Pith. sign in

REVIEW 53 references

Embedded Mean Field Reinforcement Learning for Perimeter-defense Game

T0 review · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read The paper derives optimal breach and interception strategies for a 3D perimeter-defense game and introduces an embedded mean-field actor-critic method for large-scale defender coordination.

arxiv 2505.14209 v1 pith:AXCMF6U7 submitted 2025-05-20 cs.AI

Li Wang , Xin Yu , Xuxin Lv , Gangzheng Ai , Wenjun Wu This is my paper
classification cs.AI
keywords emfacdefendersgamelarge-scalemean-fieldperimeter-defenseacrossattackers
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

In this game, an attacker outside a dome tries to reach the dome's surface, while a defender inside tries to stop it. The paper first analyzes a one-on-one version. It claims that both players should fly in a straight line at top speed to a particular point on the dome, found by solving two coupled equations. This point balances the attacker's time to arrival against the defender's time to arrival. If the paper's theorem is right, neither player can improve by choosing a different point or path. The paper also maps every possible starting position of the attacker onto a surface that separates attacker wins from defender wins.

The larger contribution is an algorithm for many defenders with different flight models, in wind. Because enemies and allies are numerous, a defender cannot consider everyone. The method first learns a compact high-level action for each defender, so that actions from different flight models can be averaged in a meaningful way. Then a lightweight attention mechanism picks out the few nearby agents that matter most, based on predicting the reward. These two pieces are fed into a mean-field actor-critic reinforcement learner. Experiments compare against several baselines at scales of 10 to 50 defenders and report faster learning and higher success rates; a 2v2 drone test shows high success with low collisions.

The main weaknesses are in the theory: the Nash proof compares only arrival times at the chosen boundary point and does not analyze interception along the way, and the zero-payoff surface is asserted without derivation. The simulation setup also appears to swap attacker and defender speeds relative to the model.

Extended reading notes

Core claim

The paper claims the following pair of strategies forms a Nash equilibrium: the attacker selects the optimal breach point B* = argmax_B P(B) and follows the direct linear trajectory AB* at maximum speed, while the defender intercepts along the direct linear trajectory DB* at maximum speed. If true, neither player can improve by unilateral deviation. The paper also claims that the EMFAC framework outperforms IDDPG, ITD3, MADDPG, MATD3, HADDPG, HATD3, MTMFAC, and a rule-based baseline across 10v10 to 50v50 tasks in both convergence speed and final reward.

Load-bearing premise

The Nash equilibrium proof assumes the only thing that matters is which player first reaches the boundary point B*, i.e. the sign of P = tau_D - tau_A, and it only analyzes defender deviations to points C on the segment AB*. The game definition in Section II-A also permits interception anywhere, with termination condition ||ZA(t)-ZD(t)|| < epsilon at any time t. If a defender could do better by intercepting the attacker before the boundary, or if the attacker could exploit this by changing target once the defender deviates, the claimed equilibrium need not hold. The small-angle derivation of the fixed-point system in Theorem 2 is also treated as exact for finite geometries.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Assumptions & free parameters 5 free parameters · 6 assumptions · 2 invented entities

The central theoretical claim rests on the race-to-boundary payoff model, exactness of the small-angle fixed-point system, and a one-to-one assignment assumption; the algorithm rests on hand-selected reward weights and hyperparameters that are not derived. No physical entities are invented, but the high-level action embedding and attention weights are learned constructs with no independent evidence.

free parameters (5)
  • Reward weights alpha1, alpha2, alpha3, alpha4 = -0.01, 10, 10, -0.03
    Hand-selected in Section III-D; they determine the balance between task completion, guidance, and collision penalties in all reported results.
  • Attention ratio k = 0.3
    Hand-set in Table IV; sensitivity analysis in Fig. 10b shows performance depends on k at extreme values.
  • High-level action dimension = 4
    Hand-set in Table IV; Fig. 10a shows robustness, but it is a capacity choice not derived from the problem.
  • Interception and safety thresholds d_th, d_safe, capture radius epsilon = not reported
    Define success and collision in Section III-D and Section II-A, but values are absent, so the exact task measured is not reproducible.
  • Wind noise variance sigma^2 = not reported
    Equation (29) includes Gaussian noise with variance sigma^2, but sigma^2 is not given in the paper.
assumptions (6)
  • domain assumption First-order kinematics with constant speeds: defender speed 1, attacker speed v <= 1; maximum-speed straight-line motion is the relevant class of strategies.
    Stated in Section II-A and used throughout Theorems 1-3; violated if agents use higher-order dynamics, though Section III later introduces heterogeneous dynamics.
  • ad hoc to paper Outcome is fully determined by the sign of P = tau_D - tau_A evaluated at the chosen boundary point B*, with no need to model mid-course interception.
    The game definition in Section II-A includes termination when ||ZA(t)-ZD(t)|| < epsilon at any t, but Theorem 1's proof only compares arrival times at B* and deviations C in segment AB*.
  • ad hoc to paper The infinitesimal small-angle relation d_tau_A = (R cos(beta)/v) d_theta yields the exact optimum for finite geometries in Theorem 2.
    Theorem 2 proof step 2 uses a small-angle approximation to derive the fixed-point system that is then used as the exact equilibrium condition.
  • domain assumption Many-on-many game decouples into one-on-one duels through Hungarian assignment of attackers to defenders.
    Section III states the Hungarian algorithm is used for optimal target assignment but does not prove the decoupling is valid under wind and heterogeneous dynamics.
  • domain assumption Wind perturbation is additive: w = f(height, p, v) + N(0, sigma^2), with systematic part f and Gaussian noise, and f is treated as given.
    Equation (29) in Section III-B; f is never specified, so the environment is not fully defined.
  • ad hoc to paper The 16 hand-defined dynamic types cover the heterogeneity of real missiles and UAVs.
    Section III-B introduces 16 distinct dynamic types but does not list them or justify their coverage.
invented entities (2)
  • High-level action embedding space
    purpose: Maps heterogeneous raw actions into a shared representation so that mean-field action aggregation is meaningful across different dynamics.
    Trained with a state-prediction decoder (Eqs. 33-36); no external benchmark confirms the embedding is semantically correct.
  • Reward-based agent-level attention weights
    purpose: Selects which other agents to include in observations and mean-field actions, reducing effective interaction complexity.
    Trained by reward prediction (Eqs. 37-40); only internal visualizations are provided, with no external validation.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Embedded Mean Field Reinforcement Learning for Perimeter-defense Game." pith.science (2026). https://pith.science/paper/AXCMF6U7

@misc{pith2026250514209,
  author       = {Pith},
  title        = {Pith review of: Embedded Mean Field Reinforcement Learning for Perimeter-defense Game},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/AXCMF6U7}},
  note         = {Machine review of arXiv:2505.14209}
}
read the original abstract

With the rapid advancement of unmanned aerial vehicles (UAVs) and missile technologies, perimeter-defense game between attackers and defenders for the protection of critical regions have become increasingly complex and strategically significant across a wide range of domains. However, existing studies predominantly focus on small-scale, simplified two-dimensional scenarios, often overlooking realistic environmental perturbations, motion dynamics, and inherent heterogeneity--factors that pose substantial challenges to real-world applicability. To bridge this gap, we investigate large-scale heterogeneous perimeter-defense game in a three-dimensional setting, incorporating realistic elements such as motion dynamics and wind fields. We derive the Nash equilibrium strategies for both attackers and defenders, characterize the victory regions, and validate our theoretical findings through extensive simulations. To tackle large-scale heterogeneous control challenges in defense strategies, we propose an Embedded Mean-Field Actor-Critic (EMFAC) framework. EMFAC leverages representation learning to enable high-level action aggregation in a mean-field manner, supporting scalable coordination among defenders. Furthermore, we introduce a lightweight agent-level attention mechanism based on reward representation, which selectively filters observations and mean-field information to enhance decision-making efficiency and accelerate convergence in large-scale tasks. Extensive simulations across varying scales demonstrate the effectiveness and adaptability of EMFAC, which outperforms established baselines in both convergence speed and overall performance. To further validate practicality, we test EMFAC in small-scale real-world experiments and conduct detailed analyses, offering deeper insights into the framework's effectiveness in complex scenarios.

Figures

Figures reproduced from arXiv: 2505.14209 by the authors.

Figure 1
Figure 1. Hemispherical defense scenario with kinematic agents. [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. (a) The Transformed 2D Plane. (b) Proportional dis [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. EMFAC consists of two key components: the High [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (9 more)
Figure 3
Figure 3. Figure 3: Framework of the EMFAC Method. The left part shows the training of the high-level action encoder [PITH_FULL_IMAGE:figures/full_fig_p007_3.png]
Figure 4
Figure 4. Figure 4: Simulation results for Nash equilibrium validation. (a) The time required for both attackers and defenders to reach the [PITH_FULL_IMAGE:figures/full_fig_p009_4.png]
Figure 5
Figure 5. Figure 5: Comparison of average reward learning curves for different algorithms in perimeter-defense game tasks of varying [PITH_FULL_IMAGE:figures/full_fig_p009_5.png]
Figure 6
Figure 6. Figure 6: Comparison of Success and Collision Rates of Various Algorithms at Different Scales. [PITH_FULL_IMAGE:figures/full_fig_p010_6.png]
Figure 9
Figure 9. Figure 9: Comparison of 5v5 and 10v10 trajectory visualizations. [PITH_FULL_IMAGE:figures/full_fig_p010_9.png]
Figure 10
Figure 10. Figure 10: Parameter Sensitivity Analysis Metrics: Converged [PITH_FULL_IMAGE:figures/full_fig_p010_10.png]
Figure 8
Figure 8. Figure 8: Visualization of different matrices. From left to [PITH_FULL_IMAGE:figures/full_fig_p010_8.png]
Figure 11
Figure 11. Figure 11: Comparison of Algorithm Runtimes across Different [PITH_FULL_IMAGE:figures/full_fig_p011_11.png]
Figure 12
Figure 12. Figure 12: Experimental setup for real-world validation using [PITH_FULL_IMAGE:figures/full_fig_p012_12.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

53 extracted references · 44 canonical work pages

  1. [1]

    A review of multi-agent perimeter defense games,

    D. Shishika and V . Kumar, “A review of multi-agent perimeter defense games,” inProc. 11th Int. Conf. Decision and Game Theory for Security (GameSec), College Park, MD, USA, Oct. 28–30, 2020, pp. 472–485. Springer, 2020

    work page 2020

  2. [2]

    Local-game decomposition for multiplayer perimeter-defense problem,

    D. Shishika and V . Kumar, “Local-game decomposition for multiplayer perimeter-defense problem,” inProc. 2018 IEEE Conf. Decis. Control (CDC), 2018, pp. 2093–2100

    work page 2018

  3. [3]

    Modeling of target tracking system for homing missiles and air defense systems,

    Y . Alqudsi and G. El-Bayoumi, “Modeling of target tracking system for homing missiles and air defense systems,”INCAS Bulletin, vol. 10, no. 2, 2018

    work page 2018

  4. [4]

    Simulation of intelligent unmanned aerial vehicle (UA V) for military surveillance,

    M. A. Ma’Sum, M. K. Arrofi, G. Jati, F. Arifin, M. N. Kurniawan, P. Mursanto, and W. Jatmiko, “Simulation of intelligent unmanned aerial vehicle (UA V) for military surveillance,” inProc. 2013 Int. Conf. Adv. Comput. Sci. Inf. Syst. (ICACSIS), 2013, pp. 161–166

    work page 2013

  5. [5]

    A hierarchical deep reinforcement learning framework for 6-DOF UCA V air-to-air combat,

    J. Chai, W. Chen, Y . Zhu, Z.-X. Yao and D. Zhao, “A hierarchical deep reinforcement learning framework for 6-DOF UCA V air-to-air combat,” IEEE Transactions on Systems, Man, and Cybernetics: Systems, vol. 53, no. 9, pp. 5417–5429, 2023, publisher: IEEE

    work page 2023

  6. [6]

    K. S. Kappel, T. M. Cabreira, J. L. Marins, L. B. de Brisolara, and P. R. Ferreira,Strategies for patrolling missions with multiple UAVs. Journal of Intelligent & Robotic Systems, vol. 99, pp. 499–515, 2020

    work page 2020

  7. [7]

    Competitive perimeter defense on a line,

    S. Bajaj, E. Torng, and S. D. Bopardikar, “Competitive perimeter defense on a line,” inProc. 2021 American Control Conference (ACC), 2021, pp. 3196–3201

    work page 2021

  8. [8]

    Team composition for perimeter defense with patrollers and defenders,

    D. Shishika, J. Paulos, M. R. Dorothy, M. A. Hsieh, and V . Kumar, “Team composition for perimeter defense with patrollers and defenders,” inProc. 2019 IEEE 58th Conf. Decis. Control (CDC), 2019, pp. 7325–7332

    work page 2019

Show all 53 references
  1. [9]

    Cooperative team strategies for multi-player perimeter-defense games,

    D. Shishika, J. Paulos, and V . Kumar, “Cooperative team strategies for multi-player perimeter-defense games,”IEEE Robotics and Automation Letters, vol. 5, no. 2, pp. 2738–2745, 2020

    2020

  2. [10]

    Perimeter-defense game on arbitrary convex shapes,

    D. Shishika and V . Kumar, “Perimeter-defense game on arbitrary convex shapes,”arXiv preprint arXiv:1909.03989, 2019

    1909 arXiv

  3. [11]

    Multivehicle perimeter defense in conical environments,

    S. Bajaj, S. D. Bopardikar, E. Torng, A. V on Moll, and D. W. Casbeer, “Multivehicle perimeter defense in conical environments,”IEEE Trans- actions on Robotics, vol. 40, pp. 1439–1456, 2024

    2024

  4. [12]

    Perimeter-defense game between aerial defender and ground intruder,

    E. S. Lee, D. Shishika, and V . Kumar, “Perimeter-defense game between aerial defender and ground intruder,” in2020 59th IEEE Conference on Decision and Control (CDC), pp. 1530–1536, 2020. “‘

    2020

  5. [13]

    Defending a perimeter from a ground intruder using an aerial defender: Theory and practice,

    E. S. Lee, D. Shishika, G. Loianno, and V . Kumar, “Defending a perimeter from a ground intruder using an aerial defender: Theory and practice,” inProc. 2021 IEEE Int. Symp. Safety, Security, Rescue Robotics (SSRR), 2021, pp. 184–189

    2021

  6. [14]

    Learning decentralized strategies for a perimeter defense game with graph neural networks,

    E. S. Lee, L. Zhou, A. Ribeiro, and V . Kumar, “Learning decentralized strategies for a perimeter defense game with graph neural networks,” arXiv preprint arXiv:2211.01757, 2022

    2022 arXiv

  7. [15]

    Vision-based perimeter defense via multiview pose estimation,

    E. S. Lee et al., “Vision-based perimeter defense via multiview pose estimation,”arXiv preprint arXiv:2209.12136, 2022

    2022 arXiv

  8. [16]

    The role of heterogeneity in autonomous perimeter defense problems,

    A. Adler, O. Mickelin, R. K. Ramachandran, G. S. Sukhatme, and S. Karaman, “The role of heterogeneity in autonomous perimeter defense problems,”The International Journal of Robotics Research, vol. 43, no. 9, pp. 1363–1381, 2024

    2024

  9. [17]

    Intercept angle missile guidance under time vary- ing acceleration bounds,

    I. Taub and T. Shima, “Intercept angle missile guidance under time vary- ing acceleration bounds,”Journal of Guidance, Control, and Dynamics, vol. 36, no. 3, pp. 686–699, 2013

    2013

  10. [18]

    The effects of different wing configurations on missile aerodynamics,

    A. S ¸umnu and˙I. G¨uzelbey, “The effects of different wing configurations on missile aerodynamics,”Journal of Thermal Engineering, vol. 9, no. 5, pp. 1260–1271, 2023

    2023

  11. [19]

    Dynamic Modeling, Guidance, and Control of Missiles,

    M. A. Ma’Sum et al., “Dynamic Modeling, Guidance, and Control of Missiles,”Middle East Technical University (METU), 2024

    2024

  12. [20]

    Nonlinear Autopilot for Improving Guidance Performance of Dual-controlled Missiles With Lateral Thrust Regulation,

    I. H. Jeong and H. G. Kim, “Nonlinear Autopilot for Improving Guidance Performance of Dual-controlled Missiles With Lateral Thrust Regulation,” inProc. 39th Institute of Control, Robotics and Systems Conference, 2024, pp. 129–130

    2024

  13. [21]

    Wind compensation framework for unpowered aircraft using online waypoint correction,

    N. Cho, S. Lee, J. Kim, Y . Kim, S. Park, and C. Song, “Wind compensation framework for unpowered aircraft using online waypoint correction,”IEEE Transactions on Aerospace and Electronic Systems, vol. 56, no. 1, pp. 698–710, 2019

    2019

  14. [22]

    Attitude control in ascent phase of missile considering actuator non-linearity and wind disturbance,

    B. Fu, H. Qi, J. Xu, Y . Yang, S. Wang, and Q. Gao, “Attitude control in ascent phase of missile considering actuator non-linearity and wind disturbance,”Applied Sciences, vol. 9, no. 23, p. 5113, 2019

    2019

  15. [23]

    Safe multi-agent reinforcement learning for multi-robot control,

    S. Gu, J. Kuba, Y . Chen, Y . Du, L. Yang, A. Knoll, and Y . Yang, “Safe multi-agent reinforcement learning for multi-robot control,”Artificial Intelligence, vol. 319, p. 103905, 2023

    2023

  16. [24]

    Deep reinforcement learning for autonomous driving: A survey,

    B. R. Kiran, I. Sobh, V . Talpaert, P. Mannion, A. A. Al Sallab, S. Yogamani, and P. P ´erez, “Deep reinforcement learning for autonomous driving: A survey,”IEEE Transactions on Intelligent Transportation Systems, vol. 23, no. 6, pp. 4909–4926, 2021

    2021

  17. [25]

    Cooperative control for multi-player pursuit-evasion games with reinforcement learning,

    Y . Wang, L. Dong, and C. Sun, “Cooperative control for multi-player pursuit-evasion games with reinforcement learning,”Neurocomputing, vol. 412, pp. 101–114, 2020

    2020

  18. [26]

    An approach to multi-agent pursuit evasion games using reinforcement learning,

    A. T. Bilgin and E. Kadioglu-Urtis, “An approach to multi-agent pursuit evasion games using reinforcement learning,” in2015 International Conference on Advanced Robotics (ICAR), pp. 164–169, 2015

    2015

  19. [27]

    Game of drones: Multi-UA V pursuit-evasion game with online motion planning by deep reinforcement learning,

    R. Zhang, Q. Zong, X. Zhang, L. Dou, and B. Tian, “Game of drones: Multi-UA V pursuit-evasion game with online motion planning by deep reinforcement learning,”IEEE Transactions on Neural Networks and Learning Systems, vol. 34, no. 10, pp. 7900–7909, 2022. 13

    2022

  20. [28]

    Maximum Entropy Heterogeneous-Agent Reinforcement Learning,

    J. Liu, Y . Zhong, S. Hu, H. Fu, Q. Fu, X. Chang, and Y . Yang, “Maximum Entropy Heterogeneous-Agent Reinforcement Learning,” inThe Twelfth International Conference on Learning Representations, 2024. [Online]. Available: ¡url id=”cv4hd7h3om1t98gvtrp0” type=”url” status=”parsed...

    2024

  21. [29]

    Heterogeneous-Agent Reinforcement Learning,

    Y . Zhong, J. Grudzien Kuba, X. Feng, S. Hu, J. Ji, and Y . Yang, “Heterogeneous-Agent Reinforcement Learning,” inJ. Mach. Learn. Res., vol. 25, no. 32, pp. 1–67, 2024. [Online]. Available: http://jmlr.org/papers/ v25/23-0488.html

    2024

  22. [30]

    Mean field multi-agent reinforcement learning,

    Y . Yang, R. Luo, M. Li, M. Zhou, W. Zhang, and J. Wang, “Mean field multi-agent reinforcement learning,” inProc. Int. Conf. Mach. Learn., 2018, pp. 5571–5580

    2018

  23. [31]

    Age of information minimization using multi-agent UA Vs based on AI-enhanced mean field resource allocation,

    Y . Emami et al., “Age of information minimization using multi-agent UA Vs based on AI-enhanced mean field resource allocation,”IEEE Transactions on Vehicular Technology, 2024

    2024

  24. [32]

    Joint Resource Allocation for V2X Communications With Multi-Type Mean-Field Reinforcement Learning,

    Y . Xu et al., “Joint Resource Allocation for V2X Communications With Multi-Type Mean-Field Reinforcement Learning,”IEEE Transactions on Intelligent Transportation Systems, 2024

    2024

  25. [33]

    Mean Field Deep Reinforcement Learning for Fair and Efficient UA V Control,

    D. Chen, Q. Qi, Z. Zhuang, J. Wang, J. Liao, and Z. Han, “Mean Field Deep Reinforcement Learning for Fair and Efficient UA V Control,” IEEE Internet of Things Journal, vol. 8, no. 2, pp. 813–828, 2021. DOI: 10.1109/JIOT.2020.3008299

    2021

  26. [34]

    Multi type mean field reinforcement learning,

    S. G. Subramanian, P. Poupart, M. E. Taylor, and N. Hegde, “Multi type mean field reinforcement learning,”arXiv preprint arXiv:2002.02513, 2020

    2002 arXiv

  27. [35]

    Hierarchical mean-field deep reinforcement learning for large- scale multiagent systems,

    C. Yu, “Hierarchical mean-field deep reinforcement learning for large- scale multiagent systems,” inProc. AAAI Conf. Artif. Intell., vol. 37,

  28. [36]

    Weighted mean-field multi-agent reinforcement learning via reward attribution decomposition,

    T. Wu, W. Li, B. Jin, W. Zhang, and X. Wang, “Weighted mean-field multi-agent reinforcement learning via reward attribution decomposition,” inProc. Int. Conf. Database Syst. Adv. Appl., 2022, pp. 301–316. Springer

    2022

  29. [37]

    Attention Is All You Need,

    A. Vaswani et al., “Attention Is All You Need,” inProc. 31st Int. Conf. Neural Information Processing Systems (NIPS), 2017, pp. 5998–6008

    2017

  30. [38]

    Unsupervised representation learning in deep rein- forcement learning: A review,

    N. Botteghi et al., “Unsupervised representation learning in deep rein- forcement learning: A review,”arXiv preprint arXiv:2208.14226, 2022

    2022 arXiv

  31. [39]

    Fraccaro, S

    M. Fraccaro, S. Kamronn, U. Paquet, and O. Winther,A disentangled recognition and nonlinear dynamics model for unsupervised learning. Advances in Neural Information Processing Systems, vol. 30, 2017

    2017

  32. [40]

    Ha and J

    D. Ha and J. Schmidhuber,Recurrent world models facilitate policy evolution. Advances in Neural Information Processing Systems, vol. 31, 2018

    2018

  33. [41]

    Ha and J

    D. Ha and J. Schmidhuber,World models. arXiv preprint arXiv:1803.10122, 2018

    2018 arXiv

  34. [42]

    Van der Pol, T

    E. Van der Pol, T. Kipf, F. A. Oliehoek, and M. Welling,Plannable approximations to MDP homomorphisms: Equivariance under actions. arXiv preprint arXiv:2002.11963, 2020

    2002 arXiv

  35. [43]

    Dulac-Arnold et al.,Deep reinforcement learning in large discrete action spaces

    G. Dulac-Arnold et al.,Deep reinforcement learning in large discrete action spaces. arXiv preprint arXiv:1512.07679, 2015

    2015 arXiv

  36. [44]

    MA2CL: Masked Attentive Contrastive Learning for Multi-Agent Reinforcement Learning,

    H. Song et al., “MA2CL: Masked Attentive Contrastive Learning for Multi-Agent Reinforcement Learning,”arXiv preprint arXiv:2306.02006, 2023

    2023 arXiv

  37. [45]

    Learning action representations for reinforcement learning,

    Y . Chandak et al., “Learning action representations for reinforcement learning,” inProc. Int. Conf. Mach. Learn., 2019, pp. 941–950

    2019

  38. [46]

    The Hungarian method for the assignment problem,

    H. W. Kuhn, “The Hungarian method for the assignment problem,” Naval Research Logistics Quarterly, vol. 2, pp. 83–97, 1955

    1955

  39. [47]

    He, S., Wang, W., Lin, D., & Lei, H. (2017). Consensus-based two-stage salvo attack guidance.IEEE Transactions on Aerospace and Electronic Systems,54(3), 1555–1566. https://doi.org/10.1109/TAES.2017.2703360

    2017

  40. [48]

    Robust estimation of a location parameter,

    P. J. Huber, “Robust estimation of a location parameter,”Annals of Mathematical Statistics, vol. 35, no. 1, pp. 73–101, 1964

    1964

  41. [49]

    Continuous control with deep reinforcement learning,

    T. P. Lillicrap, J. J. Hunt, A. Pritzel, N. Heess, T. Erez, Y . Tassa, D. Silver, and D. Wierstra, “Continuous control with deep reinforcement learning,”arXiv preprint arXiv:1509.02971, 2015

    2015 arXiv

  42. [50]

    Addressing function approxima- tion error in actor-critic methods,

    S. Fujimoto, H. Hoof, and D. Meger, “Addressing function approxima- tion error in actor-critic methods,” inProc. Int. Conf. Mach. Learn., 2018, pp. 1587–1596

    2018

  43. [51]

    Multi-agent actor-critic for mixed cooperative-competitive envi- ronments,

    R. Lowe, Y . I. Wu, A. Tamar, J. Harb, P. Abbeel, and I. Mor- datch, “Multi-agent actor-critic for mixed cooperative-competitive envi- ronments,” inAdv. Neural Inf. Process. Syst., vol. 30, 2017. Li Wangreceived the B.S. degree from the School of Artificial Intelligence, Beiha...

    2017

  44. [53]

    degree at the same institution, under the supervision of Prof

    He is currently pursuing the M.S. degree at the same institution, under the supervision of Prof. Wenjun Wu. His research interests include swarm intelligence and robotics. Gangzheng Aireceived the B.S. degree from School of Aeronautics and Astronautics, Sun Yat-Sen Uni- versit...

    2022

  45. [2024]

    degree at the same institution, under the supervision of Prof

    He is currently pursuing the Ph.D. degree at the same institution, under the supervision of Prof. Wenjun Wu. His research interests include multi- agent reinforcement learning and large language models. Xin Yuis a Ph.D. student at the School of Computer Science and Engineering...

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.