Agentic AI-Native Collaborative Task Offloading and Resource Orchestration for 6G Space–Air-Integrated Computing Power Networks

Haoxiang Luo; Gang Sun; Long Luo; Hongfang Yu; Hongke Zhang

doi:10.1016/j.eng.2026.08.022

Engineering ›› :202608022 DOI: 10.1016/j.eng.2026.08.022

research-article

Agentic AI-Native Collaborative Task Offloading and Resource Orchestration for 6G Space–Air-Integrated Computing Power Networks

Author information +

History +

PDF (2695KB)

Abstract

The advent of the sixth-generation (6G) wireless communication era heralds a paradigm shift from terrestrial-centric connectivity to a three-dimensional, ubiquitous ecosystem known as the space–air-integrated computing power network (SAICPN). This architecture integrates low-Earth-orbit satellites and unmanned aerial vehicles (UAVs) with terrestrial networks to provide seamless computation offloading and edge intelligence services to ground users, particularly in underserved remote regions and emerging low-altitude economies. However, the realization of SAICPN is hindered by severe resource orchestration challenges. These challenges arise from the high-speed orbital dynamics of satellites, the nonlinear aerodynamic energy constraints of UAVs, and the intricate dependency structures of modern computational tasks modeled as directed acyclic graphs. Conventional centralized optimization schemes incur prohibitive signaling overhead, while traditional multi-agent reinforcement learning frameworks struggle with the instability and partial observability inherent in such large-scale, heterogeneous networks. This paper proposes a novel agentic artificial intelligence framework: the generative agentic interface-based multi-agent deep deterministic policy gradient (GAI-MADDPG) algorithm. We mathematically model the computation offloading and resource pricing problem as a hierarchical Stackelberg game, providing a rigorous proof of the existence and uniqueness of the equilibrium using backward induction and Hessian matrix analysis. Uniquely, our framework integrates variational autoencoders to compress high-dimensional, continuous environmental states into robust latent representations, enabling proactive, intent-based agentic decision-making. Extensive simulations demonstrate that GAI-MADDPG significantly outperforms comparison approaches, achieving a 15%–20% reduction in energy consumption and a 96.5% task completion rate under dynamic, constraints-intensive 6G scenarios. Ablation studies corroborate that each core module makes an indispensable contribution to overall performance.

Keywords

Sixth-generation wireless communication / Space–air-integrated computing power network / Agentic artificial intelligence / Multi-agent reinforcement learning / Non-terrestrial networks

Cite this article

Download citation ▾

Haoxiang Luo, Gang Sun, Long Luo, Hongfang Yu, Hongke Zhang. Agentic AI-Native Collaborative Task Offloading and Resource Orchestration for 6G Space–Air-Integrated Computing Power Networks. Engineering 202608022 DOI:10.1016/j.eng.2026.08.022

登录浏览全文

4963

注册一个新账户忘记密码

References

Publishing order | Descend order by publishing year | Descend order by cited within

[1]	Luo H, Sun G, Wang J, Yu H, Niyato D, Dustdar S, et al. Wireless blockchain meets 6G: the future trustworthy and ubiquitous connectivity. IEEE Commun Surv Tutor 2026; 28: 3596-36.

[2]	Jamshed MA, Kaushik A, Dajer M, Guidotti A, Parzysz F, Lagunas E, et al. Non—terrestrial networks for 6G: integrated, intelligent and ubiquitous connectivity. IEEE Commun Stand Mag 2025; 9(3): 86-93.

[3]	Zhou H, Liu X, Zhang X, Qin X, Zhang M, Liu Y, et al. High fidelity and efficiency simulator for 6G integrated space—ground network. Engineering 2026; 56: 62-78.

[4]	Sun Y, Liu Y, Guo S, Wang Z . Toward integrated air—ground computing and communications: a synergy of computing power networks and low—altitude economy network. 2025. arXiv:2511.18720.

[5]	Du M, Liu Y, Zhao C, Tian W, Shi H, Han Z . GDP—UAV: a Gaussian differential privacy—empowered multi—UAV task offloading in low—altitude economy networks. IEEE Trans Cogn Commun Netw 2025; 12: 5001-13.

[6]	Ren J, Sun Y, Du H, Yuan W, Wang C, Wang X, et al. Generative semantic communication: architectures, technologies, and applications. Engineering 2026; 56: 45-61.

[7]	Zhang R, Liu G, Liu Y, Zhao C, Wang J, Xu Y, et al. Toward edge general intelligence with agentic AI and agentification: concepts, technologies, and future directions. IEEE Commun Surv Tutor 2026; 28: 4285-318.

[8]	Luo H, Zhang R, Liu Y, Sun G, Wang J . Real world assets on—chain assistance low—altitude computility networks: architecture, methodology, and challenges. IEEE Internet Things Mag. 2026. arXiv:2508.17911.

[9]	Shang X, Gao D, Yang D, Li J, Zhang T, Liu S, et al. Generative coflow scheduling for cross—silo federated large language models synchronization in computing power networks. IEEE Trans Netw Sci Eng 2025; 13: 3992-4007.

[10]	Hsu YH, Phan TTT . A DRL—based energy—efficient service caching and task offloading scheme for 6G MEC SAGINs. IEEE Trans Commun 2025; 73(12): 13967—82.

[11]	Luo L, Zhang C, Yu H, Li Z, Sun G, Luo S . Energy—efficient hierarchical collaborative learning over LEO satellite constellations. IEEE J Sel Areas Commun 2024; 42(12): 3366-79.

[12]	Kuang L, Shi Y, Liu K, Jiang C . Space computing power networks: fundamentals and techniques. Engineering 2025; 54: 26-34.

[13]	Zhu S, Han G, Lin C, Chen C, Wang Z, Yang F, et al. Smart multi—scenario task deployment for AUV cluster network: a large language model—driven exploration—enhanced MARL approach. IEEE Trans Mob Comput 2026; 25(7): 10507-24.

[14]	Tao S, Yuan M, Wu Q, Wang R, Hao J . Generative AI—aided vertical handover decision in SAGIN for IoT with integrated sensing and communication. IEEE Internet Things J 2025; 12(15): 13297-310.

[15]	Gronauer S, Diepold K . Multi—agent deep reinforcement learning: a survey. Artif Intell Rev 2022; 55(2): 895-943.

[16]	Tang J, Peng S, Guo J, Song D, Gao D, Liu W, et al. DT and LLM driven intelligent maintenance system for L—DED and DAG—based LLM fault diagnosis evaluation framework. Appl Soft Comput 2025; 185: 113942.

[17]	Sun Y, Liu Y, Guo S, Li H . IGAA: intent—driven general agentic AI for edge services scheduling using generative meta learning. 2026. arXiv:2601.13702.

[18]	Luo H, Sun G, Liu Y, Wang J . A trustworthy agentic multi—LLM network: challenges, solutions, and a use case. IEEE Wirel Commun. In press.

[19]	Chen Q, Meng W, Quek TQS, Chen S . Multi—tier hybrid offloading for computation—aware IoT applications in civil aircraft—augmented SAGIN. IEEE J Sel Areas Commun 2023; 41(2): 399-417.

[20]	Zhang Y, Wang X, Gang Y, Li J, Liu T . Evolutionary game resource distribution strategy for 6G SAGIN. IEEE Trans Veh Technol 2025; 75(6): 11234-49.

[21]	Qin X, Zhang T, Yu K, Zhang X, Zhou H, Zhuang W, et al. Dynamic time—difference QoS guarantee in satellite—terrestrial integrated networks: an online learning—based resource scheduling scheme. Engineering 2025; 54: 127-42.

[22]	Gao Y, Ye Z, Yu H . Cost—efficient computation offloading in SAGIN: a deep reinforcement learning and perception—aided approach. IEEE J Sel Areas Commun 2024; 42(12): 3462-76.

[23]	Liu Y, Jiang L, Qi Q, Xie K, Xie S . Online computation offloading for collaborative space/aerial—aided edge computing toward 6G system. IEEE Trans Veh Technol 2023; 73(2): 2495-505.

[24]	Betalo ML, Ullah I, Tesema FB, Lee S, Kim J, Park J, et al. Generative AI—driven multi—agent DRL for task allocation in UAV—assisted EMPD within 6G—enabled SAGIN networks. IEEE Internet Things J 2025; 12(7): 35890-907.

[25]	Sun Y, Di B, Deng R, Song L . On an ultra—dense LEO—satellite—based computing network constellation. Engineering 2025; 64: 103-14.

[26]	Huang J, Cao M, Yang C, Han Z, Li T . Learning—based matching game for task scheduling and resource collaboration in intent—driven task—oriented networks. Engineering 2025; 64: 143-54.

[27]	Shi Q, Zhang J, Wang KM, Li Y, Park S, Kim T, et al. Traversal—flyby planning of Walker—Delta mega constellation using Hohmann maneuvers and phase sequencing. Aerosp Sci Technol 2026; 168: 110982.

[28]	rd Generation Partnership Project. Study on new radio (NR) to support non—terrestrial networks (release 16). Report. Valbonne: 3GPP; 2023 Mar.

[29]	International Telecommunication Union . Attenuation by atmospheric gases and related effects. Report. Geneva: ITU; 2022 Aug.

[30]	International Telecommunication Union . Propagation data and prediction methods required for the design of Earth—space telecommunication systems. Report. Geneva: ITU; 2023 Aug.

[31]	Guo X, Liu X, Meng Y, Cheng W, Wang W, Zhu L . Energy—efficient path planning scheme of multiple UAVs for reliable data collection. IEEE Internet Things J 2025; 12(23): 50882-98.

[32]	Geng W, Yu F, Jiao Y, Yin R, Zhao Z, Liu H . Orbital—attitude coupled sensing model for power optimization of single—axis solar arrays in Earth—orbiting satellites. IEEE Sens J 2025; 25(15): 28493—502.

[33]	Wang S, Xia W, Zhao H, Wei K, Quek TQS, Zhu H . Stackelberg game—based hierarchical incentive mechanism for clustered vehicular federated learning. IEEE Trans Commun 2025; 73(10): 9071-86.

[34]	Yang Q, Chu SC, Hu CC, Kong L, Pan JS . A task offloading method based on user satisfaction in C—RAN with mobile edge computing. IEEE Trans Mob Comput 2023; 23(4): 3452-65.

[35]	Yu H, Li P, Huang W, Du R, Xu Q, Nie L, et al. Social—aware incentive mechanism for data quality in mobile crowdsensing: a three—stage Stackelberg game approach. IEEE Internet Things J 2025; 12(7): 7980-94.

[36]	Koushki J, Shahbeyk S . Characterization of generalized FJ and KKT conditions for robust optimization. J Optim Theory Appl 2025; 206(2): 28.

[37]	Darzanos G, Koutsopoulos I, Stamoulis GD . Economics models and policies for cloud federations. In: Proceedings of the IFIP Networking Conference and Workshops; 2016 May 17—19; Vienna, Austria. New York City: IEEE; 2016. p. 485-93.

[38]	Sargent TJ . Beyond demand and supply curves in macroeconomics. Am Econ Rev 1982; 72(2): 382—9.

[39]

Däubener S, Damm S, Fischer A . ELBO, regularized maximum likelihood, and their common one—sample approximation for training stochastic neural networks. In: Proceedings of the 41st Conference on Uncertainty in Artificial Intelligence; 2025 Jul 21—25; Rio de Janeiro, Brazil. Rome: Association for Uncertainty in Artificial Intelligence (AUAI); 2025. p. 1-18.

[40]	Hou X, Wang J, Du J, Jiang C, Ren Y . Distributed machine learning for autonomous agent swarm: a survey. IEEE Commun Surv Tutor 2026; 28: 1597-636.

[41]	Rao J, Wang J, Xu J, Zhao S . Optimal control of nonlinear system based on deterministic policy gradient with eligibility traces. Nonlinear Dyn 2023; 111(21): 20041-53.

[42]	Federal Communications Commission. Space exploration holdings, LLC, request for modification of the authorization for the SpaceX NGSO satellite system: order and authorization and order on reconsideration (FCC 21—48). Report. Washington, DC: Federal Communications Commission ; 2021.

[43]	[Matrice 300 RTK: specifications] [Internet]. Shenzhen: DJI Technology Co., Ltd; c2026 [cited 2026 Aug 21]. Available from: https://www.dji.com/support/product/matrice—300.

[44]	Qin P, Li H, Fu Y, Hu J, Wu X, Zhang X . Learning—based NOMA—enabled queue—aware task offloading and UAV 3D trajectory planning for SAGIN. IEEE Trans Veh Technol 2025; 74(8): 12364—75.

[45]	Sun G, Wang Y, Yu H, Guizani M . Proportional fairness—aware task scheduling in space—air—ground integrated networks. IEEE Trans Serv Comput 2024; 17(6): 4125—37.

[46]	Lin Y, Xiao L, Tao Y, Zhang Y, Shu F, Li J . Multi—agent computing—energy—efficiency optimization in vehicular edge computing: non—cooperative versus cooperative solutions. IEEE Trans Wirel Commun 2025; 24(7): 5461-76.