A Multi-UAV Task Allocation Method Based on Multi-objective Cross-attention Reinforcement Learning

Xingchen Ge; Lulu Chen; Xinshang Qin; Pengfei Zheng; Zhenbao Liu

doi:10.1007/978-981-95-7656-2_13

A Multi-UAV Task Allocation Method Based on Multi-objective Cross-attention Reinforcement Learning

Xingchen Ge
, Lulu Chen
, Xinshang Qin
, Pengfei Zheng
, Zhenbao Liu

School of Civil Aviation

Northwestern Polytechnical University Xian
Ltd.

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

Abstract

This paper proposes a Multi-Objective Cross-Attention (MOCA) reinforcement learning model based on PPO, tailored for multi-UAV task allocation. By leveraging a multi-head cross-attention mechanism, it effectively captures complex dependencies between UAVs and tasks, optimizing multiple objectives such as minimizing the longest flight time, reducing total travel distance, and balancing task load. MOCA overcomes limitations of traditional heuristic and optimization methods, including poor adaptability to dynamic environments and difficulty handling complex constraints. Extensive simulations show that MOCA outperforms Particle Swarm Optimization, greedy algorithms, and Multi-Agent PPO in flight time control, path efficiency, and load balancing. Ablation studies confirm the crucial role of the multi-channel encoder and cross-attention modules, demonstrating MOCA’s robustness and potential for real-world multi-UAV coordination and mission planning.

Original language	English
Title of host publication	Proceedings of 5th 2025 International Conference on Autonomous Unmanned Systems, ICAUS - Volume 5
Editors	Shaorong Xie, Yifeng Niu, Wenxing Fu, Yi Qu
Publisher	Springer Science and Business Media Deutschland GmbH
Pages	132-141
Number of pages	10
ISBN (Print)	9789819576555
DOIs	https://doi.org/10.1007/978-981-95-7656-2_13
State	Published - 2026
Event	5th International Conference on Autonomous Unmanned Systems, ICAUS 2025 - Shanghai, China Duration: 17 Oct 2025 → 19 Oct 2025

Publication series

Name	Lecture Notes in Electrical Engineering
Volume	1578 LNEE
ISSN (Print)	1876-1100
ISSN (Electronic)	1876-1119

Conference

Conference	5th International Conference on Autonomous Unmanned Systems, ICAUS 2025
Country/Territory	China
City	Shanghai
Period	17/10/25 → 19/10/25

Keywords

PPO
cross-attention
multi-objective
task allocation

Access to Document

10.1007/978-981-95-7656-2_13

Cite this

Ge, X., Chen, L., Qin, X., Zheng, P., & Liu, Z. (2026). A Multi-UAV Task Allocation Method Based on Multi-objective Cross-attention Reinforcement Learning. In S. Xie, Y. Niu, W. Fu, & Y. Qu (Eds.), Proceedings of 5th 2025 International Conference on Autonomous Unmanned Systems, ICAUS - Volume 5 (pp. 132-141). (Lecture Notes in Electrical Engineering; Vol. 1578 LNEE). Springer Science and Business Media Deutschland GmbH. https://doi.org/10.1007/978-981-95-7656-2_13

Ge, Xingchen ; Chen, Lulu ; Qin, Xinshang et al. / A Multi-UAV Task Allocation Method Based on Multi-objective Cross-attention Reinforcement Learning. Proceedings of 5th 2025 International Conference on Autonomous Unmanned Systems, ICAUS - Volume 5. editor / Shaorong Xie ; Yifeng Niu ; Wenxing Fu ; Yi Qu. Springer Science and Business Media Deutschland GmbH, 2026. pp. 132-141 (Lecture Notes in Electrical Engineering).

@inproceedings{2d790c75994f4b77a043b6f69debfac2,

title = "A Multi-UAV Task Allocation Method Based on Multi-objective Cross-attention Reinforcement Learning",

abstract = "This paper proposes a Multi-Objective Cross-Attention (MOCA) reinforcement learning model based on PPO, tailored for multi-UAV task allocation. By leveraging a multi-head cross-attention mechanism, it effectively captures complex dependencies between UAVs and tasks, optimizing multiple objectives such as minimizing the longest flight time, reducing total travel distance, and balancing task load. MOCA overcomes limitations of traditional heuristic and optimization methods, including poor adaptability to dynamic environments and difficulty handling complex constraints. Extensive simulations show that MOCA outperforms Particle Swarm Optimization, greedy algorithms, and Multi-Agent PPO in flight time control, path efficiency, and load balancing. Ablation studies confirm the crucial role of the multi-channel encoder and cross-attention modules, demonstrating MOCA{\textquoteright}s robustness and potential for real-world multi-UAV coordination and mission planning.",

keywords = "PPO, cross-attention, multi-objective, task allocation",

author = "Xingchen Ge and Lulu Chen and Xinshang Qin and Pengfei Zheng and Zhenbao Liu",

note = "Publisher Copyright: {\textcopyright} Beijing HIWING Scientific and Technological Information Institute 2026.; 5th International Conference on Autonomous Unmanned Systems, ICAUS 2025 ; Conference date: 17-10-2025 Through 19-10-2025",

year = "2026",

doi = "10.1007/978-981-95-7656-2\_13",

language = "英语",

isbn = "9789819576555",

series = "Lecture Notes in Electrical Engineering",

publisher = "Springer Science and Business Media Deutschland GmbH",

pages = "132--141",

editor = "Shaorong Xie and Yifeng Niu and Wenxing Fu and Yi Qu",

booktitle = "Proceedings of 5th 2025 International Conference on Autonomous Unmanned Systems, ICAUS - Volume 5",

}

Ge, X, Chen, L, Qin, X, Zheng, P & Liu, Z 2026, A Multi-UAV Task Allocation Method Based on Multi-objective Cross-attention Reinforcement Learning. in S Xie, Y Niu, W Fu & Y Qu (eds), Proceedings of 5th 2025 International Conference on Autonomous Unmanned Systems, ICAUS - Volume 5. Lecture Notes in Electrical Engineering, vol. 1578 LNEE, Springer Science and Business Media Deutschland GmbH, pp. 132-141, 5th International Conference on Autonomous Unmanned Systems, ICAUS 2025, Shanghai, China, 17/10/25. https://doi.org/10.1007/978-981-95-7656-2_13

A Multi-UAV Task Allocation Method Based on Multi-objective Cross-attention Reinforcement Learning. / Ge, Xingchen; Chen, Lulu; Qin, Xinshang et al.
Proceedings of 5th 2025 International Conference on Autonomous Unmanned Systems, ICAUS - Volume 5. ed. / Shaorong Xie; Yifeng Niu; Wenxing Fu; Yi Qu. Springer Science and Business Media Deutschland GmbH, 2026. p. 132-141 (Lecture Notes in Electrical Engineering; Vol. 1578 LNEE).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

TY - GEN

T1 - A Multi-UAV Task Allocation Method Based on Multi-objective Cross-attention Reinforcement Learning

AU - Ge, Xingchen

AU - Chen, Lulu

AU - Qin, Xinshang

AU - Zheng, Pengfei

AU - Liu, Zhenbao

N1 - Publisher Copyright: © Beijing HIWING Scientific and Technological Information Institute 2026.

PY - 2026

Y1 - 2026

N2 - This paper proposes a Multi-Objective Cross-Attention (MOCA) reinforcement learning model based on PPO, tailored for multi-UAV task allocation. By leveraging a multi-head cross-attention mechanism, it effectively captures complex dependencies between UAVs and tasks, optimizing multiple objectives such as minimizing the longest flight time, reducing total travel distance, and balancing task load. MOCA overcomes limitations of traditional heuristic and optimization methods, including poor adaptability to dynamic environments and difficulty handling complex constraints. Extensive simulations show that MOCA outperforms Particle Swarm Optimization, greedy algorithms, and Multi-Agent PPO in flight time control, path efficiency, and load balancing. Ablation studies confirm the crucial role of the multi-channel encoder and cross-attention modules, demonstrating MOCA’s robustness and potential for real-world multi-UAV coordination and mission planning.

AB - This paper proposes a Multi-Objective Cross-Attention (MOCA) reinforcement learning model based on PPO, tailored for multi-UAV task allocation. By leveraging a multi-head cross-attention mechanism, it effectively captures complex dependencies between UAVs and tasks, optimizing multiple objectives such as minimizing the longest flight time, reducing total travel distance, and balancing task load. MOCA overcomes limitations of traditional heuristic and optimization methods, including poor adaptability to dynamic environments and difficulty handling complex constraints. Extensive simulations show that MOCA outperforms Particle Swarm Optimization, greedy algorithms, and Multi-Agent PPO in flight time control, path efficiency, and load balancing. Ablation studies confirm the crucial role of the multi-channel encoder and cross-attention modules, demonstrating MOCA’s robustness and potential for real-world multi-UAV coordination and mission planning.

KW - PPO

KW - cross-attention

KW - multi-objective

KW - task allocation

UR - https://www.scopus.com/pages/publications/105038881332

U2 - 10.1007/978-981-95-7656-2_13

DO - 10.1007/978-981-95-7656-2_13

M3 - 会议稿件

AN - SCOPUS:105038881332

SN - 9789819576555

T3 - Lecture Notes in Electrical Engineering

SP - 132

EP - 141

BT - Proceedings of 5th 2025 International Conference on Autonomous Unmanned Systems, ICAUS - Volume 5

A2 - Xie, Shaorong

A2 - Niu, Yifeng

A2 - Fu, Wenxing

A2 - Qu, Yi

PB - Springer Science and Business Media Deutschland GmbH

T2 - 5th International Conference on Autonomous Unmanned Systems, ICAUS 2025

Y2 - 17 October 2025 through 19 October 2025

ER -

Ge X, Chen L, Qin X, Zheng P, Liu Z. A Multi-UAV Task Allocation Method Based on Multi-objective Cross-attention Reinforcement Learning. In Xie S, Niu Y, Fu W, Qu Y, editors, Proceedings of 5th 2025 International Conference on Autonomous Unmanned Systems, ICAUS - Volume 5. Springer Science and Business Media Deutschland GmbH. 2026. p. 132-141. (Lecture Notes in Electrical Engineering). doi: 10.1007/978-981-95-7656-2_13