pith. machine review for the scientific record. sign in

arxiv: 2605.06557 · v1 · submitted 2026-05-07 · 💻 cs.MA · cs.AI· cs.LG

Recognition: unknown

Coordination Matters: Evaluation of Cooperative Multi-Agent Reinforcement Learning

Afsaneh Doryab, Maria Ana Cardei, Matthew Landers

Authors on Pith no claims yet

Pith reviewed 2026-05-08 03:30 UTC · model grok-4.3

classification 💻 cs.MA cs.AIcs.LG
keywords cooperative multi-agent reinforcement learningcoordination evaluationtask allocationprocess-level diagnosticsvalue-based MARLassignment redundancyscaling in MARL
0
0 comments X

The pith

Similar returns in cooperative multi-agent reinforcement learning often mask distinct coordination mechanisms among agents.

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

The paper argues that common success metrics such as total return or completion time do not reveal how agents actually coordinate their actions, especially when the number of agents and tasks increases. It supplements these metrics with process-level checks that track redundant assignments, variety in task choices, and how efficiently tasks get completed. The approach is demonstrated in a controlled test environment that changes the scale of agents and tasks while keeping observation rules and task constraints the same. A reader would care because many real-world uses of multi-agent systems, from robotics to logistics, depend on reliable coordination rather than just high average scores. The results indicate that methods with matching returns can rely on quite different internal strategies for handling overlapping decisions.

Core claim

Similar return trends can reflect distinct coordination mechanisms, including differences in redundant assignment, assignment diversity, and task-completion efficiency. In commitment-constrained task allocation, performance under scale is shaped not only by nominal action-space size, but also by assignment pressure, sparse decision opportunities, and redundant choices among interdependent agents.

What carries the argument

The STAT testbed, a controlled commitment-constrained spatial task-allocation environment that varies the number of agents, tasks, and environment size while holding observation access and task rules fixed, paired with process-level diagnostics that measure redundant assignment, assignment diversity, and task-completion efficiency.

If this is right

  • Return-only evaluation can fail to distinguish coordination quality among value-based MARL methods even when aggregate scores look the same.
  • Scaling performance in task allocation depends on assignment pressure and the presence of redundant choices, not solely on the size of the action space.
  • Different centralization levels in MARL produce measurable variations in how often agents duplicate tasks or complete them efficiently.
  • Process diagnostics provide a practical way to compare methods beyond what success rate or total return alone can show.

Where Pith is reading between the lines

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

  • The same diagnostic approach could be adapted to other cooperative domains such as multi-robot navigation to detect hidden coordination failures.
  • Training procedures might improve by adding explicit penalties for redundant assignments during learning.
  • Benchmark suites for MARL could incorporate these process checks as standard reporting requirements to support fairer comparisons at larger scales.

Load-bearing premise

The selected process-level diagnostics and the controlled scale variations in the testbed capture the essential features of coordination in cooperative multi-agent settings.

What would settle it

A follow-up experiment in which two methods with different diagnostic profiles produce identical real-world coordination outcomes, such as the same rate of successful joint task completions without overlap, would challenge the claim that the diagnostics reveal meaningful differences.

Figures

Figures reproduced from arXiv: 2605.06557 by Afsaneh Doryab, Maria Ana Cardei, Matthew Landers.

Figure 1
Figure 1. Figure 1: Conflict rate provides a com￾plementary diagnostic beyond return. To instantiate this evaluation perspective, we use STAT (the Spatial Task Allocation Testbed), a configurable co￾operative MARL testbed that scales agents, tasks, and en￾vironment size under full observability. STAT uses action masking and finite-state commitment to isolate high-level task-allocation coordination. We leverage STAT to com￾par… view at source ↗
Figure 2
Figure 2. Figure 2: Illustration of STAT. Agents start at a fixed origin and must coordinate to com￾plete spatially distributed tasks efficiently. Agents distribute themselves across spatially dis￾tributed tasks, commit to selected assignments, and complete all tasks efficiently. This induces a structured combinatorial coordination problem, where each as￾signment decision interacts with the choices of other agents, while comm… view at source ↗
Figure 3
Figure 3. Figure 3: Illustration of the assignment-based process-level diagnostics used in this work. The top view at source ↗
Figure 4
Figure 4. Figure 4: Coordination-aware scaling analysis. Each row isolates one scaling axis: (A) environment view at source ↗
Figure 5
Figure 5. Figure 5: Finite-state commitment structure representing agent modes and valid transitions. view at source ↗
Figure 6
Figure 6. Figure 6: Learning curves for Baseline STAT configurations. Curves show mean evaluation per view at source ↗
Figure 7
Figure 7. Figure 7: Learning curves for Extreme STAT configurations. Curves show mean evaluation perfor view at source ↗
Figure 8
Figure 8. Figure 8: Full benchmark overview across STAT configurations. Final return summarizes task view at source ↗
Figure 9
Figure 9. Figure 9: Full benchmark overview across STAT configurations. Each heatmap reports the final view at source ↗
Figure 10
Figure 10. Figure 10: Additional mechanism-level scaling diagnostics. Each row isolates one controlled scaling view at source ↗
read the original abstract

Cooperative multi-agent reinforcement learning (MARL) benchmarks commonly emphasize aggregate outcomes such as return, success rate, or completion time. While essential, these metrics often fail to reveal how agents coordinate, particularly in settings where agents, tasks, and joint assignment choices scale combinatorially. We propose a coordination-aware evaluation perspective that supplements return with process-level diagnostics. We instantiate this perspective using STAT, a controlled commitment-constrained spatial task-allocation testbed that systematically varies agents, tasks, and environment size while holding observation access and task rules fixed. We evaluate six representative value-based MARL methods across varying levels of centralization. Our results show that similar return trends can reflect distinct coordination mechanisms, including differences in redundant assignment, assignment diversity, and task-completion efficiency. We find that in commitment-constrained task allocation, performance under scale is shaped not only by nominal action-space size, but also by assignment pressure, sparse decision opportunities, and redundant choices among interdependent agents. Our findings motivate coordination-aware evaluation as a necessary complement to return-based benchmarking for cooperative MARL.

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 claims that aggregate metrics such as return are insufficient to reveal coordination mechanisms in cooperative MARL, particularly under combinatorial scaling of agents and tasks. It introduces the STAT testbed—a commitment-constrained spatial task-allocation environment that varies agent count, task count, and environment size while fixing observation access and task rules—and evaluates six value-based MARL methods across centralization levels. The central empirical finding is that similar return trends can mask distinct coordination mechanisms (differences in redundant assignment, assignment diversity, and task-completion efficiency), and that performance under scale is shaped by assignment pressure, sparse decision opportunities, and redundant choices among interdependent agents.

Significance. If the empirical distinctions hold, the work is significant for motivating process-level diagnostics as a necessary complement to return-based benchmarking in cooperative MARL. The controlled STAT testbed enables systematic isolation of scaling factors, which is a methodological strength, and the focus on commitment-constrained allocation addresses a practically relevant setting. This could encourage more nuanced algorithm evaluation and design, especially for interdependent agent settings.

major comments (2)
  1. [Experiments section] Experiments section: the claims that similar return trends reflect distinct coordination mechanisms rest on observed differences in the process-level diagnostics, yet no statistical significance tests, error bars, number of independent runs, or data-exclusion rules are reported. Without these, it is impossible to determine whether the reported differences in redundant assignment or task-completion efficiency exceed experimental variance.
  2. [STAT testbed section] STAT testbed section: the central claim that the diagnostics isolate coordination mechanisms (assignment pressure, sparse decisions, redundant choices) independent of other factors is load-bearing for the generalizability argument. Because observation access and task rules are held fixed by construction, an ablation that perturbs the observation model or relaxes commitment constraints while holding nominal action-space size constant is needed to confirm the differences are not artifacts of the specific spatial layout or joint-action enumeration.
minor comments (2)
  1. [Abstract] Abstract: the six methods are referred to only as 'representative value-based MARL methods'; naming them (e.g., QMIX, VDN, etc.) would improve immediate clarity for readers.
  2. [Figures] Figures: legends and axis labels for the process-level diagnostic plots should explicitly define 'redundant assignment' and 'assignment diversity' to ensure the metrics are interpretable without returning to the text.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed feedback. We address each major comment point by point below, indicating where revisions have been made to strengthen the manuscript.

read point-by-point responses

  1. Referee: [Experiments section] Experiments section: the claims that similar return trends reflect distinct coordination mechanisms rest on observed differences in the process-level diagnostics, yet no statistical significance tests, error bars, number of independent runs, or data-exclusion rules are reported. Without these, it is impossible to determine whether the reported differences in redundant assignment or task-completion efficiency exceed experimental variance.

    Authors: We agree that the original submission lacked sufficient statistical reporting to rigorously support the observed differences in process-level diagnostics. In the revised manuscript, we now report all results as means over 5 independent runs with distinct random seeds, include error bars denoting one standard deviation, explicitly state that no data points were excluded beyond standard convergence checks, and add paired statistical significance tests (t-tests with p < 0.05 threshold) on the differences in redundant assignment rates and task-completion efficiency. These additions confirm that the reported distinctions between methods exceed experimental variance. revision: yes

  2. Referee: [STAT testbed section] STAT testbed section: the central claim that the diagnostics isolate coordination mechanisms (assignment pressure, sparse decisions, redundant choices) independent of other factors is load-bearing for the generalizability argument. Because observation access and task rules are held fixed by construction, an ablation that perturbs the observation model or relaxes commitment constraints while holding nominal action-space size constant is needed to confirm the differences are not artifacts of the specific spatial layout or joint-action enumeration.

    Authors: We appreciate the referee's concern regarding potential artifacts. However, the STAT testbed is deliberately designed to hold observation access and task rules fixed precisely to isolate the effects of combinatorial scaling on coordination under commitment constraints. Introducing ablations that perturb the observation model or relax commitments would change the fundamental problem class, confounding the very scaling factors under study. Our generalizability claims are scoped to commitment-constrained spatial allocation with fixed rules; we have added a clarifying paragraph in the revised STAT testbed section explaining why the current controlled variations suffice to attribute differences to assignment pressure and inter-agent redundancy rather than layout or enumeration artifacts. revision: partial

Circularity Check

0 steps flagged

No circularity: purely empirical evaluation of existing methods on new testbed

full rationale

The paper introduces the STAT testbed and applies process-level diagnostics (redundant assignment, assignment diversity, task-completion efficiency) to evaluate six existing value-based MARL algorithms under controlled variations in agents, tasks, and environment size. No derivations, equations, fitted parameters, or predictions are claimed; results are direct experimental outcomes. No self-citations are load-bearing for the central claims, and the evaluation does not reduce any quantity to its own inputs by construction. The work is self-contained as an empirical benchmarking study.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The central claims rest on the assumption that the selected diagnostics validly measure coordination quality and that STAT is representative of broader cooperative MARL challenges; no free parameters or invented physical entities.

axioms (1)
  • domain assumption Value-based MARL methods are representative of the broader class of cooperative algorithms for the purpose of this evaluation.
    Abstract states evaluation of six representative value-based methods without justifying why this class suffices.
invented entities (1)
  • STAT testbed no independent evidence
    purpose: Controlled environment for varying agents, tasks, and size while fixing observation access and task rules to isolate coordination effects.
    Newly proposed in the paper; no independent evidence outside this work.

pith-pipeline@v0.9.0 · 5488 in / 1260 out tokens · 52790 ms · 2026-05-08T03:30:46.401936+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

40 extracted references · 4 canonical work pages

  1. [1]

    Rtaw: An attention inspired reinforcement learning method for multi-robot task allocation in warehouse environments

    Aakriti Agrawal, Amrit Singh Bedi, and Dinesh Manocha. Rtaw: An attention inspired reinforcement learning method for multi-robot task allocation in warehouse environments. arXiv preprint arXiv:2209.05738, 2022

    work page arXiv 2022

  2. [2]

    Dc-mrta: Decen- tralized multi-robot task allocation and navigation in complex environments

    Aakriti Agrawal, Senthil Hariharan, Amrit Singh Bedi, and Dinesh Manocha. Dc-mrta: Decen- tralized multi-robot task allocation and navigation in complex environments. In2022 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2022

    2022

  3. [3]

    Dynamic multi-agent task allocation with spatial and temporal constraints

    Sofia Amador, Steven Okamoto, and Roie Zivan. Dynamic multi-agent task allocation with spatial and temporal constraints. InProceedings of the Twenty-Eighth AAAI Conference on Artificial Intelligence, pages 1384–1390, 2014

    2014

  4. [4]

    Who is helping whom? analyzing inter-dependencies to evaluate cooperation in human-ai teaming

    Upasana Biswas, Vardhan Palod, Siddhant Bhambri, and Subbarao Kambhampati. Who is helping whom? analyzing inter-dependencies to evaluate cooperation in human-ai teaming. Proceedings of the AAAI Conference on Artificial Intelligence, 40(21):17347–17356, 2026

    2026

  5. [5]

    Multi-agent reinforcement learning: A review of challenges and applications.Applied Sciences, 11(11):4948, 2021

    Lorenzo Canese, Gian Carlo Cardarilli, Luca Di Nunzio, Rocco Fazzolari, Daniele Giardino, Marco Re, and Sergio Spanò. Multi-agent reinforcement learning: A review of challenges and applications.Applied Sciences, 11(11):4948, 2021

    2021

  6. [6]

    Practical heuristics for victim tagging during a mass casualty incident emergency medical response

    Maria Ana Cardei and Afsaneh Doryab. Practical heuristics for victim tagging during a mass casualty incident emergency medical response. In2024 IEEE 20th International Conference on Automation Science and Engineering (CASE), pages 165–172, 2024

    2024

  7. [7]

    Factorized deep q-network for cooperative multi-agent reinforcement learning in victim tagging.IEEE Transactions on Automation Science and Engineering, 23:3109–3120, 2026

    Maria Ana Cardei and Afsaneh Doryab. Factorized deep q-network for cooperative multi-agent reinforcement learning in victim tagging.IEEE Transactions on Automation Science and Engineering, 23:3109–3120, 2026

    2026

  8. [8]

    Ho, Thomas L

    Micah Carroll, Rohin Shah, Mark K. Ho, Thomas L. Griffiths, Sanjit A. Seshia, Pieter Abbeel, and Anca Dragan.On the utility of learning about humans for human-AI coordination. Curran Associates Inc., Red Hook, NY , USA, 2019

    2019

  9. [9]

    Shared experience actor-critic for multi-agent reinforcement learning

    Filippos Christianos, Lukas Schäfer, and Stefano V Albrecht. Shared experience actor-critic for multi-agent reinforcement learning. InAdvances in Neural Information Processing Systems (NeurIPS), 2020

    2020

  10. [10]

    Oliehoek, Karl Tuyls, Daniel Hennes, and Wiebe van der Hoek

    Daniel Claes, Philipp Robbel, Frans A. Oliehoek, Karl Tuyls, Daniel Hennes, and Wiebe van der Hoek. Effective approximations for multi-robot coordination in spatially distributed tasks. InProceedings of the 2015 International Conference on Autonomous Agents and Multiagent Systems, pages 881–890, Richland, SC, 2015. International Foundation for Autonomous ...

    2015

  11. [11]

    Zhenhui Feng, Renbin Xiao, and Mingzhi Xiao. Spatial crowdsourcing task allocation for heterogeneous multi-task hybrid scenarios: A model-embedded role division approach.Frontiers of Information Technology & Electronic Engineering, 26:1144–1163, 2025

    2025

  12. [12]

    Counterfactual multi-agent policy gradients

    Jakob Foerster, Gregory Farquhar, Triantafyllos Afouras, Nantas Nardelli, and Shimon Whiteson. Counterfactual multi-agent policy gradients. InProceedings of the AAAI conference on artificial intelligence, volume 32, 2018

    2018

  13. [13]

    A formal analysis and taxonomy of task allocation in multi-robot systems.The International journal of robotics research, 23(9):939–954, 2004

    Brian P Gerkey and Maja J Matari ´c. A formal analysis and taxonomy of task allocation in multi-robot systems.The International journal of robotics research, 23(9):939–954, 2004

    2004

  14. [14]

    A survey and critique of multiagent deep reinforcement learning.Autonomous Agents and Multi-Agent Systems, 33(6):750–797, 2019

    Pablo Hernandez-Leal, Bilal Kartal, and Matthew E Taylor. A survey and critique of multiagent deep reinforcement learning.Autonomous Agents and Multi-Agent Systems, 33(6):750–797, 2019

    2019

  15. [15]

    Policy diagnosis via measuring role diversity in cooperative multi-agent reinforcement learning

    Siyi Hu, Fengda Zhu, Xiaojun Chang, and Xiaodan Liang. Policy diagnosis via measuring role diversity in cooperative multi-agent reinforcement learning. InProceedings of the 39th Inter- national Conference on Machine Learning, volume 162 ofProceedings of Machine Learning Research, pages 9041–9071. PMLR, 2022. 10

    2022

  16. [16]

    Steleac, Jonathan D

    Aleksandar Krnjaic, Raul D. Steleac, Jonathan D. Thomas, Georgios Papoudakis, Lukas Schäfer, Andrew Wing Keung To, Kuan-Ho Lao, Murat Cubuktepe, Matthew Haley, Peter Börsting, and Stefano V . Albrecht. Scalable multi-agent reinforcement learning for warehouse logistics with robotic and human co-workers. In2024 IEEE/RSJ International Conference on Intellig...

    2024

  17. [17]

    Spatial crowd- sourcing task allocation scheme for massive data with spatial heterogeneity.arXiv preprint arXiv:2310.12433, 2023

    Kun Li, Shengling Wang, Hongwei Shi, Xiuzhen Cheng, and Minghui Xu. Spatial crowd- sourcing task allocation scheme for massive data with spatial heterogeneity.arXiv preprint arXiv:2310.12433, 2023

    work page arXiv 2023

  18. [18]

    Michael L. Littman. Markov games as a framework for multi-agent reinforcement learning. In Proceedings of the Eleventh International Conference on Machine Learning, pages 157–163, 1994

    1994

  19. [19]

    Multi-agent actor- critic for mixed cooperative-competitive environments

    Ryan Lowe, Yi Wu, Aviv Tamar, Jean Harb, Pieter Abbeel, and Igor Mordatch. Multi-agent actor- critic for mixed cooperative-competitive environments. InProceedings of the 31st International Conference on Neural Information Processing Systems, NIPS’17, page 6382–6393, Red Hook, NY , USA, 2017. Curran Associates Inc

    2017

  20. [20]

    Maven: Multi-agent variational exploration

    Anuj Mahajan, Tabish Rashid, Mikayel Samvelyan, and Shimon Whiteson. Maven: Multi-agent variational exploration. InAdvances in Neural Information Processing Systems, volume 32, 2019

    2019

  21. [21]

    Maheswaran, Pedro A

    Rajiv T. Maheswaran, Pedro A. Szekely, Marcel Becker, Stephen Fitzpatrick, Gergely Gati, Jing Jin, Robert Neches, Narges Noori, Craig Milo Rogers, Romeo Sanchez, Kevin Smyth, and Chris VanBuskirk. Predictability and criticality metrics for coordination in complex environments. In Proceedings of the 7th International Joint Conference on Autonomous Agents a...

    2008

  22. [22]

    Human-level control through deep reinforcement learning.Nature, 518(7540):529–533, 2015

    V olodymyr Mnih, Koray Kavukcuoglu, David Silver, Andrei A Rusu, Joel Veness, Marc G Bellemare, Alex Graves, Martin Riedmiller, Andreas K Fidjeland, Georg Ostrovski, Stig Pe- tersen, Charles Beattie, Amir Sadik, Ioannis Antonoglou, Helen King, Dharshan Kumaran, Daan Wierstra, Shane Legg, and Demis Hassabis. Human-level control through deep reinforcement l...

    2015

  23. [23]

    Springer, 2016

    Frans A Oliehoek, Christopher Amato, et al.A concise introduction to decentralized POMDPs, volume 1. Springer, 2016

    2016

  24. [24]

    A review of cooperative multi-agent deep reinforce- ment learning.Applied Intelligence, 53:13677–13722, 2023

    Afshin Oroojlooy and Davood Hajinezhad. A review of cooperative multi-agent deep reinforce- ment learning.Applied Intelligence, 53:13677–13722, 2023

    2023

  25. [25]

    An extended benchmarking of multi-agent reinforcement learning algorithms in complex fully cooperative tasks

    George Papadopoulos, Andreas Kontogiannis, Foteini Papadopoulou, Chaido Poulianou, Ioannis Koumentis, and George V ouros. An extended benchmarking of multi-agent reinforcement learning algorithms in complex fully cooperative tasks. InProceedings of the 24th International Conference on Autonomous Agents and Multiagent Systems, AAMAS ’25, page 1613–1622, Ri...

    2025

  26. [26]

    Albrecht

    Georgios Papoudakis, Filippos Christianos, Lukas Schäfer, and Stefano V . Albrecht. Benchmark- ing multi-agent deep reinforcement learning algorithms in cooperative tasks. InProceedings of the Neural Information Processing Systems Track on Datasets and Benchmarks (NeurIPS), 2021

    2021

  27. [27]

    Qmix: Monotonic value function factorisation for deep multi-agent reinforcement learning

    Tabish Rashid, Mikayel Samvelyan, Christopher de Witt, Gregory Farquhar, Jakob N Foerster, and Shimon Whiteson. Qmix: Monotonic value function factorisation for deep multi-agent reinforcement learning. InProceedings of the 35th International Conference on Machine Learning (ICML), volume 80, pages 4295–4304. PMLR, 2018

    2018

  28. [28]

    The starcraft multi-agent challenge

    Mikayel Samvelyan, Tabish Rashid, Christian Schroeder de Witt, Gregory Farquhar, Nantas Nardelli, Tim G J Rudner, Philip H S Torr, Jakob Foerster, and Shimon Whiteson. The starcraft multi-agent challenge. InProceedings of the 18th International Conference on Autonomous Agents and MultiAgent Systems (AAMAS), pages 2186–2188, 2019. 11

    2019

  29. [29]

    Qtran: Learning to factorize with transformation for cooperative multi-agent reinforcement learning

    Kyunghwan Son, Daewoo Kim, Wan Ju Kang, David Hostallero, and Yung Yi. Qtran: Learning to factorize with transformation for cooperative multi-agent reinforcement learning. InPro- ceedings of the 36th International Conference on Machine Learning (ICML), pages 5887–5896. PMLR, 2019

    2019

  30. [30]

    When collaboration beats ability: Mixed- ability teams can outperform high-ability teams under coordination demands

    Younes Strittmatter, Rachael Skye, Samuel Lozano Iglesias, Samuel Liebana, Andrew Saxe, Miguel Ruiz-Garcia, Erin Teich, and Markus Spitzer. When collaboration beats ability: Mixed- ability teams can outperform high-ability teams under coordination demands. InProceedings of the Annual Meeting of the Cognitive Science Society, 2026

    2026

  31. [31]

    Value-Decomposition Networks For Cooperative Multi-Agent Learning

    Peter Sunehag, Guy Lever, Audrunas Gruslys, Wojciech M. Czarnecki, Vinícius Flores Zambaldi, Max Jaderberg, Marc Lanctot, Nicolas Sonnerat, Joel Z. Leibo, Karl Tuyls, and Thore Graepel. Value-decomposition networks for cooperative multi-agent learning.ArXiv, abs/1706.05296, 2017

    work page Pith review arXiv 2017

  32. [32]

    Multi-agent reinforcement learning: Independent vs

    Ming Tan. Multi-agent reinforcement learning: Independent vs. cooperative agents. InProceed- ings of the Tenth International Conference on Machine Learning (ICML 1993), pages 330–337, San Francisco, CA, USA, 1993. Morgan Kaufmann

    1993

  33. [33]

    QPLEX: Duplex dueling multi-agent q-learning

    Jianhao Wang, Zhizhou Ren, Terry Liu, Yang Yu, and Chongjie Zhang. QPLEX: Duplex dueling multi-agent q-learning. InInternational Conference on Learning Representations (ICLR), 2021

    2021

  34. [34]

    B. L. Welch. The generalization of Student’s problem when several different population variances are involved.Biometrika, 34(1/2):28–35, 1947

    1947

  35. [35]

    Task allocation with geographic partition in spatial crowdsourcing

    Guanyu Ye, Yan Zhao, Xuanhao Chen, and Kai Zheng. Task allocation with geographic partition in spatial crowdsourcing. InProceedings of the 30th ACM International Conference on Information & Knowledge Management, pages 2404–2413, 2021

    2021

  36. [36]

    The surprising effectiveness of ppo in cooperative multi-agent games

    Chao Yu, Akash Velu, Eugene Vinitsky, Jiaxuan Gao, Yu Wang, Alexandre Bayen, and Yi Wu. The surprising effectiveness of ppo in cooperative multi-agent games. InAdvances in Neural Information Processing Systems, volume 35, pages 24611–24624, 2022

    2022

  37. [37]

    Coordination between individual agents in multi-agent reinforcement learning

    Yongchao Zhang, Qingyu Yang, Dou An, and Weidong Chen. Coordination between individual agents in multi-agent reinforcement learning. InProceedings of the AAAI Conference on Artificial Intelligence, volume 35, pages 11387–11394, 2021

    2021

  38. [38]

    Task alloca- tion in spatial crowdsourcing: An efficient geographic partition framework.IEEE Transactions on Knowledge and Data Engineering, 36(9):4943–4955, 2024

    Yan Zhao, Xuanlei Chen, Guanyu Ye, Fangda Guo, Kai Zheng, and Xiaofang Zhou. Task alloca- tion in spatial crowdsourcing: An efficient geographic partition framework.IEEE Transactions on Knowledge and Data Engineering, 36(9):4943–4955, 2024

    2024

  39. [39]

    Ofcourse: A multi-agent reinforcement learning environment for order fulfillment.Advances in Neural Information Processing Systems, 36:34765–34777, 2023

    Yiheng Zhu, Yang Zhan, Xuankun Huang, Yuwei Chen, Jiangwen Wei, Wei Feng, Yinzhi Zhou, Haoyuan Hu, Jieping Ye, et al. Ofcourse: A multi-agent reinforcement learning environment for order fulfillment.Advances in Neural Information Processing Systems, 36:34765–34777, 2023. 12 A Code Release We release the STAT environment together with executable training a...

    2023

  40. [40]

    Centralized Training and Centralized ExecutionIn the CTCE paradigm, both learning and action selection are performed centrally over the full multi-agent system

    results are reported in Appendix C.1. Centralized Training and Centralized ExecutionIn the CTCE paradigm, both learning and action selection are performed centrally over the full multi-agent system. We employ a DQN and FDQN. 16 DQN.We extend Deep Q-Networks (DQN) [ 22], originally proposed for single-agent reinforcement learning, to a fully centralized mu...

    work page arXiv 2000