Collaborative coverage trajectory planning for stratospheric airship via multi-agent reinforcement learning

In mission-critical areas requiring periodic data collection and transmission, a multi-stratospheric airship system offers significant advantages over single-airship systems by enabling collaborative coverage of large areas and improving data collection rates. This paper proposes a multi-agent reinforcement learning-based collaborative trajectory planning algorithm to optimize coverage efficiency and data transmission. By considering airship energy cycle constraints and the dynamic wind field environment, the algorithm achieves dynamic optimized scheduling, enhancing coverage through real-time adjustments to environmental changes and ensuring timely data collection. Coverage data is organized into a dynamically updating the information matrix for efficient coverage management and timely revisits to previously covered areas. To foster collaboration, a heterogeneous reward fusion mechanism that combines individual and team-based rewards, accelerating coverage optimization and data collection. Experimental results demonstrate that the proposed algorithm outperforms other multi-agent algorithms under the same settings, achieving a success rate of 95% across multiple test scenarios. This highlights its effectiveness in collaborative coverage tasks under multi-source constraints and robustness in real-world wind fields.