A Scalable Multi-UUVs Collaborative Multi-Objective Area Protection Framework Using Multi-Task MADRL
IEEE Transactions on Mobile Computing · 2026-09 · 25 (9) · 13490-13507
DOI: 10.1109/tmc.2026.3673341 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Transactions on Mobile Computing, 2026-09. Verfasst von Wenhao Gan und 2 weiteren Autor:innen. Die Arbeit wurde bisher 1-mal zitiert (Crossref) und nennt 35 Literaturangaben. Gefördert von: National Natural Science Foundation of China, Natural Science Foundation of Shanghai.
Abstract (Originalsprache)
In area protection tasks of unmanned underwater vehicles (UUVs), to counter attackers’ high maneuverability and varying tendencies, defenders must perform targeted maneuvers. To address this problem, the paper presents a multi-task multi-agent deep reinforcement learning (MT-MADRL) framework, considering obstacle-dense environments and limited perception and communication. Firstly, a hierarchical maneuvering framework is designed to enhance inter-layer coupling, enabling defenders to learn from teammates and opponents while promoting cross-subteam cooperation. A hybrid reward mechanism is further introduced to facilitate multi-task policy learning. Then, a customized network architecture with weighted soft attention, recurrent networks, and hypernetworks is developed to extract task differences and commonalities, enabling defenders to learn interception and repulsion skills while adapting to varying game scales. Complementary network updates and training methods are introduced to enhance UUV cooperation and stabilize the training process. Finally, a Sim2Real transfer validation method facilitates the deployment of DRL strategies in real-world scenarios. Ablation and comparative studies validate the superior learning capability of our scheme over mainstream DRL variants. Comparisons with recent defense strategies, together with tests across different many-to-many game settings, diverse attacker strategies, and various communication and perception configurations, demonstrate the proposed scheme’s strong multi-task execution capability and robustness. Complexity analysis and field experiments confirm its practical feasibility.
Förderung
- National Natural Science Foundation of China
- Natural Science Foundation of Shanghai
Zitieren
APA
Gan, W., Guo, K., & Qiao, L. (2026). A Scalable Multi-UUVs Collaborative Multi-Objective Area Protection Framework Using Multi-Task MADRL. IEEE Transactions on Mobile Computing, 25(9), 13490-13507. https://doi.org/10.1109/tmc.2026.3673341
Vancouver
Gan W, Guo K, Qiao L. A Scalable Multi-UUVs Collaborative Multi-Objective Area Protection Framework Using Multi-Task MADRL. IEEE Transactions on Mobile Computing. 2026;25(9):13490-13507. doi:10.1109/tmc.2026.3673341
BibTeX
@article{Gan2026,
author = {Gan, Wenhao and Guo, Kai and Qiao, Lei},
title = {A Scalable Multi-UUVs Collaborative Multi-Objective Area Protection Framework Using Multi-Task MADRL},
journal = {IEEE Transactions on Mobile Computing},
year = {2026},
volume = {25},
number = {9},
pages = {13490--13507},
doi = {10.1109/tmc.2026.3673341}
}
EUVIMED