perform tasks, which may include complex Line of Sight (LoS) and Non-Line of Sight (NLoS) channel states. To overcome these limitations, this research proposes a trajectory design problem with the objective of minimizing flight time while collecting data in designated areas under no-fly zone constraints, utilizing a Dueling Double Deep Q-Networks (DDQN) algorithm based on multi-step learning. The final simulation results demonstrate that this algorithm exhibits good performance in practical scenarios.