Unsupervised Partner Design Enables Robust Ad-hoc Teamwork

Andreas Bulling; Anna Penzkofer; Constantin Ruhdorfer; Matteo Bortoletto; Victor Oei

REVIEW 3 major objections 5 minor 1 cited by

Unsupervised Partner Design Enables Robust Ad-hoc Teamwork

T0 review · 3 major / 5 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read Unsupervised Partner Design (UPD) claims that robust ad-hoc teamwork can be trained by generating partners on the fly and selecting them by return variance, removing pre-trained partner populations and per-task tuning, and that the same rul

desk verdict The empirical package is strong and the method is useful, but the headline mechanism—learnability-based selection—is not cleanly isolated, and the theory cited for it doesn't quite apply. read the letter →

arxiv 2508.06336 v2 pith:SR5SASPJ submitted 2025-08-08 cs.LG cs.AIcs.HCcs.MA

Constantin Ruhdorfer , Matteo Bortoletto , Victor Oei , Anna Penzkofer , Andreas Bulling This is my paper · ORCID

classification cs.LGcs.AIcs.HCcs.MA

keywords ad-hocteamworkmulti-agentreinforcementlearningunsupervisedenvironmentdesignlearnabilitycurriculumzero-shotcoordinationpartnergenerationOvercooked-AI

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

Unsupervised Partner Design (UPD) targets ad-hoc teamwork—cooperating with partners never encountered during training. The paper's central claim is that robustness to unseen partners does not require the standard expensive ingredients: a pre-trained population of diverse partner policies or hand-tuned mixing coefficients. Instead, UPD generates candidate partners on the fly, scores each by the variance of returns it produces with the current agent, and trains only against the most learnable ones, turning partner selection into an adaptive curriculum. The same scoring rule, renormalized across reward scales, also selects procedurally generated levels jointly with partners. If the claim holds, training for ad-hoc teamwork becomes nearly as cheap as self-play while matching or beating population-based methods in both artificial-partner and human evaluations.

What carries the argument

The load-bearing mechanism is the pair (generator, scorer). The partner generator samples a mixing coefficient $\epsilon \sim U(0,1)$ and, with probability 0.5, a persistent action-bias mask $m\sim\mathrm{Dirichlet}(\alpha \mathbf{1}_A)$, producing a partner policy $\pi_p = \epsilon \pi_{r,m} + (1-\epsilon)\pi_{\mathrm{ego}}$. The scorer then estimates $\ell_{\mathrm{var}} = \mathrm{Var}_{\tau\sim G_{\pi_p,\theta}}[R(\tau)]$ from rollout returns (or the mean-normalized coefficient-of-variation-squared variant in joint settings) and admits the top-$|B|$ candidates to a buffer refreshed every $R$ training loops. This turns the induced single-agent game $G_{\pi_p,\theta}$ into a training instan

What would settle it

Train two UPD variants whose scoring rollouts are matched for mean return but differ in return variance, and measure the ego policy's one-step PPO improvement; the paper's learnability justification predicts the high-variance partners produce larger improvement. At the method level, compare UPD with UPD without the learnability filter at matched compute on the layout where the filter matters most (Asymmetric Advantages, reported 181.4 vs 164.0): if filtering does not beat unfiltered generation, the central learnability claim is falsified.

Watch

Extended reading notes

Core claim

UPD's central claim is that a partner can be treated as a training instance to be generated and filtered, exactly as unsupervised environment design treats levels. The paper defines a stochastic partner generator that mixes the ego policy with a biased random policy, samples many such partners, scores each by the variance of episodic return when paired with the current ego, and keeps only the highest-variance candidates in a replay buffer for PPO updates. This learnability filter is the whole method: it replaces the pre-trained partner population of fictitious co-play and maximum-entropy population training and the fixed mixture coefficient of the E3T baseline with an adaptive curriculum tha

Load-bearing premise

The load-bearing premise is that the variance of returns across rollouts with a fixed partner tracks how much the ego policy is expected to improve by training with that partner; if that proxy fails, the learnability filter loses its justification.

Editorial extensions

If this is right

Ad-hoc teamwork training becomes single-stage and population-free: UPD needs only the ego policy plus a stochastic partner generator, removing the $O(N C)$ partner-pretraining cost; the paper's cost analysis gives a break-even population size of about 2.5 in the simulation-dominated limit, so any practical population of size 3 or more is more expensive.
One hyperparameter configuration transfers across all five Overcooked layouts, whereas E3T requires a per-layout sweep of its mixture coefficient; this removes task-specific tuning for population-free ad-hoc teamwork.
The same mechanism solves joint partner-and-level generalization: JUPD jointly selects partners and procedurally generated levels and outperforms random level-partner sampling, self-play cross-environment cooperation, and SFL-level selection combined with E3T partners on the Overcooked Generalisation Challenge.
Learnability-based selection produces emergent convention-breaking dynamics—partner action biases switch direction over training—so UPD implicitly avoids overfitting to a single coordination equilibrium without an explicit mechanism.
In a human-AI study, UPD-trained agents achieve higher returns and are rated significantly more adaptive, more human-like, and less frustrating than self-play, MEP, and E3T agents.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

Because UPD's partner space is defined by a stochastic generator, richer partner distributions (learned latent partner models or stored populations) could plug into the same learnability filter; the paper names this as future work, which suggests UPD is best read as a general curriculum layer rather than a single partner recipe.
The emergent bias switches imply a testable bridge to convention-breaking methods: UPD-trained agents should cooperate better with partners that share none of the ego's learned action conventions than self-play agents do, because learnability actively selects such partners.
The JUPD results leave open whether joint selection is additive: a natural ablation the paper does not run is levels chosen by SFL with partners chosen uniformly, or vice versa, which would isolate the gain that comes from selecting both dimensions at once.
The appendix's transparency note reports that an earlier version contained baseline configuration issues; after correction baselines improved but the paper states conclusions were unchanged, so the current comparative numbers are the ones to rely on.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

Quick take: this is a solid empirical paper on a simple, useful idea—generate partners online via random mixtures of ego and biased random policies, score them by return variance, and train against the high-variance ones. The method is population-free, works across LBF, Overcooked, and the OGC, and the human study is a genuine plus. The paper is worth a serious referee. But the advertised central mechanism—the learnability filter—is not cleanly isolated, and the theory cited for it doesn't quite apply.

What's new: applying SFL-style learnability to partner policies, the bias-mask generator, and the joint partner-environment extension (JUPD). Those are real increments over E3T and prior UED work. The empirical work is extensive: 282 trained policies, multiple seeds, ablations, a per-partner breakdown, and a careful double-blind human study with proper multiple-comparison corrections. The transparency note about fixing earlier baseline bugs is a point in their favor.

Soft spots. First, the ablations. UPD w/o ℓ removes not just scoring but the whole buffer/refresh machinery and samples random partners per rollout. So the 3.6-point average gap between UPD and UPD w/o ℓ cannot be attributed specifically to learnability selection; it could come from the buffer infrastructure. The paper itself acknowledges that large-scale generation alone is a strong baseline, and includes helpful analysis in the appendix showing that the convention-breaking dynamics do require the learnability filter. That partially rescues the mechanism, but the headline claim "selects them adaptively based on a learnability criterion" overstates what the ablation establishes.

Second, the theory. Section 4.1 invokes Foster et al. (2026) that expected policy improvement is proportional to variance of the scalar learning signal. That's about advantage variance, not episodic return variance in a non-stationary induced game. The paper says "applying the result ... implies" but that transfer is not proven. Minor in practice—the empirical results stand—but the bridge is rhetorical, not formal.

Third, the tuning-free claim. They search over UPD hyperparameters (buffer size, N, refresh frequency) in Appendix E.1 and report a fine-tuned version that does better. So "no manual parameter tuning" should be softened to "one shared configuration across layouts, found with a modest search."

Who this is for: anyone working on zero-shot coordination or ad-hoc teamwork, especially with human-AI evaluation. The practical recipe is cheap and reproducible.

Recommendation: send it to review. The empirical core is sound and the method is useful; the learnability mechanism needs a cleaner ablation or a substantially toned-down claim. A referee can sort that out, but the paper should not be desk-rejected.

Referee Report

3 major / 5 minor

Summary. The paper introduces Unsupervised Partner Design (UPD), a population-free method for training ad-hoc teamwork agents. UPD generates candidate partner policies on the fly by mixing the ego policy with a biased random policy using a stochastic mixing coefficient, scores candidates by the variance of episodic returns (a learnability signal), maintains a buffer of high-scoring partners, and trains the ego policy with PPO. The same mechanism is extended to joint partner-environment curricula (JUPD) using a coefficient-of-variation-squared score. The method is evaluated on Level-Based Foraging, Overcooked-AI, the Overcooked Generalisation Challenge, and in a human-AI study, with 282 trained policies overall. The central claims are that UPD achieves robust ad-hoc teamwork without pre-trained partner populations or manual parameter tuning, and that the adaptive learnability-based selection is the mechanism responsible for the gains.

Significance. If the learnability-based partner-selection mechanism is indeed responsible for the reported gains, this is a valuable contribution: it extends unsupervised environment design to the partner-policy space, removes the need for pre-trained partner populations, and shows a simple adaptive curriculum can compete with or outperform population-based methods across multiple benchmarks. The empirical effort is substantial and is a clear strength: multiple seeds, diverse evaluation partners, an additional ROTATE comparison, per-partner results, curriculum-dynamics analyses, alternative learnability functions, a human study with corrected multiple-comparison statistics, and a transparency note about earlier baseline issues. The paper also honestly discusses the possibility that large-scale randomized partner generation alone may be the active ingredient. However, the evidence isolating the specific contribution of the learnability criterion is incomplete, and the tuning-related claims are broader than what the experiments support. The significance is therefore conditional on resolving these attribution issues.

major comments (3)

[§5.2.2, Table 1] The only ablation targeting learnability is confounded. The text states that UPD w/o ℓ "removes learnability scoring and instead samples random partners per rollout," which eliminates not only scoring but also the buffer-admission (top-|B|) and refresh (R) steps of Algorithm 1. The observed gap (UPD 94.4±2.3 vs. UPD w/o ℓ 90.8±6.0) can therefore be attributed to the buffer/refresh infrastructure or to the interaction of ϵ~U(0,1) and bias masking, not specifically to the variance signal. This matters because the abstract's central claim is adaptive selection "based on a learnability criterion." The paper needs an ablation that keeps the buffer/refresh pipeline but selects partners randomly, and/or an ablation that keeps learnability scoring but removes the buffer, to isolate the mechanism. Without this, the data are fully consistent with the alternative—raised by the authors themselves in
[§4.1, Eq. (5)] The formal connection between return variance and expected policy improvement is not established. The paragraph invokes Foster et al. (2026), whose result concerns the variance of the scalar learning signal used in advantage estimation, and then equates this with Varτ[R(τ)] in the induced single-agent game Gπp,θ. Episodic return variance is not the same as the variance of the per-step advantage estimator, and the induced game is non-stationary because πego changes during training even between buffer refreshes. As written, Eq. (5) is a heuristic. Since the learnability criterion is the advertised conceptual novelty, the paper should either state the precise transfer assumptions or provide a direct derivation for this setting. This issue is compounded by the confounded ablation in Table 1, so the empirical evidence for the criterion is currently weak.
[Abstract, §5.2.1, Appendix E] The claim that UPD removes "manual parameter tuning" is overstated. UPD introduces several curriculum hyperparameters (|B|, N, R, ρ, α, pbias, δ, K) and the paper uses different values across tasks: LBF uses |B|=64, N=5; Overcooked-AI uses |B|=512, N=10; OGC uses buffer sizes 4,096/16,384, ρ=1.0, and R=2. Appendix E, Table 7 describes a hyperparameter search whose choices were made using evaluation with a BRDiv population as a proxy, and Appendix D shows that per-layout fine-tuning of curriculum parameters improves UPD from 94.4 to 102.3 average return. The claim would be defensible if narrowed to "removes per-layout tuning of the E3T mixture coefficient ϵ in Overcooked-AI" or if accompanied by sensitivity evidence showing that default hyperparameters are robust across tasks. As written, the abstract overstates the elimination of manual tuning.

minor comments (5)

[§5.2.2, Table 1] Rename "UPD w/o ℓ" to something like "UPD w/o ℓ (random partners, no buffer)" to make clear that this ablation removes both learnability scoring and the buffer/refresh infrastructure.
[§6, Figure 15] Table 2 uses "DR-DR" but Figure 15 uses "DR"; align the naming.
[§4.2, Eq. (6)] The stability constant δ is used in the main text but defined only in Appendix E. Add a one-line definition in the main text.
[§5.3, Figure 7] The significance stars are informative, but reporting exact p-values and effect sizes (e.g., median/mean differences with confidence intervals) for the human study would improve transparency, especially given the small sample size (n=12).
[Algorithm 1] Line 5 says "for each desired partner" without specifying the number of candidates; the number K (the SFL batch size) is only given in Appendix E. State K directly in the algorithm or its caption.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: UPD's learnability score is derived online from generated partners, evaluation is on held-out partners and humans, and self-citations are to external benchmarks/artifacts, not to the method's validity.

full rationale

UPD's claimed derivation chain is not circular. The learnability score ℓvar (Eq. 5) is computed online from rollouts pairing the current ego with randomly generated partners (Alg. 1) and is never fitted to evaluation partners or test returns; evaluation uses held-out BRDiv/planning/hardcoded agents and a human study, so no fitted input is relabeled as a prediction. The §4.1 justification that return variance tracks expected policy improvement is an external citation (Foster et al. 2026) applied to the induced single-agent game; whether that transfer is valid is a correctness/threat-to-validity issue, not a circular reduction, because the paper does not define 'expected improvement' in terms of ℓvar. The §4.4 matrix-game example is explicitly illustrative and only shows that Bernoulli variance peaks at p=0.5, a mathematical property of variance, not a derivation of AHT success. Self-citations (Ruhdorfer et al. 2025a,b) are to an unrelated environment and to the OGC benchmark, which are external artifacts rather than load-bearing premises of the method. The paper's own limitations (Sec. 7, App. C) acknowledge that UPD w/o ℓ already performs strongly and that large-scale generation alone may induce useful curricula; this weakens attribution of gains to the learnability filter, but it is an empirical/attribution concern, not a case of the result being equivalent to its inputs. The transparency note in App. B discloses corrected baselines but states conclusions were unchanged; again, this is an integrity disclosure, not a circularity. Therefore no circular step can be exhibited from the paper's equations or citations.

Assumptions & free parameters 8 free parameters · 4 assumptions · 0 invented entities

The central claim depends on several hand-set hyperparameters (buffer size, N, R, rho, p_bias, alpha, K) and on the theoretical assumption that return variance is a valid learnability signal. The partner generator is a design choice, not an invented entity. No new objects (particles, forces) are introduced.

free parameters (8)

bias probability p_bias = 0.5
Probability of sampling a biased random policy in the partner generator (Alg. 2); chosen by hand.
Dirichlet concentration alpha = 1.0
Controls strength of action biases in partner generation (Sec. 4.3, Alg. 2).
buffer size |B| = 512 (Overcooked), 64 (LBF), 4096/16384 (OGC)
Number of partners admitted to the learnability buffer (Alg. 1); layout specific and tuned in Table 7.
scoring rollouts N = 10 (Overcooked), 5 (LBF/OGC)
Number of rollouts per candidate partner to estimate return variance (Alg. 1).
buffer refresh frequency R = 4 (Overcooked/LBF), 2 (OGC)
How often the buffer is replaced (Alg. 1).
SFL sample ratio rho = 0.5 (Overcooked), 1.0 (OGC)
Proportion of partners drawn from buffer vs newly generated (Alg. 1; Appendix E).
CV^2 stability constant delta = 1e-8
Added to denominator in Eq. 6 to avoid division by zero.
generated partners K (SFL batch) = 4000 (Overcooked), 8192 (OGC)
Candidates sampled per refresh; cost scales with this (Appendix E).

assumptions (4)

domain assumption Foster et al. (2026): expected policy improvement is proportional to variance of the scalar learning signal, and this transfers to return variance in the induced single-agent game (Sec. 4.1).
The learnability score's theoretical justification rests on this external result; the mapping from advantage variance to episodic return variance is not proven in this paper.
domain assumption The partner generator Sp (mixture of ego and biased random policy) spans a diverse enough partner space for training AHT agents, including human-like partners.
Inherited from E3T (Yan et al., 2023); if the generated partners do not cover the evaluation distribution, the curriculum cannot generalize.
domain assumption The evaluation populations (BRDiv, planning, hardcoded, humans) are representative of the unknown Pi_eval in ad-hoc teamwork.
Standard AHT benchmarking protocol, following Wang et al. (2025). The paper does not justify this beyond prior usage.
standard math PPO with the reported hyperparameters converges stably in all settings.
Training curves in Appendix H support this, but convergence is an empirical fact, not proven.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Unsupervised Partner Design Enables Robust Ad-hoc Teamwork." pith.science (2026). https://pith.science/paper/SR5SASPJ

@misc{pith2026250806336,
  author       = {Pith},
  title        = {Pith review of: Unsupervised Partner Design Enables Robust Ad-hoc Teamwork},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/SR5SASPJ}},
  note         = {Machine review of arXiv:2508.06336}
}

read the original abstract

We introduce Unsupervised Partner Design (UPD), a population-free multi-agent reinforcement learning method for robust ad-hoc teamwork. UPD generates training partners on-the-fly and selects them adaptively based on a learnability criterion, removing the need for pre-trained partner populations or manual parameter tuning. We show that this simple mechanism enables effective partner diversity and can be extended to joint partner-environment selection when a procedural level generator is available. Across Level-Based Foraging, Overcooked-AI, and the Overcooked Generalisation Challenge, UPD consistently achieves strong performance compared to both population-based and population-free baselines. In a human-AI user study, agents trained with UPD achieve higher returns and are rated as more adaptive, more human-like, and less frustrating than all evaluated baseline methods.

Figures

Figures reproduced from arXiv: 2508.06336 by the authors.

**Figure 1.** Unsupervised partner design is a novel population-free, multi-agent learning framework for ad-hoc teamwork that uses learnability to find training partners for the ego agent to generate an open-ended curriculum. populations becomes increasingly expensive as tasks and partner diversity scale. Efficient end-to-end training (E3T) (Yan et al., 2023) partially addresses this challenge by generating training partners as … view at source ↗

**Figure 2.** Conceptual illustration: We illustrate that as ego competence improves over training (black), E3T generates partners using a fixed mixture coefficient (here ϵ = 0.5), resulting in a fixed fraction of ego competence (blue). In contrast, UPD samples ϵ ∼ U(0, 1) (green) and filters partners using a learnability criterion, leading to a dynamic range of partner competences (orange). 3.3. Unsupervised Environment Design… view at source ↗

**Figure 3.** Average returns in cooperative LBF with ten evaluation partners. Bars show mean ± standard deviation. UPD achieves higher average returns than E3T across all tested ϵ. 5. Experiments in Fixed Environments We split our experiments into two sections. In this section, we evaluate UPD in environments without procedural generation where θ is fixed, using Level-Based Foraging and Overcooked-AI. Our goal here is to assess … view at source ↗

Figures from the paper (18 more)

**Figure 4.** Figure 4: We compare how ℓvar selects different average ϵ values across layouts during training. For layouts that are known to feature narrower coordination challenges (CC, CR & FC), UPD favours smaller, while for CRoom and AA, UPD favours higher ϵ. We train 6 seeds for 107 time…

**Figure 5.** Figure 5: We plot learnability vs. return in three representative layouts: CR, AA, and FC. Each dot represents a single potential partner. The bar plots on the axis count partners in their respective return/learnability bands. UPD identifies partners of intermediate difficulty. …

**Figure 6.** Figure 6: Action biases added in partner generation over training. We find that UPD induces emergent convention breaking, with partners initially biased toward one action before switching. zero. This is consistent with the motivation in Sec. 4.1: most partners are not equally be…

**Figure 7.** Figure 7: Human evaluation of partners: Frust = ‘frustrating?’ (↓), Adapt = ‘adapted well?’ (↑), Human = ‘human-like?’ (↑), Coord = ‘coordinated well?’ (↑). We performed one-sided Wilcoxon signed-rank tests on individual survey questions and one-sided paired t-tests on return, c…

**Figure 9.** Figure 9: We redraw the five evaluation layouts introduced by Carroll et al. (2019) using the JaxMARL visualisation pipeline (Rutherford et al., 2023). From left to right: CRoom, AA, CR, CC, and FC. G.2. Overcooked-AI Overcooked-AI is a multi-agent coordination benchmark based o…

**Figure 10.** Figure 10: SP training curves. We average over 6 seeds and show standard deviation. 0 1 2 3 4 5 Environment Step 1e7 0 50 100 150 200 Average Episode Return FCP Training Returns Over Time by Layout (Mean ± Std) Layout AA CR CC CRoom FC [PITH_FULL_IMAGE:figures/full_fig_p022_10.png]

**Figure 11.** Figure 11: FCP training curves. We average over 6 seeds and show standard deviation. 0 1 2 3 4 5 Environment Step 1e7 0 50 100 150 200 Average Episode Return MEP Training Returns Over Time by Layout (Mean ± Std) Layout AA CR CC CRoom FC [PITH_FULL_IMAGE:figures/full_fig_p022_11.png]

**Figure 12.** Figure 12: MEP training curves. We average over 6 seeds and show standard deviation. 0 1 2 3 4 5 Environment Step 1e7 0 50 100 150 200 Average Episode Return E3T Training Returns Over Time by Layout (Mean ± Std) Layout AA CR CC CRoom FC [PITH_FULL_IMAGE:figures/full_fig_p022_12.png]

**Figure 13.** Figure 13: E3T training curves. We average over 6 seeds and show standard deviation. 22 [PITH_FULL_IMAGE:figures/full_fig_p022_13.png]

**Figure 14.** Figure 14: UPD training curves. We average over 6 seeds and show standard deviation. 0.0 0.2 0.4 0.6 0.8 1.0 Environment Step 1e9 0 5 10 15 20 25 Average Episode Return OGC Mean ± Std Training Curves Method DR CEC SFLE3T JUPD [PITH_FULL_IMAGE:figures/full_fig_p023_14.png]

**Figure 15.** Figure 15: OGC training curves. We average over 6 seeds and show standard deviation for all methods. (a) Early (b) Middle (c) Final [PITH_FULL_IMAGE:figures/full_fig_p023_15.png]

**Figure 16.** Figure 16: Learnability vs. return over training in Coordination Ring. (a) Early (b) Middle (c) Final [PITH_FULL_IMAGE:figures/full_fig_p023_16.png]

**Figure 17.** Figure 17: Learnability vs. return over training in Asymmetric Advantages. 23 [PITH_FULL_IMAGE:figures/full_fig_p023_17.png]

**Figure 18.** Figure 18: Learnability vs. return in Forced Coordination. 50 60 70 80 90 100 Average Returns mean adaptive-SR gauss var [PITH_FULL_IMAGE:figures/full_fig_p024_18.png]

**Figure 19.** Figure 19: Performance of multiple learnability functions against the evaluation population averaged over layouts. Here ℓmean selects partners based on their mean return, ℓgauss weights partners down that are far from the average global return (Monette et al., 2025) and adaptive…

**Figure 20.** Figure 20: Per-Partner results averaged over layouts in Overcooked-AI. UPD outperforms other methods with most partners. Some methods show similar or even slightly better performance with the Onion and the Plate agent. However, these methods are not consistently better than UPD.…

**Figure 21.** Figure 21: We show the same analysis as in [PITH_FULL_IMAGE:figures/full_fig_p026_21.png]

**Figure 22.** Figure 22: Distribution of human ratings for each survey question across all agents. Each bar represents the number of responses given to each Likert item (x-axis), with colors indicating the agent. Questions are grouped vertically and include both subjective impressions (e.g., …

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

Dynamic Forcing Behind Rapid Intensification of Hurricane Lidia
physics.ao-ph 2025-08 unverdicted novelty 4.0 of 10

Based on the abstract, the paper claims that stronger upper-level Trenberth forcing preceded Hurricane Lidia's rapid intensification in the higher-intensification IFS-ECMWF ensemble members.

Reference graph

Works this paper leans on

13 extracted references · 10 canonical work pages · cited by 1 Pith paper

[1]

I enjoyed playing with the agent

work page
[2]

I felt that the agent’s ability to coordinate with me was: (very poor, poor, neutral, good, very good)

work page
[3]

The agent adapted to me when making decisions

work page
[4]

The agent frequently got in my way.(negative)

work page
[5]

Yan, X., Guo, J., Lou, X., Wang, J., Zhang, H., and Du, Y

URL http://proceedings.mlr.press/ v119/wang20l.html. Yan, X., Guo, J., Lou, X., Wang, J., Zhang, H., and Du, Y . An efficient end-to-end training approach for zero-shot human-ai coordination. In Oh, A., Naumann, T., Globerson, A., Saenko, K., Hardt, M., and Levine, S. (eds.),Advances in Neural Information Processing Systems 36: Annual Conference on Neural...

work page 2023
[6]

Zhao, R., Song, J., Yuan, Y ., Hu, H., Gao, Y ., Wu, Y ., Sun, Z., and Yang, W

URL https://openreview.net/forum ?id=TrwE8l9aJzs. Zhao, R., Song, J., Yuan, Y ., Hu, H., Gao, Y ., Wu, Y ., Sun, Z., and Yang, W. Maximum entropy population- based training for zero-shot human-ai coordination. In Williams, B., Chen, Y ., and Neville, J. (eds.),Thirty- Seventh AAAI Conference on Artificial Intelligence, AAAI 2023, Thirty-Fifth Conference o...

work page 2023
[7]

Responses were numerically mapped from 1 to 5

The agent’s behaviour was frustrating.(negative) We additionally allowed users to give free-form feedback at the end of the study. Responses were numerically mapped from 1 to 5. Negative-valence questions were inverted before aggregation for our analysis on the overall subjective preference. To assess internal consistency of the question responses, we com...

work page
[11]

The agent was consistent in its actions

work page

Show all 13 references

[12]

The agent’s actions were human-like
[2020]

URL https://doi.org/10.1016/j.artint.2019

doi: 10.1016/J.ARTINT.2019.103216. URL https://doi.org/10.1016/j.artint.2019. 103216. Barrett, S., Agmon, N., Hazon, N., Kraus, S., and Stone, P. Communicating with unknown teammates. In Bazzan, A. L. C., Huhns, M. N., Lomuscio, A., and Scerri, P. (eds.),International conferen...

2019
[2023]

URL https: //dl.acm.org/doi/10.5555/3545946.359 8700

doi: 10.5555/3545946.3598700. URL https: //dl.acm.org/doi/10.5555/3545946.359 8700. Mirsky, R., Carlucho, I., Rahman, A., Fosong, E., Macke, W., Sridharan, M., Stone, P., and Albrecht, S. V . A sur- vey of ad hoc teamwork research. In Baumeister, D. and Rothe, J. (eds.),Multi-...

2022
[2025]

URL https: //doi.org/10.48550/arXiv.2504.15457

doi: 10.48550/ARXIV.2504.15457. URL https: //doi.org/10.48550/arXiv.2504.15457. Cho, K., van Merri ¨enboer, B., Bahdanau, D., and Ben- gio, Y . On the properties of neural machine transla- tion: Encoder–decoder approaches. In Wu, D., Carpuat, M., Carreras, X., and Vecchi, E. M...

2014 doi
[4410]

URL http://proceedings

PMLR, 2020. URL http://proceedings. mlr.press/v119/hu20a.html. Hunter, J. D. Matplotlib: A 2d graphics environment.Com- puting in Science & Engineering, 9(3):90–95, 2007. doi: 10.1109/MCSE.2007.55. Jha, K., Carvalho, W., Liang, Y ., Du, S. S., Kleiman-Weiner, M., and Jaques, N...

2020 arXiv

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.

[1] [1]

I enjoyed playing with the agent

work page

[2] [2]

I felt that the agent’s ability to coordinate with me was: (very poor, poor, neutral, good, very good)

work page

[3] [3]

The agent adapted to me when making decisions

work page

[4] [4]

The agent frequently got in my way.(negative)

work page

[5] [5]

Yan, X., Guo, J., Lou, X., Wang, J., Zhang, H., and Du, Y

URL http://proceedings.mlr.press/ v119/wang20l.html. Yan, X., Guo, J., Lou, X., Wang, J., Zhang, H., and Du, Y . An efficient end-to-end training approach for zero-shot human-ai coordination. In Oh, A., Naumann, T., Globerson, A., Saenko, K., Hardt, M., and Levine, S. (eds.),Advances in Neural Information Processing Systems 36: Annual Conference on Neural...

work page 2023

[6] [6]

Zhao, R., Song, J., Yuan, Y ., Hu, H., Gao, Y ., Wu, Y ., Sun, Z., and Yang, W

URL https://openreview.net/forum ?id=TrwE8l9aJzs. Zhao, R., Song, J., Yuan, Y ., Hu, H., Gao, Y ., Wu, Y ., Sun, Z., and Yang, W. Maximum entropy population- based training for zero-shot human-ai coordination. In Williams, B., Chen, Y ., and Neville, J. (eds.),Thirty- Seventh AAAI Conference on Artificial Intelligence, AAAI 2023, Thirty-Fifth Conference o...

work page 2023

[7] [7]

Responses were numerically mapped from 1 to 5

The agent’s behaviour was frustrating.(negative) We additionally allowed users to give free-form feedback at the end of the study. Responses were numerically mapped from 1 to 5. Negative-valence questions were inverted before aggregation for our analysis on the overall subjective preference. To assess internal consistency of the question responses, we com...

work page

[8] [11]

The agent was consistent in its actions

work page