pith. sign in

arxiv: 2509.17676 · v2 · submitted 2025-09-22 · 💻 cs.NI

GLo-MAPPO: Multi-Agent Deep Reinforcement Learning for Energy-Efficient UAV-Assisted LoRa Networks

Abdullahi Isa Ahmed , Jamal Bentahar , El Mehdi Amhoud This is my paper

Pith reviewed 2026-05-18 14:53 UTC · model grok-4.3

classification 💻 cs.NI
keywords LoRa networksUAV gatewaysMulti-agent reinforcement learningEnergy efficiencyIoT coverageSpreading factor optimizationTrajectory planningPartially observable stochastic game
0
0 comments X

The pith

GLo-MAPPO uses multi-agent reinforcement learning to let UAVs serve as mobile gateways that jointly optimize spreading factors, powers, trajectories, and associations for higher energy efficiency in LoRa networks.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper proposes a multi-UAV setup where drones act as moving gateways to gather data from ground LoRa end devices, replacing fixed towers that leave coverage gaps or satellites that waste energy. It frames the combined choice of spreading factors, transmit powers, flight paths, and device-to-drone links as a partially observable stochastic game and solves it with a new algorithm called GLo-MAPPO that trains centrally but runs decisions locally. Simulations across different densities show the method delivers better weighted energy efficiency and lower total power draw than existing multi-agent reinforcement learning approaches. The work also includes ablation checks confirming that the association scheme and each optimized variable contribute to the gains. A reader would care because longer battery life for remote IoT sensors could reduce maintenance trips and expand coverage without new ground infrastructure.

Core claim

The authors formulate the joint optimization of spreading factors, transmission powers, UAV trajectories, and ED-UAV associations as a partially observable stochastic game and solve it with GLo-MAPPO, a multi-agent proximal policy optimization method that uses centralized training with decentralized execution together with a gain-based association scheme; simulation results show this yields significantly higher energy efficiency and lower power consumption than prior multi-agent reinforcement learning benchmarks across varying network densities.

What carries the argument

GLo-MAPPO, the multi-agent proximal policy optimization algorithm that solves the partially observable stochastic game formulation of the joint optimization problem via centralized training with decentralized execution and a gain-based ED-UAV association scheme.

If this is right

  • Higher weighted energy efficiency is achieved by jointly tuning spreading factors, powers, trajectories, and associations.
  • Power consumption drops compared with prior multi-agent reinforcement learning methods at multiple network densities.
  • Each optimization component and the gain-based association scheme are necessary, as shown by ablation studies.
  • The centralized-training decentralized-execution structure allows scalable decisions while capturing global system goals.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same joint-optimization approach could be adapted to other low-power wide-area technologies that need mobile gateways.
  • Real hardware trials would reveal how much the simulated gains shrink when wind, interference, or battery aging appear.
  • If the energy savings hold outdoors, operators might reduce reliance on satellite backhaul for sparse rural IoT deployments.
  • The framework suggests a route to dynamic gateway placement that could later incorporate predictive traffic or weather data.

Load-bearing premise

The simulation model of LoRa propagation, UAV mobility, energy consumption, and channel dynamics matches real-world conditions without large unmodeled discrepancies.

What would settle it

Deploy the GLo-MAPPO controller on physical UAVs and LoRa end devices in an outdoor testbed and compare measured energy efficiency and power draw against the simulation predictions for the same densities and trajectories.

Figures

Figures reproduced from arXiv: 2509.17676 by Abdullahi Isa Ahmed, El Mehdi Amhoud, Jamal Bentahar.

Figure 1
Figure 1. Figure 1: Illustration of the studied system model. UAVs equipped with [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: LoRa-assisted UAV navigation through cone-constrained [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: GLo-MAPPO Architecture. where the clipped value prediction is defined as: V clip µ (ob,u) =    Vµold (ob,u) − ϵ, if Vµ(ob,u) < Vµold (ob,u) − ϵ Vµold (ob,u) + ϵ, if Vµ(ob,u) > Vµold (ob,u) + ϵ Vµ(ob,u), otherwise. Here, Rˆ b denotes the estimated reward-to-go for agent u and b refers to batch size. This training structure enables decentralized policy learning by agents while benefiting from the stab… view at source ↗
Figure 4
Figure 4. Figure 4: Training performance under different hyperparameter settings: (a) Effect of learning rate [PITH_FULL_IMAGE:figures/full_fig_p012_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: (a) Effect of the discount factor γ on the training performance and long-term reward accumulation. (b) Scalability evaluation with varying numbers of EDs, demonstrating performance consistency in larger networks. (c) Scalability with respect to the number of GWs, highlighting the system’s adaptability to increased infrastructure. B. Training Performance [PITH_FULL_IMAGE:figures/full_fig_p012_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Performance analysis of a 2-UAV system: (a) 2D projection of UAV trajectories and End Device distribution, (b) 3D view of UAV [PITH_FULL_IMAGE:figures/full_fig_p013_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Performance analysis of a 4-UAV system: (a) 2D projection of UAV trajectories and End Device distribution, (b) 3D view of UAV [PITH_FULL_IMAGE:figures/full_fig_p013_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Performance comparison of the proposed approach against various benchmarking methods: (a) Training reward curves and convergence [PITH_FULL_IMAGE:figures/full_fig_p014_8.png] view at source ↗
read the original abstract

The rapid advancement of Low-Power Wide Area Networks (LPWANs), particularly Long Range (LoRa) systems, has positioned them as a cornerstone for Next-Generation Internet of Things (NG-IoT) applications within 5G/6G ecosystems. Despite their long-range and low-power advantages, achieving high energy efficiency in LoRa networks remains a significant challenge in highly dynamic environments. Traditional terrestrial gateway deployments often suffer from coverage gaps and non-line-of-sight propagation, while satellite-based alternatives incur excessive energy consumption and prohibitive latency. To address these limitations, we propose a multi-UAV architecture where unmanned aerial vehicles (UAVs) serve as mobile LoRa gateways to dynamically collect data from ground-based end devices (EDs). We formulate a joint optimization problem to maximize the system's weighted energy efficiency by jointly optimizing spreading factors, transmission powers, UAV trajectories, and ED-UAV associations. This problem is transformed into a partially observable stochastic game (POSG), which we solve using our proposed Green LoRa Multi-Agent Proximal Policy Optimization (GLo-MAPPO). Our framework leverages centralized training with decentralized execution (CTDE) and is enhanced by a gain-based ED-UAV association scheme. Simulation results show that GLo-MAPPO significantly outperforms state-of-the-art multi-agent reinforcement learning (MARL) benchmarks in energy efficiency and power consumption across varying network densities. Furthermore, ablation studies validate the necessity of each optimization component and the effectiveness of the proposed association scheme.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper proposes GLo-MAPPO, a variant of multi-agent proximal policy optimization, to solve a joint optimization problem in a multi-UAV LoRa gateway architecture. The goal is to maximize weighted energy efficiency by optimizing spreading factors, transmission powers, UAV trajectories, and ED-UAV associations, formulated as a partially observable stochastic game solved under centralized training with decentralized execution. Simulation results claim significant outperformance over state-of-the-art MARL baselines in energy efficiency and power consumption across varying network densities, with ablation studies supporting the contribution of each component and the proposed association scheme.

Significance. If the reported gains prove robust, the work could advance practical UAV-assisted LPWAN deployments for dynamic IoT scenarios by demonstrating how CTDE-based MARL can jointly handle discrete (SF, power) and continuous (trajectory) decisions. The gain-based association heuristic and explicit energy model are concrete contributions that could be reused. However, the significance is limited by the absence of external validation or real traces, so the claimed improvements remain tied to the fidelity of the chosen propagation, mobility, and energy models.

major comments (2)
  1. [§5] §5 (Simulation Results) and the energy-efficiency definition in §3.2: the central performance claim rests on comparisons whose statistical significance, number of independent runs, and baseline hyperparameter tuning protocol are not reported. Without these, it is impossible to determine whether the reported margins exceed what could arise from simulation variance or post-hoc tuning.
  2. [§4] §4 (System Model) and §5.1 (LoRa Propagation and Energy Model): the path-loss, SF-threshold, and energy-consumption equations omit log-normal shadowing, co-channel interference from external sources, wind-induced UAV control power, and altitude-dependent coverage. Because the energy-efficiency metric is defined directly from these equations, any systematic bias in the model directly inflates the reported gains relative to real deployments.
minor comments (2)
  1. [§3.3] The notation for the POSG tuple and the reward function in §3.3 could be clarified with an explicit table mapping each component to its mathematical symbol.
  2. [Figure 3] Figure 3 (trajectory plots) would benefit from an overlay of the ground-truth ED locations and a legend indicating which UAV is which across time steps.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments on our manuscript. We address each major comment point by point below, indicating where revisions have been made to strengthen the presentation and analysis.

read point-by-point responses

  1. Referee: [§5] §5 (Simulation Results) and the energy-efficiency definition in §3.2: the central performance claim rests on comparisons whose statistical significance, number of independent runs, and baseline hyperparameter tuning protocol are not reported. Without these, it is impossible to determine whether the reported margins exceed what could arise from simulation variance or post-hoc tuning.

    Authors: We agree that these experimental details should have been reported explicitly. Our simulations were performed over 5 independent runs per scenario using distinct random seeds, with results presented as averages; we have now added standard deviations to all figures and tables in the revised Section 5. Baseline hyperparameters were taken directly from the respective original papers to maintain fairness, while our own parameters were selected via a modest grid search on a held-out validation scenario. A new paragraph has been inserted in Section 5.1 to document the run count, statistical reporting, and tuning protocol. revision: yes

  2. Referee: [§4] §4 (System Model) and §5.1 (LoRa Propagation and Energy Model): the path-loss, SF-threshold, and energy-consumption equations omit log-normal shadowing, co-channel interference from external sources, wind-induced UAV control power, and altitude-dependent coverage. Because the energy-efficiency metric is defined directly from these equations, any systematic bias in the model directly inflates the reported gains relative to real deployments.

    Authors: We acknowledge the simplifications in the models of Section 4. Log-normal shadowing and external interference were deliberately excluded to focus on the joint optimization of the proposed MARL framework rather than on channel impairments; we have added an explicit limitations paragraph in the revised Section 4.2 noting that the reported gains represent an upper bound under idealized propagation. Altitude dependence has been incorporated into the path-loss formula. Wind-induced control power is omitted because the UAVs are modeled as hovering at constant altitude with negligible additional consumption in the simulated scenarios; this assumption is now stated and its implications discussed in the updated energy model section. revision: partial

Circularity Check

0 steps flagged

No significant circularity; derivation and evaluation remain independent of inputs

full rationale

The paper formulates a standard joint optimization over spreading factors, powers, trajectories and associations as a POSG, then applies a CTDE-enhanced MAPPO variant (GLo-MAPPO) with an added gain-based association heuristic. Simulation results compare the learned policy against other MARL baselines on the same objective; this is an empirical performance claim, not a reduction by construction. No equation equates the reported energy-efficiency gains to a fitted parameter or self-citation chain, and the simulation models (path-loss, energy equations) are presented as modeling assumptions rather than derived outputs. The central result therefore retains independent content from the chosen algorithm and benchmark comparisons.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract supplies no explicit free parameters, axioms, or invented entities; typical RL papers of this type implicitly rely on many hyperparameters and simulation assumptions not listed here.

pith-pipeline@v0.9.0 · 5810 in / 1009 out tokens · 39896 ms · 2026-05-18T14:53:59.874881+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

48 extracted references · 48 canonical work pages · 2 internal anchors

  1. [1]

    A survey on scalable lorawan for massive iot: Recent advances, potentials, and challenges,

    M. Jouhari, N. Saeed, M.-S. Alouini, and E. M. Amhoud, “A survey on scalable lorawan for massive iot: Recent advances, potentials, and challenges,”IEEE Communications Surveys & Tutorials, 2023

    work page 2023

  2. [2]

    6g internet of things: A comprehensive survey,

    D. C. Nguyen, M. Ding, P. N. Pathirana, A. Seneviratne, J. Li, D. Niyato, O. Dobre, and H. V . Poor, “6g internet of things: A comprehensive survey,”IEEE Internet of Things Journal, vol. 9, no. 1, pp. 359–383, 2022. This work has been submitted to the IEEE for possible publication. Copyright may be transferred without notice, after which this version may ...

    work page 2022

  3. [3]

    Optimizing energy-efficient cooperative mac strategies for data collection in iot networks with terrestrial and nonterrestrial relays,

    Z. Zhang, S. Atapattu, B. Ren, Y . Wang, and M. Di Renzo, “Optimizing energy-efficient cooperative mac strategies for data collection in iot networks with terrestrial and nonterrestrial relays,”IEEE Internet of Things Journal, vol. 12, no. 17, pp. 35 556–35 576, 2025

    work page 2025

  4. [4]

    A unified deep transfer learning model for accurate iot localization in diverse environments,

    A. I. Ahmed, Y . Etiabi, A. W. Azim, and E. M. Amhoud, “A unified deep transfer learning model for accurate iot localization in diverse environments,” inIEEE 35th International Symposium on Personal, Indoor and Mobile Radio Communications, 2024

    work page 2024

  5. [5]

    Enhancing smart home device identification in wifi environments for futuristic smart networks-based iot,

    H. F. Fakhruldeen, M. J. Saadh, S. Khan, N. A. Salim, N. Jhamat, and G. Mustafa, “Enhancing smart home device identification in wifi environments for futuristic smart networks-based iot,”International Journal of Data Science and Analytics, 2024

    work page 2024

  6. [6]

    A first step toward an iot network dedicated to the sustainable development of a territory,

    G. Orazi, G. Fontaine, P. Chemla, M. Zhao, P. Cousin, and F. Le Gall, “A first step toward an iot network dedicated to the sustainable development of a territory,” inGlobal Internet of Things Summit, 2018

    work page 2018

  7. [7]

    Spreading factor assisted lora localization with deep reinforcement learning,

    Y . Etiabi, M. Jouhari, A. Burg, and E. M. Amhoud, “Spreading factor assisted lora localization with deep reinforcement learning,” inIEEE 97th Vehicular Technology Conference, 2023

    work page 2023

  8. [8]

    Energy efficient resource allocation for uplink lora networks,

    B. Su, Z. Qin, and Q. Ni, “Energy efficient resource allocation for uplink lora networks,” inIEEE Global Communications Conference, 2018

    work page 2018

  9. [9]

    Internet of things (iot) using lora technology,

    A. Zourmand, A. L. K. Hing, C. W. Hung, and M. AbdulRehman, “Internet of things (iot) using lora technology,” inIEEE international conference on automatic control and intelligent systems, 2019

    work page 2019

  10. [10]

    Energy-efficient computing offloading with trajectory optimization and resource allo- cation in uavs aided industrial iot,

    Z. Yu, Z. Zhang, S. Zeadally, B. Shen, and X. Pei, “Energy-efficient computing offloading with trajectory optimization and resource allo- cation in uavs aided industrial iot,”IEEE Internet of Things Journal, vol. 12, no. 17, pp. 34 780–34 792, 2025

    work page 2025

  11. [11]

    Fast deployment of uav networks for optimal wireless coverage,

    X. Zhang and L. Duan, “Fast deployment of uav networks for optimal wireless coverage,”IEEE Transactions on Mobile Computing, vol. 18, no. 3, pp. 588–601, 2018

    work page 2018

  12. [12]

    College admissions and the stability of marriage,

    D. Gale and L. S. Shapley, “College admissions and the stability of marriage,”The American mathematical monthly, vol. 69, no. 1, pp. 9– 15, 1962

    work page 1962

  13. [13]

    Energy efficiency in short and wide-area iot technologies—a survey,

    E. Zanaj, G. Caso, L. De Nardis, A. Mohammadpour, ¨O. Alay, and M.-G. Di Benedetto, “Energy efficiency in short and wide-area iot technologies—a survey,”Technologies, vol. 9, no. 1, 2021

    work page 2021

  14. [14]

    A study of lora: Long range & low power networks for the internet of things,

    A. Augustin, J. Yi, T. Clausen, and W. M. Townsley, “A study of lora: Long range & low power networks for the internet of things,”Sensors, vol. 16, no. 9, 2016

    work page 2016

  15. [15]

    The narrowband internet of things (nb-iot) resources management performance state of art, challenges, and opportunities,

    E. M. Migabo, K. D. Djouani, and A. M. Kurien, “The narrowband internet of things (nb-iot) resources management performance state of art, challenges, and opportunities,”IEEE Access, vol. 8, 2020

    work page 2020

  16. [16]

    The surprising effectiveness of PPO in cooperative multi-agent games,

    C. Yu, A. Velu, E. Vinitsky, J. Gao, Y . Wang, A. Bayen, and Y . Wu, “The surprising effectiveness of PPO in cooperative multi-agent games,” inThirty-sixth Conference on Neural Information Processing Systems Datasets and Benchmarks Track, 2022

    work page 2022

  17. [17]

    Spreading fac- tor allocation strategy for lora networks under imperfect orthogonality,

    L. Amichi, M. Kaneko, N. El Rachkidy, and A. Guitton, “Spreading fac- tor allocation strategy for lora networks under imperfect orthogonality,” inIEEE International Conference on Communications, 2019

    work page 2019

  18. [18]

    Energy efficient uplink transmissions in lora networks,

    B. Su, Z. Qin, and Q. Ni, “Energy efficient uplink transmissions in lora networks,”IEEE Transactions on Communications, vol. 68, no. 8, 2020

    work page 2020

  19. [19]

    Dynamic spreading factor assignment in lora wireless networks,

    R. Hamdi, M. Qaraqe, and S. Althunibat, “Dynamic spreading factor assignment in lora wireless networks,” inIEEE international conference on communications, 2020

    work page 2020

  20. [20]

    Flyinglora: Towards energy efficient data collection in uav-assisted lora networks,

    R. Xiong, C. Liang, H. Zhang, X. Xu, and J. Luo, “Flyinglora: Towards energy efficient data collection in uav-assisted lora networks,”Computer Networks, vol. 220, p. 109511, 2023

    work page 2023

  21. [21]

    Efficient heuristic for optimal milp-lora adaptive resource allocation for aquaculture

    M. I. Arasu, S. S. Rani, and G. R. Geoffery, “Efficient heuristic for optimal milp-lora adaptive resource allocation for aquaculture.” Intelligent Automation & Soft Computing, vol. 33, no. 2, 2022

    work page 2022

  22. [22]

    Enhanced lorawan performance through advanced spread factor allocation empowered by machine learning,

    M. R. Rao and S. Sundar, “Enhanced lorawan performance through advanced spread factor allocation empowered by machine learning,” Engineering Research Express, vol. 6, no. 4, p. 045354, 2024

    work page 2024

  23. [23]

    Intelligent resource allocation in lorawan using machine learning techniques,

    S. U. Minhaj, A. Mahmood, S. F. Abedin, S. A. Hassan, M. T. Bhatti, S. H. Ali, and M. Gidlund, “Intelligent resource allocation in lorawan using machine learning techniques,”IEEE Access, vol. 11, 2023

    work page 2023

  24. [24]

    Mix-mab: Reinforcement learning-based resource allocation algorithm for lorawan,

    F. Azizi, B. Teymuri, R. Aslani, M. Rasti, J. Tolvaneny, and P. H. J. Nardelli, “Mix-mab: Reinforcement learning-based resource allocation algorithm for lorawan,” inIEEE 95th Vehicular Technology Conference, 2022

    work page 2022

  25. [25]

    Efficient online resource allocation in large- scale lorawan networks: A multi-agent approach,

    C. Garrido-Hidalgo, L. Roda-Sanchez, F. J. Ram ´ırez, A. Fern ´andez- Caballero, and T. Olivares, “Efficient online resource allocation in large- scale lorawan networks: A multi-agent approach,”Computer Networks, vol. 221, 2023

    work page 2023

  26. [26]

    M. R. Rao and S. Sundar, “Enhancement in optimal resource-based data transmission over lpwan using a deep adaptive reinforcement learning model aided by novel remora with lotus effect optimization algorithm,” IEEE Access, 2024

    work page 2024

  27. [27]

    Q-learning aided resource allocation and environment recognition in lorawan with csma/ca,

    N. Aihara, K. Adachi, O. Takyu, M. Ohta, and T. Fujii, “Q-learning aided resource allocation and environment recognition in lorawan with csma/ca,”IEEE Access, vol. 7, 2019

    work page 2019

  28. [28]

    A lightweight, fully- distributed ai framework for energy-efficient resource allocation in lora networks,

    A. Scarvaglieri, S. Palazzo, and F. Busacca, “A lightweight, fully- distributed ai framework for energy-efficient resource allocation in lora networks,” inProceedings of the IEEE/ACM 16th International Conference on Utility and Cloud Computing, 2023

    work page 2023

  29. [29]

    Multi-agent q-learning algorithm for dynamic power and rate allocation in lora networks,

    Y . Yu, L. Mroueh, S. Li, and M. Terr´e, “Multi-agent q-learning algorithm for dynamic power and rate allocation in lora networks,” inIEEE 31st Annual International Symposium on Personal, Indoor and Mobile Radio Communications, 2020

    work page 2020

  30. [30]

    Deep reinforcement learning-based energy efficiency optimization for flying lora gateways,

    M. Jouhari, K. Ibrahimi, J. B. Othman, and E. M. Amhoud, “Deep reinforcement learning-based energy efficiency optimization for flying lora gateways,” inIEEE International Conference on Communications, 2023

    work page 2023

  31. [31]

    Alternating optimization based hybrid precoding strategies for millimeter wave mimo systems,

    X. Qiao, Y . Zhang, M. Zhou, and L. Yang, “Alternating optimization based hybrid precoding strategies for millimeter wave mimo systems,” IEEE Access, vol. 8, pp. 113 078–113 089, 2020

    work page 2020

  32. [32]

    Convergence of alternating optimiza- tion,

    J. C. Bezdek and R. J. Hathaway, “Convergence of alternating optimiza- tion,”Neural, Parallel & Scientific Computations, vol. 11, no. 4, 2003

    work page 2003

  33. [33]

    Ris-assisted integrated sensing and communication system with physical layer security enhance- ment by drl approach,

    P. Jiang, X. Cao, Y . He, X. Song, and Z. Lyu, “Ris-assisted integrated sensing and communication system with physical layer security enhance- ment by drl approach,” inIEEE 99th Vehicular Technology Conference, 2024

    work page 2024

  34. [34]

    The surprising effectiveness of ppo in cooperative multi-agent games,

    C. Yu, A. Velu, E. Vinitsky, J. Gao, Y . Wang, A. Bayen, and Y . Wu, “The surprising effectiveness of ppo in cooperative multi-agent games,” Advances in Neural Information Processing Systems, vol. 35, 2022

    work page 2022

  35. [35]

    Bench- marking multi-agent deep reinforcement learning algorithms in coopera- tive tasks,

    G. Papoudakis, F. Christianos, L. Sch ¨afer, and S. V . Albrecht, “Bench- marking multi-agent deep reinforcement learning algorithms in coopera- tive tasks,” inProceedings of the Neural Information Processing Systems Track on Datasets and Benchmarks, 2021

    work page 2021

  36. [36]

    Multi-agent reinforcement learning- based resource allocation for uav networks,

    J. Cui, Y . Liu, and A. Nallanathan, “Multi-agent reinforcement learning- based resource allocation for uav networks,”IEEE Transactions on Wireless Communications, vol. 19, no. 2, pp. 729–743, 2019

    work page 2019

  37. [37]

    Multi-uav cooperative searching and tracking for moving targets based on multi-agent reinforcement learning,

    K. Su and F. Qian, “Multi-uav cooperative searching and tracking for moving targets based on multi-agent reinforcement learning,”Applied Sciences, vol. 13, no. 21, p. 11905, 2023

    work page 2023

  38. [38]

    Joint optimization of mobility and reliability-guaranteed air-to-ground communication for uavs,

    J. Zhou, D. Tian, Y . Yan, X. Duan, and X. Shen, “Joint optimization of mobility and reliability-guaranteed air-to-ground communication for uavs,”IEEE Transactions on Mobile Computing, vol. 23, no. 1, pp. 566–580, 2022

    work page 2022

  39. [39]

    Joint gateway selection and resource allocation for cross-tier communication in space-air-ground integrated iot networks,

    Y . Shi, Y . Xia, and Y . Gao, “Joint gateway selection and resource allocation for cross-tier communication in space-air-ground integrated iot networks,”IEEE Access, vol. 9, pp. 4303–4314, 2021

    work page 2021

  40. [40]

    Why the shannon and hartley entropies are ‘natural’,

    J. Acz ´el, B. Forte, and C. T. Ng, “Why the shannon and hartley entropies are ‘natural’,”Advances in applied probability, vol. 6, no. 1, pp. 131– 146, 1974

    work page 1974

  41. [41]

    Modeling power consumptions for multirotor uavs,

    H. Gong, B. Huang, B. Jia, and H. Dai, “Modeling power consumptions for multirotor uavs,”IEEE Transactions on Aerospace and Electronic Systems, vol. 59, no. 6, 2023

    work page 2023

  42. [42]

    Energy minimization for wireless communication with rotary-wing uav,

    Y . Zeng, J. Xu, and R. Zhang, “Energy minimization for wireless communication with rotary-wing uav,”IEEE Transactions on Wireless communications, 2019

    work page 2019

  43. [43]

    M. R. Garey and D. S. Johnson,Computers and intractability. wh freeman New York, 2002

    work page 2002

  44. [44]

    High-Dimensional Continuous Control Using Generalized Advantage Estimation

    J. Schulman, P. Moritz, S. Levine, M. Jordan, and P. Abbeel, “High- dimensional continuous control using generalized advantage estimation,” arXiv preprint arXiv:1506.02438, 2015

    work page internal anchor Pith review Pith/arXiv arXiv 2015

  45. [45]

    Counterfactual multi-agent policy gradients,

    J. Foerster, G. Farquhar, T. Afouras, N. Nardelli, and S. Whiteson, “Counterfactual multi-agent policy gradients,” inProceedings of the AAAI conference on artificial intelligence, vol. 32, no. 1, 2018

    work page 2018

  46. [46]

    Value-Decomposition Networks For Cooperative Multi-Agent Learning

    P. Sunehag, G. Lever, A. Gruslys, W. M. Czarnecki, V . Zambaldi, M. Jaderberg, M. Lanctot, N. Sonnerat, J. Z. Leibo, K. Tuylset al., “Value-decomposition networks for cooperative multi-agent learning,” arXiv preprint arXiv:1706.05296, 2017

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  47. [47]

    Monotonic value function factorisation for deep multi- agent reinforcement learning,

    T. Rashid, M. Samvelyan, C. S. De Witt, G. Farquhar, J. Foerster, and S. Whiteson, “Monotonic value function factorisation for deep multi- agent reinforcement learning,”Journal of Machine Learning Research, vol. 21, no. 178, pp. 1–51, 2020

    work page 2020

  48. [48]

    Is indepen- dent learning all you need in the starcraft multi-agent challenge?

    C. S. De Witt, T. Gupta, D. Makoviichuk, V . Makoviychuk, P. H. Torr, M. Sun, and S. Whiteson, “Is independent learning all you need in the starcraft multi-agent challenge?”arXiv preprint arXiv:2011.09533, 2020

    work page arXiv 2011