A Deep Reinforcement Learning Approach for Multi-UAV Collaborative Coverage with Adaptive Step Size and Dynamic Reward Mechanism

In multi-UAV cooperative area coverage tasks, the effectiveness of path planning directly impacts coverage efficiency and mission completion time. Traditional reinforcement learning (RL) approaches suffer from limitations such as fixed action granularity and insufficient environmental perception, restricting their application in complex urban environments. To address these limitations, this paper proposes a multi-UAV deep reinforcement learning (DRL) area coverage method that integrates adaptive step size control and a dynamic reward mechanism. First, the method introduces adaptive step size control through a local obstacle density metric as a new state variable, which expands the action space. By employing Gaussian kernel weighting to evaluate obstacle distribution, the approach enhances environmental perception and enables adaptive multi-cell movement, thereby improving path selection sensitivity. The method then develops a dynamic segmented reward-penalty mechanism that accounts for both the interplay between step size and obstacle density, as well as UAV battery limitations, thereby optimizing decision-making. Experimental results demonstrate that the proposed approach significantly outperforms conventional fixed-step-size methods in terms of convergence speed, coverage rate, and path planning efficiency, providing a more robust and intelligent solution for multi-UAV cooperative coverage in complex scenarios.