Pith. sign in

REVIEW 3 major objections 4 minor 295 references

For Other-Play, the standard zero-shot-coordination evaluation—inter-seed cross-play from a single implementation—survives variation in implementation details, with no meaningful gap to full cross-implementation cross-play.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

For Other-Play in Yokai, agents trained with different implementation details coordinate across implementations about as well as across seeds, supporting inter-seed cross-play as a proxy for cross-implementation evaluation.

T0 review reviewed 2026-08-05 challenge →

load-bearing objection Useful XIXP evaluation framework and a carefully run null result, but the 'reliable proxy' claim outruns the evidence: unexamined threshold discards half the implementations, and one-at-a-time variation from a single codebase is not independent implementation. the 3 major comments →

arxiv 2608.03644 v1 pith:XDIEVRFD submitted 2026-08-04 cs.AI cs.MA

Is Inter-Seed Cross-Play Enough? Evaluating the Robustness of Zero-Shot Coordination Algorithms to Implementation Details

Maksymilian Wolski , Nicholas Hoernle , Johannes Forkel , Jakob Foerster This is my paper
classification cs.AI cs.MA
keywords Zero-shot coordinationOther-PlayIndependent PPOCross-play evaluationImplementation detailsPPO implementation detailsYokaiMulti-agent reinforcement learning
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The reading

Zero-shot coordination (ZSC) algorithms are meant to let agents trained by separate parties coordinate without any shared training. Because evaluating that ideal requires multiple independent implementations—too expensive for regular use—the field has settled for a cheaper proxy: train one implementation with many random seeds and cross-play the resulting policies. This paper asks whether that proxy is trustworthy. It builds a new evaluation scheme, cross-implementation cross-play (XIXP), that varies code-level implementation details of the underlying PPO algorithm, and runs it on Other-Play, a leading ZSC method, in the Yokai environment. The result is a null gap between within-implementation and cross-implementation cross-play, supporting the standard practice for this algorithm and benchmark.

Core claim

The paper's claim is that inter-seed cross-play is a reliable proxy for cross-implementation evaluation in zero-shot coordination. To support this, it defines XIXP as the average cross-play score over all policy pairs drawn from two different implementations, and WIXP as the average within-implementation cross-play score; the gap WIXP−XIXP is the measure of implementation-induced miscoordination. The authors train 22 implementations of Other-Play with IPPO in the Yokai environment, varying λGAE, learning-rate scheduling, gradient and value-function clipping, weight initialization, hidden-layer counts, minibatch counts, discount factor, entropy coefficient, and network architecture. After dis

What carries the argument

Cross-implementation cross-play (XIXP): a scheme that turns the standard single-implementation, multi-seed evaluation into a full matrix of implementations. For each pair of distinct implementations it averages cross-play over all seed pairs, and the within-implementation average (WIXP) is the traditional inter-seed cross-play score. The decisive object is the gap WIXP−XIXP; if it stays near zero, implementation details are not creating new coordination failures beyond those already seen across seeds. The paper also uses a paired standard-error estimator, in which each seed contributes exactly one inter-seed pairing per implementation, to keep confidence intervals valid.

Load-bearing premise

The load-bearing premise is that the curated list of varied implementation details (λGAE, clipping, initialization, architecture, etc.) faithfully represents the spread of genuinely independent implementations—if independent codebases differ in ways not covered here, the null result may not generalize.

What would settle it

Train two or more genuinely independent implementations of Other-Play from the same specification (ideally written by separate teams, then exchanged), apply the same XIXP protocol on Yokai, and see whether the WIXP−XIXP gap stays within the overlapping-confidence-interval range; a large gap would refute the proxy. A cheaper check: add an unvaried detail such as optimizer choice (Adam vs SGD), observation preprocessing, or network width; if the resulting gap is statistically meaningful, inter-seed cross-play is not enough even within the simulated-variation setting.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • ZSC papers can keep using inter-seed cross-play as the primary evaluation without commissioning multiple independent implementations, at least for Other-Play-style algorithms on benchmarks where the competence filter is applied.
  • The XIXP protocol—generate implementations, filter by self-play competence, compare WIXP to XIXP—gives future work a concrete template for testing whether a new ZSC algorithm is more or less sensitive to implementation details.
  • Nearly half of the generated implementations (11 of 22) failed the self-play competence threshold, so the proxy only holds once implementations are competent; the null result does not license skipping quality control.
  • The seed-pairing standard-error estimator shows how a small seed count can distort conclusions; adopting it would make reported confidence intervals in ZSC papers more honest.
  • The finding gives an empirical precedent for treating Other-Play as robust to specification ambiguity in the Yokai environment, shifting the burden of proof onto claims that implementation details do break coordination.

Where Pith is reading between the lines

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

  • Because the paper varies only a curated list of PPO details, the representative-details assumption is a testable extension: compare XIXP against implementations written truly independently from the same specification, and if a gap shows up, the null result is limited to simulated variation.
  • If the null result generalizes to other ZSC algorithms, it would suggest that the symmetry-avoiding design principle behind Other-Play also absorbs implementation noise, making algorithmic robustness an emergent property rather than a separate engineering concern.
  • The WIXP−XIXP gap could become a standard regression metric in ZSC research—reported alongside seed counts—so that a claimed ZSC algorithm is judged on how much its coordination survives code-level variation, not just seed variation.
  • A natural extension is to run the same protocol on Off-Belief Learning and Q-learning variants, since the paper explicitly leaves those open; finding a large gap there would map the boundary of the proxy's validity.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 4 minor

Summary. The paper asks whether the standard zero-shot coordination (ZSC) evaluation practice — training a single implementation across random seeds and using inter-seed cross-play as a proxy for cross-implementation coordination — is justified. It introduces a new evaluation scheme, cross-implementation cross-play (XIXP), in which implementation variants of the same ZSC algorithm are trained and evaluated against each other. Using Other-Play with IPPO in the Yokai environment, the authors vary nine PPO implementation details (λGAE, learning-rate schedule, gradient/value clipping, initialization, hidden layers, minibatches, discount factor, entropy coefficient, architecture). They train 176 policies across 22 implementations with 8 seeds each, discard 11 implementations whose self-play score is below 5, and report overlapping 95% confidence intervals for WIXP (4.8892, CI 4.6109–5.1675) and XIXP (4.8487, CI 4.7548–4.9427). They conclude that there is no meaningful WIXP−XIXP gap and that inter-seed cross-play is a reliable proxy for cross-implementation evaluation. The paper explicitly limits its scope to one environment, one base algorithm, and one ZSC algorithm, leaving generalization as an open question.

Significance. If the result holds, this is a valuable contribution: it is the first systematic empirical evaluation of a widely used but unexamined evaluation shortcut in ZSC research, and it introduces a reusable XIXP framework. The study is methodologically careful in several respects: it uses 8 seeds per implementation, adopts a CI estimator that avoids the non-independence of all-pairs cross-play, and is attentive to multimodality in cross-play scores. These are real strengths and should be credited. However, the central claim is currently supported only by an informal reading of overlapping confidence intervals and by a filtering step that removes half of the implementations without a sensitivity analysis. The paper is therefore suggestive rather than conclusive, and the breadth of the conclusion in the abstract and Section 6 exceeds what the evidence can bear.

major comments (3)
  1. [Section 5, Table 1] The central inference — 'no meaningful WIXP−XIXP gap' — is based solely on overlapping 95% confidence intervals. Overlap of two CIs is not evidence of equivalence; it is compatible with a range of true differences, including ones that would undermine the 'reliable proxy' claim. The authors should report a confidence interval for the pairwise difference WIXP−XIXP, or perform a two one-sided tests (TOST) equivalence analysis against a pre-specified and justified bound. Additionally, the CI estimator is only cited to Forkel et al. (2025) and not described; without the estimator formula, the reported intervals are not reproducible from the text.
  2. [Section 5, SP<5 filter (Figures 4 and 5)] Eleven of the 22 trained implementations are discarded because their self-play score is below 5. This threshold is arbitrary and no sensitivity analysis is provided. As Figure 4 shows, the filter removes all feedforward implementations and several entropy/architecture/PPO variants; these are precisely the implementations that could exhibit a WIXP−XIXP gap if their poor self-play does not translate equally to cross-play. The conclusion is therefore conditional on an unexamined selection step. The authors should justify the threshold from the score distribution, report results for alternative thresholds (e.g., 3, 4, 5, 6), or include all 22 implementations in a supplementary analysis.
  3. [Sections 4.2 and 6] The implementations used in the study are generated by varying one PPO detail at a time from a single reference codebase. Independent implementations typically differ in combinations of details and in choices not varied here (optimizer details, observation preprocessing, training budgets, replay buffer, etc.). Yet the abstract and Section 6 generalize to 'the standard evaluation practice in ZSC research.' This is a large inferential leap from one environment (Yokai), one base algorithm (IPPO), and one ZSC algorithm (Other-Play). The Section 6 limitations paragraph acknowledges this, but the main claim should be proportionately restricted or supported by additional experiments with combined variations or independently written code.
minor comments (4)
  1. [Equations (4)-(5) and Figure 2/5 captions] There is an inconsistency: Eq. (4) defines XIXP(L_k,L_k) = XP(L_k), i.e., within-implementation inter-seed cross-play, and WIXP is the average of these diagonal entries. However, the main text says the diagonal tiles in Figure 2 'represent self-play scores for each policy.' Please clarify whether the diagonal in the XIXP matrix shows self-play or within-implementation cross-play; if it is self-play, then WIXP is not computed from the displayed matrix.
  2. [Section 4.2 and Appendix] The paper does not provide a full hyperparameter table, code release, or environment details needed to reproduce the 22 implementations. Given that the paper's entire argument is about implementation details, a complete list of all varied and fixed hyperparameters (optimizer, learning rate, normalizations, etc.) is essential. Also specify the score range in Yokai and the units of the SP<5 threshold.
  3. [Section 4.2, CI estimator] The 'estimators proposed by Forkel et al. [2025]' are central to the statistical conclusion but are not defined. Please include the estimator equations or an appendix derivation so the reader can verify the claimed independence properties.
  4. [Throughout] Minor language and labeling issues: Figure 3/4/5 axis labels are incomplete (e.g., x-axis labeled 'Metric'), and the duplicated implementation names in Figure 5 make it hard to count the 11 retained implementations. Please clean up the figure presentation.

Circularity Check

0 steps flagged

No significant circularity: the WIXP-vs-XIXP comparison is a direct empirical measurement, not a construction or fitted prediction, and the self-citations are non-load-bearing tools.

full rationale

The paper is an empirical study, not a formal derivation, so the main circularity patterns do not apply. The central quantity WIXP−XIXP is measured directly from trained policies via Equations (4) and (5): WIXP is the average inter-seed cross-play within each implementation, and XIXP is the average cross-play across different implementations. These are operationally distinct objects, and no parameter is fitted to make the gap small; the gap is simply observed to be near zero. The implementation variations (Section 4.2) are grounded in PPO implementation-detail literature (Huang et al., 2022) and are varied one at a time from a single base codebase; this is a limitation on external validity, not a circularity. The competence filter SP<5 (Section 5) discards incompetent policies, which could in principle affect the conclusion, but the retained implementations are still distinct and the comparison is not definitionally forced. The paper explicitly acknowledges its limited scope in Section 6: 'The computational cost of XIXP evaluation restricted our analysis to a single environment (Yokai), a single base algorithm (IPPO), and a single ZSC algorithm (Other-Play).' This honest limitation statement further shows the conclusion is not presented as a tautology. Citations to prior work by the same authors (e.g., Forkel et al. 2025 for standard-error estimators, Ruhdorfer et al. 2026 for the Yokai environment, Hu et al. 2020 for Other-Play) are used as tools or background; the empirical WIXP-vs-XIXP comparison does not reduce to any of these citations. No equation in the paper is equivalent to its input by construction, and the main claim is an empirical finding rather than a derived identity.

Axiom & Free-Parameter Ledger

1 free parameters · 3 axioms · 0 invented entities

The central claim rests on two empirical assumptions: that the selected implementation details are representative of real independent implementations, and that Yokai is representative of ZSC environments. The free parameter is the self-play threshold used to filter implementations. No new entities are postulated.

free parameters (1)
  • self-play competence threshold = 5
    Implementations with SP(Lk)<5 were discarded; this threshold is chosen by hand and directly affects which implementations enter the XIXP analysis, potentially biasing the null result.
axioms (3)
  • domain assumption The implementation details varied (lambda-GAE, LR schedule, gradient clipping, value function clipping, weight init, hidden layers, minibatches, discount factor) are the ones that matter for independent implementations of IPPO/Other-Play.
    The paper selects these details based on prior work (Huang et al. 2022, Yu et al. 2022, Forkel et al. 2025), but there is no guarantee that the full space of implementer differences is covered; this is Section 4.2.
  • domain assumption The Yokai environment is a representative ZSC benchmark for evaluating robustness to implementation details.
    The paper evaluates only Yokai and acknowledges the limitation in Section 6; generalizing to other environments is an open question.
  • domain assumption The seed-pairing estimator from Forkel et al. 2025 preserves independence for valid confidence intervals.
    The paper relies on this estimator (Section 4.2), and the validity of the CI overlap conclusion depends on it.

reviewed 2026-08-05 · how reviews work

0 comments
Cite this review

Pith. "Pith review of Is Inter-Seed Cross-Play Enough? Evaluating the Robustness of Zero-Shot Coordination Algorithms to Implementation Details." pith.science (2026). https://pith.science/paper/XDIEVRFD

@misc{pith2026260803644,
  author       = {Pith},
  title        = {Pith review of: Is Inter-Seed Cross-Play Enough? Evaluating the Robustness of Zero-Shot Coordination Algorithms to Implementation Details},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XDIEVRFD}},
  note         = {Machine review of arXiv:2608.03644}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

AI agents deployed in real-world settings must be capable of coordinating with humans and other AI agents they have not encountered before. Zero-shot coordination (ZSC) algorithms aim to achieve this by specifying high-level learning rules such that independently engineered agents can coordinate with each other at test time. Rigorous evaluation of ZSC algorithms remains difficult: ideally, multiple independent implementations of each proposed algorithm must be used, reflecting the variation that arises when independent parties interpret and implement the same specification. In practice, however, ZSC algorithms have almost exclusively been evaluated using a single implementation trained across different random seeds, with only a handful of works additionally varying the neural network architecture. This leaves open questions about robustness to specification ambiguities and implementation details. In this work, we provide the first systematic evaluation of this robustness. We introduce a new evaluation scheme, cross-implementation cross-play, varying implementation details that prior work has shown to affect the performance of multi-agent reinforcement learning (MARL) algorithms, and we evaluate Other-Play, a popular ZSC algorithm, with this scheme. Our findings are encouraging and suggest that, for Other-Play, the standard ZSC evaluation is, in fact, a reasonable proxy for this more thorough cross-implementation evaluation.

Figures

Figures reproduced from arXiv: 2608.03644 by Jakob Foerster, Johannes Forkel, Maksymilian Wolski, Nicholas Hoernle.

Figure 1
Figure 1. Figure 1: Overview of the cross-implementation cross-play (XIXP) evaluation procedure. For a [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: XIXP matrix of Other-Play trained with IPPO across GRU and LSTM architectures and [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Self-play scores of IPPO in Yokai across three neural network architectures (GRU, LSTM, feedforward), with and without Other￾Play, at entropy coefficients 0.01 and 0.05. A higher entropy coefficient (0.05) consistently im￾proves self-play performance across all architec￾tures. Only IPPO with the LSTM and GRU archi￾tectures generate competent policies. 0 2 4 6 8 Metric GRU+OP (ent=0.05, 8 minibatches) LSTM+… view at source ↗
Figure 4
Figure 4. Figure 4: Self-play and cross-play scores for all IPPO implementations using GRU and LSTM architectures trained with Other-Play, across all implementation details considered in this study (prior to filtering). Several implementation details produce policies with insufficient self-play perfor￾mance and are excluded from the XIXP analysis. 9 [PITH_FULL_IMAGE:figures/full_fig_p009_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: XIXP matrix of Other-Play trained with IPPO across GRU and LSTM architectures and [PITH_FULL_IMAGE:figures/full_fig_p010_5.png] view at source ↗

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

295 extracted references · 53 canonical work pages · 28 internal anchors

  1. [1]

    Benchmarking Open-Ended Multi-Agent Coordination in Language Agents

    Tessera, Kale-ab Abebe and Szecsenyi, Andras and Barker, Cameron and Rutherford, Alexander and Paglieri, Davide and Scannell, Aidan and Gouk, Henry and Crowley, Elliot J. and Rocktäschel, Tim and Storkey, Amos , month = jun, year =. Benchmarking. doi:10.48550/arXiv.2606.08340 , abstract =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2606.08340

  2. [2]

    doi:10.48550/arXiv.2603.19312 , abstract =

    Maes, Lucas and Lidec, Quentin Le and Scieur, Damien and LeCun, Yann and Balestriero, Randall , month = mar, year =. doi:10.48550/arXiv.2603.19312 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2603.19312

  3. [3]

    Expected

    Muglich, Darius and Forkel, Johannes and Pol, Elise van der and Foerster, Jakob Nicolaus , year =. Expected. The

  4. [4]

    Equivariant networks for zero-shot coordination , isbn =

    Muglich, Darius and de Witt, Christian Schroeder and van der Pol, Elise and Whiteson, Shimon and Foerster, Jakob , month = nov, year =. Equivariant networks for zero-shot coordination , isbn =. Proceedings of the 36th

  5. [5]

    Cui, Brandon and Hu, Hengyuan and Pineda, Luis and Foerster, Jakob , editor =. K-level. Advances in. 2021 , pages =

    2021

  6. [6]

    , month = nov, year =

    Cui, Brandon and Hu, Hengyuan and Lupu, Andrei and Sokota, Samuel and Foerster, Jakob N. , month = nov, year =. Off-team learning , isbn =. Proceedings of the 36th

  7. [7]

    Dizdarević, Tin and Hammond, Ravi and Gessler, Tobias and Calinescu, Anisoara and Cook, Jonathan and Gallici, Matteo and Lupu, Andrei and Foerster, Jakob Nicolaus , editor =. Ad-. Proceedings of the 42nd. 2025 , pages =

    2025

  8. [8]

    How foundation models will revolutionize robot swarms , volume =

    Strobel, Volker and Dorigo, Marco and Fritz, Mario , month = apr, year =. How foundation models will revolutionize robot swarms , volume =. Science Robotics , publisher =. doi:10.1126/scirobotics.adz1543 , abstract =

    work page doi:10.1126/scirobotics.adz1543

  9. [9]

    Ye, Seonghyeon and Ge, Yunhao and Zheng, Kaiyuan and Gao, Shenyuan and Yu, Sihyun and Kurian, George and Indupuru, Suneel and Tan, You Liang and Zhu, Chuning and Xiang, Jiannan and Malik, Ayaan and Lee, Kyungmin and Liang, William and Ranawaka, Nadun and Gu, Jiasheng and Xu, Yinzhen and Wang, Guanzhi and Hu, Fengyuan and Narayan, Avnish and Bjorck, Johan ...

    Pith/arXiv arXiv doi:10.48550/arxiv.2602.15922

  10. [10]

    Yu, Chao and Velu, Akash and Vinitsky, Eugene and Gao, Jiaxuan and Wang, Yu and Bayen, Alexandre and Wu, Yi , month = jun, year =. The

  11. [11]

    Witt, Christian Schroeder de and Gupta, Tarun and Makoviichuk, Denys and Makoviychuk, Viktor and Torr, Philip H. S. and Sun, Mingfei and Whiteson, Shimon , month = nov, year =. Is. doi:10.48550/arXiv.2011.09533 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2011.09533 2011

  12. [12]

    Treutlein, Johannes and Dennis, Michael and Oesterheld, Caspar and Foerster, Jakob , month = jul, year =. A. Proceedings of the 38th

  13. [13]

    Proximal

    Schulman, John and Wolski, Filip and Dhariwal, Prafulla and Radford, Alec and Klimov, Oleg , month = aug, year =. Proximal. doi:10.48550/arXiv.1707.06347 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.1707.06347

  14. [14]

    and Spaan, Matthijs T

    Oliehoek, Frans A. and Spaan, Matthijs T. J. and Vlassis, Nikos , month = may, year =. Dec-

  15. [15]

    Expected

    Muglich, Darius and Forkel, Johannes and Pol, Elise van der and Foerster, Jakob Nicolaus , month = oct, year =. Expected

  16. [16]

    , month = may, year =

    Lucas, Keane and Allen, Ross E. , month = may, year =. Any-. Proceedings of the 21st

  17. [17]

    Hu, Hengyuan and Lerer, Adam and Cui, Brandon and Pineda, Luis and Brown, Noam and Foerster, Jakob , month = jul, year =. Off-. Proceedings of the 38th

  18. [18]

    Hu, Hengyuan and Lerer, Adam and Peysakhovich, Alex and Foerster, Jakob , month = nov, year =. “. Proceedings of the 37th

  19. [19]

    Schulman, John and Moritz, Philipp and Levine, Sergey and Jordan, Michael and Abbeel, Pieter , month = oct, year =. High-. doi:10.48550/arXiv.1506.02438 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.1506.02438

  20. [20]

    Ruhdorfer, Constantin and Bortoletto, Matteo and Forkel, Johannes and Foerster, Jakob and Bulling, Andreas , month = mar, year =. The. doi:10.48550/arXiv.2508.12480 , abstract =

    work page doi:10.48550/arxiv.2508.12480

  21. [21]

    Continuous Coordination As a Realistic Scenario for Lifelong Learning

    Nekoei, Hadi and Badrinaaraayanan, Akilesh and Courville, Aaron and Chandar, Sarath , month = jun, year =. Continuous. doi:10.48550/arXiv.2103.03216 , abstract =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2103.03216

  22. [22]

    ZSC-Eval: An Evaluation Toolkit and Benchmark for Multi-agent Zero-shot Coordination

    Wang, Xihuai and Zhang, Shao and Zhang, Wenhao and Dong, Wentao and Chen, Jingxiao and Wen, Ying and Zhang, Weinan , month = sep, year =. doi:10.48550/arXiv.2310.05208 , abstract =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2310.05208

  23. [23]

    , month = jan, year =

    Lucas, Keane and Allen, Ross E. , month = jan, year =. Any-. doi:10.48550/arXiv.2201.12436 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2201.12436

  24. [24]

    Nekoei, Hadi and Zhao, Xutong and Rajendran, Janarthanan and Liu, Miao and Chandar, Sarath , month = aug, year =. Towards. doi:10.48550/arXiv.2308.10284 , abstract =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2308.10284

  25. [25]

    arXiv.org , author =

    High entropy leads to symmetry equivariant policies in. arXiv.org , author =

  26. [26]

    Huang, Shengyi and Dossa, Rousslan Fernand Julien and Raffin, Antonin and Kanervisto, Anssi and Wang, Weixun , year =. The 37

  27. [27]

    Li, Tianyi and Chen, Mingda and Guo, Bowei and Shen, Zhiqiang , month = dec, year =. A. doi:10.48550/arXiv.2508.10875 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2508.10875

  28. [28]

    doi:10.48550/arXiv.2503.01935 , abstract =

    Zhu, Kunlun and Du, Hongyi and Hong, Zhaochen and Yang, Xiaocheng and Guo, Shuyi and Wang, Zhe and Wang, Zhenhailong and Qian, Cheng and Tang, Xiangru and Ji, Heng and You, Jiaxuan , month = mar, year =. doi:10.48550/arXiv.2503.01935 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2503.01935

  29. [29]

    doi:10.48550/arXiv.2602.18224 , abstract =

    Luo, Yuankai and Chen, Woping and Liang, Tong and Wang, Baiqiao and Li, Zhenguo , month = feb, year =. doi:10.48550/arXiv.2602.18224 , abstract =

    work page doi:10.48550/arxiv.2602.18224

  30. [30]

    doi:10.48550/arXiv.2502.09560 , abstract =

    Yang, Rui and Chen, Hanyang and Zhang, Junyu and Zhao, Mark and Qian, Cheng and Wang, Kangrui and Wang, Qineng and Koripella, Teja Venkat and Movahedi, Marziyeh and Li, Manling and Ji, Heng and Zhang, Huan and Zhang, Tong , month = jun, year =. doi:10.48550/arXiv.2502.09560 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2502.09560

  31. [31]

    Tan, Elizabeth S. Z. and Soubki, Adil and Cranmer, Miles , month = feb, year =. doi:10.48550/arXiv.2602.21307 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2602.21307

  32. [32]

    CoDe: Communication Delay-Tolerant Multi-Agent Collaboration via Dual Alignment of Intent and Timeliness

    Song, Shoucheng and Lin, Youfang and Han, Sheng and Yao, Chang and Wu, Hao and Wang, Shuo and Lv, Kai , month = jan, year =. doi:10.48550/arXiv.2501.05207 , abstract =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2501.05207

  33. [33]

    Qian, Chen and Xie, Zihao and Wang, YiFei and Liu, Wei and Zhu, Kunlun and Xia, Hanchen and Dang, Yufan and Du, Zhuoyun and Chen, Weize and Yang, Cheng and Liu, Zhiyuan and Sun, Maosong , month = mar, year =. Scaling. doi:10.48550/arXiv.2406.07155 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2406.07155

  34. [34]

    doi:10.48550/arXiv.2602.13476 , abstract =

    Hirose, Noriaki and Glossop, Catherine and Shah, Dhruv and Levine, Sergey , month = feb, year =. doi:10.48550/arXiv.2602.13476 , abstract =

    work page doi:10.48550/arxiv.2602.13476

  35. [35]

    Selden, Mark and Zhou, Jason and Campos, Felipe and Lambert, Nathan and Drew, Daniel and Pister, Kristofer S. J. , month = aug, year =. doi:10.48550/arXiv.2108.13606 , abstract =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2108.13606

  36. [36]

    Chen, Baiming and Xu, Mengdi and Li, Liang and Zhao, Ding , month = may, year =. Delay-. doi:10.48550/arXiv.2005.05440 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2005.05440 2005

  37. [37]

    Yang, Guang and Yang, Tianpei and Qiao, Jingwen and Wu, Yanqing and Huo, Jing and Chen, Xingguo and Gao, Yang , month = mar, year =. Multi-. doi:10.48550/arXiv.2512.03528 , abstract =

    work page doi:10.48550/arxiv.2512.03528

  38. [38]

    Asynchronous

    Yu, Chao and Yang, Xinyi and Gao, Jiaxuan and Chen, Jiayu and Li, Yunfei and Liu, Jijia and Xiang, Yunfei and Huang, Ruixin and Yang, Huazhong and Wu, Yi and Wang, Yu , month = apr, year =. Asynchronous. doi:10.48550/arXiv.2301.03398 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2301.03398

  39. [39]

    Pezeshkpour, Pouya and Hruschka, Estevam , month = feb, year =. From. doi:10.48550/arXiv.2602.02760 , abstract =

    work page doi:10.48550/arxiv.2602.02760

  40. [40]

    Benchmarking

    Ruan, Kai and Huang, Mowen and Wen, Ji-Rong and Sun, Hao , month = oct, year =. Benchmarking. doi:10.48550/arXiv.2505.04364 , abstract =

    doi:10.48550/arxiv.2505.04364

  41. [41]

    Yan, Bingyu and Zhou, Zhibo and Zhang, Litian and Zhang, Lian and Zhou, Ziyi and Miao, Dezhuang and Li, Zhoujun and Li, Chaozhuo and Zhang, Xiaoming , month = jun, year =. Beyond. doi:10.48550/arXiv.2502.14321 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2502.14321

  42. [42]

    Chen, Jingdi and Yang, Hanqing and Liu, Zongjun and Joe-Wong, Carlee , month = feb, year =. The. doi:10.48550/arXiv.2602.11583 , abstract =

    work page doi:10.48550/arxiv.2602.11583

  43. [43]

    doi:10.48550/arXiv.2510.26536 , abstract =

    Tan, Huajie and Chi, Cheng and Chen, Xiansheng and Ji, Yuheng and Zhao, Zhongxia and Hao, Xiaoshuai and Lyu, Yaoxu and Cao, Mingyu and Zhao, Junkai and Lyu, Huaihai and Zhou, Enshen and Chen, Ning and Fu, Yankai and Peng, Cheng and Guo, Wei and Liang, Dong and Chen, Zhuo and Lyu, Mengsi and He, Chenrui and Ao, Yulong and Lin, Yonghua and Wang, Pengwei and...

    work page doi:10.48550/arxiv.2510.26536

  44. [44]

    and Bishop, Hugh , year =

    Bishop, Christopher M. and Bishop, Hugh , year =. Deep. doi:10.1007/978-3-031-45468-4 , keywords =

    doi:10.1007/978-3-031-45468-4

  45. [45]

    COOP$^2$: Defining, Observing, and Repairing Cooperation in LLM Multi-Agent Systems

    Yang, Hanqing and Chen, Shiyu and Nourzad, Narjes and Siew, Marie and Chen, Jingdi and Joe-Wong, Carlee , month = feb, year =. doi:10.48550/arXiv.2603.00349 , abstract =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2603.00349

  46. [46]

    Zhu, Rui-Jie and Wang, Zixuan and Hua, Kai and Zhang, Tianyu and Li, Ziniu and Que, Haoran and Wei, Boyi and Wen, Zixin and Yin, Fan and Xing, He and Li, Lu and Shi, Jiajun and Ma, Kaijing and Li, Shanda and Kergan, Taylor and Smith, Andrew and Qu, Xingwei and Hui, Mude and Wu, Bohong and Min, Qiyang and Huang, Hongzhi and Zhou, Xun and Ye, Wei and Liu, J...

    Pith/arXiv arXiv doi:10.48550/arxiv.2510.25741

  47. [47]

    Wang, Kevin and Javali, Ishaan and Bortkiewicz, Michał , keywords =. 1000

  48. [48]

    Eckhaus, Niv and Berger, Uri and Stanovsky, Gabriel , month = sep, year =. Time to. doi:10.48550/arXiv.2506.05309 , abstract =

    work page doi:10.48550/arxiv.2506.05309

  49. [49]

    Cooperation

    Nishimoto, Keita and Asatani, Kimitaka and Sakata, Ichiro , month = feb, year =. Cooperation. doi:10.48550/arXiv.2602.11754 , abstract =

    work page doi:10.48550/arxiv.2602.11754

  50. [50]

    State of

    Reuss, Moritz , keywords =. State of

  51. [51]

    ScieNce RoboticS , author =

    Learning a thousand tasks in a day , language =. ScieNce RoboticS , author =. 2025 , keywords =

    2025

  52. [52]

    Trajectory

    Lupu, Andrei and Cui, Brandon and Hu, Hengyuan and Foerster, Jakob , month = jul, year =. Trajectory. Proceedings of the 38th

  53. [53]

    doi:10.48550/arXiv.2505.13729 , abstract =

    Rajvanshi, Abhinav and Sahu, Pritish and Shan, Tixiao and Sikka, Karan and Chiu, Han-Pang , month = may, year =. doi:10.48550/arXiv.2505.13729 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2505.13729

  54. [54]

    Zheng, Yujia and Zhao, Zhuokai and Li, Zijian and Xie, Yaqi and Gao, Mingze and Zhang, Lizhu and Zhang, Kun , month = oct, year =. Thought. doi:10.48550/arXiv.2510.20733 , abstract =

    doi:10.48550/arxiv.2510.20733

  55. [55]

    Vicinagearth , author =

    A survey on. Vicinagearth , author =. 2024 , keywords =. doi:10.1007/s44336-024-00009-2 , abstract =

    doi:10.1007/s44336-024-00009-2 2024

  56. [56]

    Blumenkamp, Jan and Morad, Steven and Gielis, Jennifer and Li, Qingbiao and Prorok, Amanda , month = may, year =. A. 2022. doi:10.1109/ICRA46639.2022.9811744 , abstract =

    arXiv 2022

  57. [57]

    and Mireshghallah, Niloofar and Ibrahim, Mark and Mahloujifar, Saeed , month = feb, year =

    Morris, John X. and Mireshghallah, Niloofar and Ibrahim, Mark and Mahloujifar, Saeed , month = feb, year =. Learning to. doi:10.48550/arXiv.2602.04118 , abstract =

    work page doi:10.48550/arxiv.2602.04118

  58. [58]

    Elements of

    Ben-Ari, Mordechai and Mondada, Francesco , year =. Elements of. doi:10.1007/978-3-319-62533-1 , language =

    work page doi:10.1007/978-3-319-62533-1

  59. [59]

    doi:10.48550/arXiv.2510.21450 , abstract =

    Danieli, Federico and Rodriguez, Pau and Sarabia, Miguel and Suau, Xavier and Zappella, Luca , month = nov, year =. doi:10.48550/arXiv.2510.21450 , abstract =

    work page doi:10.48550/arxiv.2510.21450

  60. [60]

    LAMARL: LLM-Aided Multi-Agent Reinforcement Learning for Cooperative Policy Generation

    Zhu, Guobin and Zhou, Rui and Ji, Wenkang and Zhao, Shiyu , month = jun, year =. doi:10.48550/arXiv.2506.01538 , abstract =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2506.01538

  61. [61]

    Robots in a human world: safety, reliability, robustness , abstract =

  62. [62]

    Li, Hao Xiang and Amir, Michael and Prorok, Amanda , month = nov, year =. Scaling. doi:10.48550/arXiv.2511.03100 , abstract =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2511.03100

  63. [63]

    IEEE Robotics and Automation Letters , author =

    Learning-. IEEE Robotics and Automation Letters , author =. 2023 , pages =. doi:10.1109/LRA.2023.3234809 , abstract =

    arXiv 2023

  64. [64]

    doi:10.48550/arXiv.2510.04898 , abstract =

    Xiong, Zheng and Li, Kang and Wang, Zilin and Jackson, Matthew and Foerster, Jakob and Whiteson, Shimon , month = oct, year =. doi:10.48550/arXiv.2510.04898 , abstract =

    work page doi:10.48550/arxiv.2510.04898

  65. [65]

    Wan, Weikang and Zhu, Yifeng and Shah, Rutav and Zhu, Yuke , month = may, year =. 2024. doi:10.1109/ICRA57147.2024.10611129 , abstract =

    arXiv 2024

  66. [66]

    Nature , author =

    Loss of plasticity in deep continual learning , volume =. Nature , author =. 2024 , pages =. doi:10.1038/s41586-024-07711-7 , abstract =

    doi:10.1038/s41586-024-07711-7 2024

  67. [67]

    Intelligence, Physical and Black, Kevin and Brown, Noah and Darpinian, James and Dhabalia, Karan and Driess, Danny and Esmail, Adnan and Equi, Michael and Finn, Chelsea and Fusai, Niccolo and Galliker, Manuel Y. and Ghosh, Dibya and Groom, Lachy and Hausman, Karol and Ichter, Brian and Jakubczak, Szymon and Jones, Tim and Ke, Liyiming and LeBlanc, Devin a...

    Pith/arXiv arXiv doi:10.48550/arxiv.2504.16054

  68. [68]

    Robotics and Autonomous Systems , author =

    Perceived safety in physical human–robot interaction—. Robotics and Autonomous Systems , author =. 2022 , pages =. doi:10.1016/j.robot.2022.104047 , abstract =

    arXiv 2022

  69. [69]

    Foundation models: toward real-robot and real-time capable systems , abstract =

  70. [70]

    doi:10.48550/arXiv.2503.17821 , abstract =

    Gessler, Tobias and Dizdarevic, Tin and Calinescu, Ani and Ellis, Benjamin and Lupu, Andrei and Foerster, Jakob Nicolaus , month = mar, year =. doi:10.48550/arXiv.2503.17821 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2503.17821

  71. [71]

    Yuan, Lei and Zhang, Ziqian and Li, Lihe and Guan, Cong and Yu, Yang , month = dec, year =. A. doi:10.48550/arXiv.2312.01058 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2312.01058

  72. [72]

    and Whiteson, Shimon , month = oct, year =

    Ellis, Benjamin and Cook, Jonathan and Moalla, Skander and Samvelyan, Mikayel and Sun, Mingfei and Mahajan, Anuj and Foerster, Jakob N. and Whiteson, Shimon , month = oct, year =. doi:10.48550/arXiv.2212.07489 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2212.07489

  73. [73]

    Yu, Chao and Velu, Akash and Vinitsky, Eugene and Gao, Jiaxuan and Wang, Yu and Bayen, Alexandre and Wu, Yi , month = nov, year =. The. doi:10.48550/arXiv.2103.01955 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2103.01955

  74. [74]

    and Kailkhura, Bhavya and Bhatele, Abhinav and Goldstein, Tom , month = feb, year =

    Geiping, Jonas and McLeish, Sean and Jain, Neel and Kirchenbauer, John and Singh, Siddharth and Bartoldson, Brian R. and Kailkhura, Bhavya and Bhatele, Abhinav and Goldstein, Tom , month = feb, year =. Scaling up. doi:10.48550/arXiv.2502.05171 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2502.05171

  75. [75]

    2025 , keywords =

    Enabling. 2025 , keywords =

    2025

  76. [77]

    Zou, Jiaru and Yang, Xiyuan and Qiu, Ruizhong and Li, Gaotang and Tieu, Katherine and Lu, Pan and Shen, Ke and Tong, Hanghang and Choi, Yejin and He, Jingrui and Zou, James and Wang, Mengdi and Yang, Ling , month = nov, year =. Latent. doi:10.48550/arXiv.2511.20639 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2511.20639

  77. [78]

    Entropy is all you need for

  78. [79]

    IEEE Transactions on Pattern Analysis and Machine Intelligence , author =

    A. IEEE Transactions on Pattern Analysis and Machine Intelligence , author =. 2024 , pages =. doi:10.1109/TPAMI.2024.3457538 , abstract =

    arXiv 2024

  79. [80]

    Language

    Radford, Alec and Wu, Jeffrey and Child, Rewon and Luan, David and Amodei, Dario and Sutskever, Ilya , keywords =. Language

  80. [81]

    Improving

    Radford, Alec and Narasimhan, Karthik and Salimans, Tim and Sutskever, Ilya , keywords =. Improving

Showing first 80 references.

This paper was first reviewed by deepseek-v4-flash on August 5, 2026.