Collaborative Coverage of Heterogeneous Unmanned Swarms Based on Deep Reinforcement Learning in Signal Interference Environments

Heterogeneous unmanned swarms, composed of unmanned aerial vehicles (UAVs) and unmanned ground vehicles (UGVs), play a pivotal role in real-world missions such as environmental exploration and disaster area assessment. However, unknown signal interference sources can severely disrupt intraswarm communication, leading to interrupted information sharing and redundant coverage, which in turn degrades mission completion efficiency. To enhance the collaborative exploration and coverage capabilities of heterogeneous swarms in unknown interference environments, this study proposes an anti-interference heterogeneous multiagent reinforcement learning algorithm based on QMIX (AIH-QMIX). First, a state-enhanced gated fusion LSTM-attention (SG-LSTMA) model is designed to effectively extract key temporal features from partially observable environments, providing high-quality state representations for decision-making. Second, a continuous reward function based on multidimensional potential field collaboration is introduced to prevent communication interruptions by constructing a potential field model, thereby enhancing the stability of communication links. Finally, an intrinsic reward function based on action consistency is formulated to guide UAVs to maneuver collaboratively with UGV after communication interruptions, accelerating communication restoration and improving mission continuity. Simulation results demonstrate that the proposed algorithm significantly enhances regional coverage efficiency, communication reliability, and recovery capabilities under varying interference intensities and swarm sizes, providing an effective technical framework for collaborative exploration by heterogeneous unmanned swarms in complex electromagnetic environments.