Pith. sign in

REVIEW 4 major objections 6 minor 61 references

Hidden partner abilities can be inferred from a few joint tasks and reused to plan safer teamwork across new tasks.

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 ↗

T0 review · grok-4.5

2026-07-30 10:53 UTC pith:UH7YPRP6

load-bearing objection Solid multi-task AHT methods paper: online capability inference without population training works in sim, diversity helps offline human matching, but closed-loop teaming gains are not shown. the 4 major comments →

arxiv 2607.27177 v1 pith:UH7YPRP6 submitted 2026-07-29 cs.AI cs.HCcs.MA

Partner Capability Estimation for Task-Agnostic Adaptation in Ad-Hoc Teamwork

Peter Tisnikar , Maja Swieczkowska , Benteng Ma , Gerard Canal , Matteo Leonetti This is my paper
classification cs.AI cs.HCcs.MA
keywords ad-hoc teamworkhuman–agent interactionbehaviour modellingcapability estimationcontextual multi-agent MDPapproximate Bayesian computationdiverse planning
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.

Most ad-hoc teamwork methods train a single ego policy for one fixed task and treat the partner as a black box. This paper instead treats repeated collaboration as joint planning under hidden, task-invariant partner capabilities—what the partner can and cannot do. CE-CM samples candidate capability vectors, simulates the joint trajectories each would produce, and keeps those that match what was observed, building an approximate posterior without population pre-training. The resulting estimate induces a capability-conditioned multi-agent model so the agent can plan the next task while executing only its own actions. When humans choose among many valid strategies, CE-CM-Div scores each hypothesis against diverse planner rollouts rather than one optimal path. In a household planning domain and in Overcooked, the method recovers capabilities, cuts infeasible assignments, and adapts when abilities change; on 225 human trajectories it needs diversity-aware matching to get reliable estimates.

Core claim

Partner capabilities—latent binary constraints on which joint transitions are feasible—are a reusable, task-agnostic representation that an ego agent can recover online from a handful of observed joint trajectories via approximate Bayesian simulate-and-compare, then use to induce contextual multi-agent MDPs for decentralised joint planning; when behaviour is diverse or suboptimal, matching against sets of diverse rollouts (CE-CM-Div) is necessary for accurate inference, especially with humans.

What carries the argument

CE-CM (Capability Estimation via Contextual Models): Approximate Bayesian Computation over discrete capability vectors—sample candidates, joint-plan under each, accept those whose simulated trajectories fall within a distance threshold of the observation, aggregate into a Bernoulli belief, and plan on the induced capability-conditioned CMMDP; CE-CM-Div replaces the single rollout with a diverse trajectory set per hypothesis.

Load-bearing premise

Observed joint trajectories are assumed to already be feasible under the partner’s true abilities because the partner can correct the ego agent and enforce the aligned plan, and the mapping from capability bits to which transitions they gate is known in advance.

What would settle it

If, with the same known capability-to-transition structure and correction channel disabled or noisy, CE-CM and CE-CM-Div fail to drive Hamming distance toward zero over a few tasks on held-out partners, or CE-CM-Div no longer beats single-trajectory CE-CM on human Overcooked trajectories, the central claim fails.

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

If this is right

  • Agents can accumulate an explicit, interpretable model of one partner across chores or layouts without retraining a population policy.
  • Feasibility-aware planning reduces assignments of actions the partner cannot execute, improving safety even when full behavioural prediction remains ambiguous.
  • When multiple strategies are equally valid, diversity-aware likelihoods are required for capability inference from human data.
  • Capability estimates can be updated online if the partner’s abilities change mid-collaboration.
  • Capability models alone do not resolve preference or convention ambiguity; joint capability-plus-preference models become the natural next representation.

Where Pith is reading between the lines

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

  • The same simulate-and-compare loop could be tried for continuous or graded ability parameters if the discrete bit vector is too coarse for physical skills.
  • Without a known capability-to-transition map, the method would need to jointly discover structure and values—closer to structure learning than pure ABC filtering.
  • Scaling beyond two agents will force factorised or role-based planning, or the joint planner cost will dominate any inference gain.
  • Correction-based alignment during data collection may understate how hard inference is when both agents act without a referee.

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

4 major / 6 minor

Summary. The paper reframes multi-task ad-hoc teamwork as joint planning with decentralised execution under hidden, task-invariant partner capabilities in a contextual MMDP. It introduces CE-CM, an ABC-style approximate Bayesian method that samples capability vectors, simulates joint trajectories under a planner, and updates an independent-Bernoulli belief from accepted samples; the MAP estimate then induces a capability-conditioned model for planning the next task. CE-CM-Div extends the likelihood comparison to a set of diverse planner rollouts to handle suboptimal or multi-modal partner behaviour. Simulated results in TidyUP and Overcooked show rapid recovery of ground-truth capability vectors, fewer infeasible partner assignments, and adaptation when capabilities change. An offline study of 225 Overcooked trajectories from 15 humans shows CE-CM-Div substantially lowers Hamming error relative to single-trajectory CE-CM.

Significance. If the results hold, the work offers a clear, interpretable alternative to population-trained AHT policies: a reusable partner model that transfers across tasks without pre-training on partner populations. The capability-as-transition-constraint formulation, the online ABC loop, and the diversity extension are concrete methodological contributions, and the paper ships public code plus a human trajectory dataset. The TidyUP results and the feasibility/safety gains in Overcooked are convincing evidence that capability estimates can improve action allocation when behaviour is largely feasibility-driven. The human study usefully documents that planner-optimal trajectories poorly match people, motivating diversity-aware inference. These strengths make the paper a solid contribution to human–AI teaming and multi-task AHT, provided the claims about closed-loop human teaming are aligned with the evidence.

major comments (4)
  1. [Abstract; §5.3; §7] Abstract and §5.3/§7 frame CE-CM-Div as essential for robust human–AI teaming, but the human evaluation is purely offline Hamming-distance recovery on recorded trajectories (225 episodes, fixed assistant policy). There is no closed-loop experiment in which CE-CM-Div estimates are used to replan with the same participants and measure corrections, plan overlap, unproductive actions, or task success. §5.2 already shows that better capability estimates need not reduce corrections under strategy underspecification (Types 1–2). The human result therefore supports only improved estimation under behavioural diversity, not the stronger teaming claim. Either add a closed-loop human evaluation or revise the abstract/conclusion to match the intermediate claim actually tested.
  2. [§3.3] §3.3 assumes observed trajectories are feasible under the partner’s true CMMDP because Ag2 can correct Ag1 and enforce the aligned joint plan. This alignment channel is load-bearing for the ABC likelihood: without it, ego misallocation or simultaneous capability errors would corrupt τ_obs and the simulate-and-compare update would not identify c. The paper notes the assumption but does not stress-test it (e.g., noisy/partial corrections, delayed corrections, or no correction channel). A sensitivity experiment or a clearly scoped limitation stating that identification holds only under aligned execution is needed for the central inference claim.
  3. [§3.2; Appendix B.2] The method assumes the capability-to-transition gating structure is known a priori (which bits enable which transitions; §3.2 and the f pruning function in App. B.2). Combined with a hand-specified discrete capability vocabulary, this weakens the “task-agnostic / no pre-coordination” framing relative to methods that learn latent partner structure from data. The paper should state explicitly what must be known in advance versus what is inferred online, and discuss how misspecified gates would affect posterior recovery.
  4. [§5.1.1; §5.2.1] Baselines are limited to optimistic (and, in TidyUP, pessimistic) non-adaptive models. There is no comparison to type-based AHT, latent-conditioned policies, behaviour cloning of the partner, or preference-learning approaches discussed in §2. Without at least one adaptive partner-modelling baseline on the same multi-task protocol, it is hard to judge whether capability vectors are competitive as a representation rather than merely better than assuming full capability. Adding one such baseline in simulation would substantially strengthen H1–H2.
minor comments (6)
  1. [§3.3; §4.1] Eq. (1) minimises a sum of trajectory losses over G_seen, but the implemented estimator is an ABC posterior mean/threshold on independent Bernoullis (§4.1). Briefly state that the MAP of the ABC approximation is treated as a surrogate for (1).
  2. [§4.1] Independent Bernoulli factorisation of the posterior (§4.1) ignores correlations among capabilities (e.g., room-linked pick/place bits in TidyUP). A short note on when this approximation fails would help.
  3. [Appendix B] Figure 1 and Algorithm 1 are clear; however, the acceptance threshold ε, prior P(c_i=1)=0.8, ψ=0.5, and δ=0.35 are scattered across Appendix B. A small hyperparameter table in the main text (or early appendix pointer) would aid reproducibility.
  4. [Appendix B.2] In Overcooked, cosine similarity with ε=0.03 on flattened states is quite tight; a brief justification or sensitivity check (analogous to App. C.3 for δ) would be useful.
  5. [§1; Appendix A] Typos/consistency: “CApability Modelling” in the CAMO paragraph; duplicate “Transitions are deterministic…” block in App. A.2; “place_study / pick_study / move_study” descriptions say “kitchen” in Table A.1.
  6. [§2] Related work on ZSC/AHT is solid; a short pointer to recent assistance games / theory-of-mind planning work beyond the cited goal-recognition papers would round out §2.2.

Circularity Check

0 steps flagged

No significant circularity: empirical ABC inference scored against held-out ground truth and independent coordination metrics.

full rationale

CE-CM is an approximate Bayesian (ABC) simulate-and-compare procedure that samples candidate capability vectors, rolls out joint plans, and accepts those whose trajectories fall within a distance threshold of observations; the MAP estimate is then used to induce a CMMDP for planning. Capability recovery is measured by Hamming distance to externally fixed ground-truth vectors, and coordination impact by corrections, plan IoU, and unproductive/feasible action rates against optimistic/pessimistic baselines—none of these targets is defined by, or fitted to equal, the inference objective. The self-citation to CAMO [5] only reuses the ABC template; the multi-task AHT setting, discrete task-invariant capabilities, decentralised execution, CE-CM-Div diversity extension, and human offline evaluation are independently specified and scored. There is no self-definitional loop, no fitted parameter renamed as a prediction, no load-bearing uniqueness theorem from the authors, and no renaming of a known empirical law. The paper is self-contained against its stated benchmarks.

Axiom & Free-Parameter Ledger

5 free parameters · 6 axioms · 2 invented entities

The central results rest on a contextual multi-agent MDP with binary capability-gated deterministic transitions, known shared goals/rewards, a known capability vocabulary and gating structure, ABC approximate inference with hand-set similarity thresholds, and the alignment assumption that observed trajectories reflect the partner’s true feasible joint plan. Capability bits are invented modeling entities with semantic labels chosen by the designers; free parameters are algorithmic thresholds and planner hyperparameters rather than physics constants.

free parameters (5)
  • ABC acceptance threshold ε = 0.24 (TidyUP); 0.03 (Overcooked)
    Controls which simulated trajectories count as matching observations; set per domain (0.24 Jaccard in TidyUP, 0.03 cosine in Overcooked) and directly gates posterior mass.
  • MAP threshold ψ on Bernoulli belief = 0.5
    Converts continuous belief coordinates into binary capability estimates used for planning.
  • Sample budget N and Bernoulli prior P(c_i=1) = N=500; prior 0.8 in TidyUP
    Determines coverage of the hypothesis space; prior bias (0.8 in TidyUP; full 2^10 enumeration/sampling in Overcooked) affects recovery speed.
  • CE-CM-Div diversity threshold δ and max trajectories k = δ=0.35; k=50
    Hand-chosen diversity filter that changes how many rollouts support each hypothesis; appendix shows non-monotonic accuracy vs δ.
  • MCTS/Q-learning planner hyperparameters = e.g. 200 sims, depth 20, γ=0.99; Q-learning 70M steps
    Simulations, rollout depth, cp, γ, training steps, etc., shape the ‘optimal’ or diverse trajectories that define the ABC likelihood.
axioms (6)
  • domain assumption Partner capabilities are static binary vectors that gate deterministic transitions via known enabling subsets; missing capability yields a self-loop.
    Section 3.1–3.2 defines CMMDP and capability-as-transition semantics used throughout inference and planning.
  • domain assumption Goals and the goal-conditioned reward/preferences are shared and known to both agents.
    Stated in Section 3; without known g and R, trajectory comparison cannot attribute mismatch to capabilities alone.
  • domain assumption Observed trajectories are aligned to the partner’s true capability model because the partner can correct conflicting ego actions.
    Section 3.3 alignment assumption; mirrored in the human UI correction protocol.
  • standard math Approximate Bayesian Computation with trajectory distance L and tolerance ε is a valid surrogate for the intractable planning likelihood P(τ|c).
    Section 4.1 invokes ABC; standard methodology but approximate, with error controlled by ε and planner fidelity.
  • ad hoc to paper Posterior factors as independent Bernoullis across capability bits for point estimation.
    Section 4.1 parametric belief update; ignores bit correlations that may exist in true capability profiles.
  • domain assumption The ego agent knows the capability vocabulary and which transitions each bit enables (structure known; values hidden).
    Implicit in C and T(·|c) design; pessimistic baseline even uses explicit transition–capability mapping in PDDL.
invented entities (2)
  • Task-invariant discrete partner capability vector c in a contextual MMDP no independent evidence
    purpose: Portable latent partner model that restricts feasible joint plans across tasks and supports planning under decentralized execution.
    Core representation introduced for multi-task AHT; semantics are designer-chosen per domain (20 bits TidyUP, 10 Overcooked).
  • CE-CM / CE-CM-Div inference procedures independent evidence
    purpose: Online ABC simulate-and-compare (and diverse-rollout variant) to estimate c and induce planning models without population pre-training.
    Methodological constructs evaluated empirically; not ontological claims beyond the capability model.

pith-pipeline@v1.2.0-daily-grok45 · 30015 in / 3801 out tokens · 69325 ms · 2026-07-30T10:53:53.220956+00:00 · methodology

0 comments
read the original abstract

Effective collaboration with novel and diverse partners is a crucial skill for autonomous agents. Most current ad-hoc teamwork (AHT) approaches assume that agents will collaborate on a single, fixed task and that the partner's capabilities, their ability to successfully execute the desired action, are already known. In reality, a partner's true capabilities are often hidden, and human collaborators may act sub-optimally on tasks with multiple valid strategies. To address these limitations, we extend ad-hoc teamwork into a multi-task setting by re-framing it as a problem of joint planning with decentralised execution under hidden partner capabilities. We introduce CE-CM (Capability Estimation via Contextual Models), an approximate Bayesian method that infers task-invariant capability vectors. By using simulation-based sampling, the agent estimates capabilities and induces a contextual Multi-agent Markov Decision Processes for planning. This approach requires no population pre-training and refines its beliefs online from just a few tasks. To account for human unpredictability, we propose CE-CM-Div, an extension that evaluates capability hypotheses against diverse planner rollouts rather than a single optimal trajectory. Simulated experiments demonstrate that CE-CM rapidly recovers hidden capabilities, reduces infeasible action assignments, and adapts to changes over time. Furthermore, in an offline human study of 225 trajectories from 15 participants, CE-CM-Div substantially improved capability estimates over the baseline CE-CM method. Our results suggest capability-based modelling is a promising interpretable, task-agnostic representation in the studied settings, demonstrating that accounting for behavioural diversity is essential for robust human-AI teaming.

Figures

Figures reproduced from arXiv: 2607.27177 by Benteng Ma, Gerard Canal, Maja Swieczkowska, Matteo Leonetti, Peter Tisnikar.

Figure 1
Figure 1. Figure 1: Overview of CE-CM. After completing a task with a partner, the agent uses the observed [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: TidyUP domain. Left: one of the possible starting states, where the blanket and a dirty bowl are on the couch, the dirty dish is in the study, and the toothbrush is in the kitchen. Right: possible goal state where the blanket is on the bed, the bowl and the dish are clean and in the kitchen, and the toothbrush is in the bathroom. We first evaluate CE-CM in TidyUP, a PDDL-based household domain in which a h… view at source ↗
Figure 3
Figure 3. Figure 3: Hamming distance between ground truth and inferred capability vectors. Solid line [PITH_FULL_IMAGE:figures/full_fig_p018_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Ratio of corrected actions across tasks. Solid line indicates the mean, and shaded area [PITH_FULL_IMAGE:figures/full_fig_p018_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Plan overlap (IoU) between joint plans. Solid line indicates the mean, and shaded area [PITH_FULL_IMAGE:figures/full_fig_p019_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Overview of Overcooked. There are three distinct layouts, and three possible recipes, creating nine distinct tasks in total. The agents must collaborate to deliver the dish as quickly as possible, ensuring that the dish consists of the correct ingredients. it to plan the next task. To evaluate open-loop generalisation, we assess performance on the remaining 4 held-out tasks. We conduct two main evaluations… view at source ↗
Figure 7
Figure 7. Figure 7: Hamming distance between ground truth capabilities and the estimated capability vector. [PITH_FULL_IMAGE:figures/full_fig_p021_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Ratio of corrected actions across tasks. Solid line indicates the mean, and shaded area [PITH_FULL_IMAGE:figures/full_fig_p021_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Ratio of unproductive (infeasible) actions across tasks. Solid line indicates the mean, and [PITH_FULL_IMAGE:figures/full_fig_p022_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Proportion of feasible actions in open-loop planning on held-out tasks. Solid line indicates [PITH_FULL_IMAGE:figures/full_fig_p022_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Overview of the Overcooked user interface. In the top left corner, the intelligent assistant agent indicates its next action. The available commands, the current recipe, and the restricted capabilities are displayed on the right hand side. resolve coordination (H2). This highlights a key limitation: capabilities alone are insufficient to predict behaviour when multiple strategies are equally valid. 5.3. C… view at source ↗
Figure 12
Figure 12. Figure 12: Number of accepted samples in the posterior across observed tasks. Solid line indicates [PITH_FULL_IMAGE:figures/full_fig_p025_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Hamming distance between ground truth and inferred capability vectors. Solid line [PITH_FULL_IMAGE:figures/full_fig_p025_13.png] view at source ↗

Review history (2 revisions) →

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

61 extracted references · 14 canonical work pages · 1 internal anchor

  1. [1]

    M. D. Zhao, R. G. Simmons, H. Admoni, The role of adaptation in collective human–ai teaming, Topics in Cognitive Science 17 (2022) 291 – 323

    2022

  2. [2]

    Natarajan, E

    M. Natarajan, E. Seraj, B. Altundas, R. Paleja, S. Ye, L. Chen, R. Jensen, K. C. Chang, M. Gombolay, Human-Robot Teaming: Grand Challenges, Current Robotics Reports 4 (3) (2023) 81–100.doi:10.1007/s43154-023-00103-1

    work page doi:10.1007/s43154-023-00103-1 2023

  3. [3]

    Mirsky, I

    R. Mirsky, I. Carlucho, A. Rahman, E. Fosong, W. Macke, M. Sridha- ran, P. Stone, S. V. Albrecht, A Survey of Ad Hoc Teamwork Research, arXiv:2202.10450 [cs.MA] (2022).arXiv:2202.10450. URLhttp://arxiv.org/abs/2202.10450

    Pith/arXiv arXiv 2022

  4. [5]

    Tisnikar, G

    P. Tisnikar, G. Canal, M. Leonetti, Probabilistic inference of human capabili- ties from passive observations, in: 2024 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), IEEE, 2024, pp. 8779–8785. 38

    2024

  5. [6]

    Other-play

    H. Hu, A. Lerer, A. Peysakhovich, J. Foerster, “Other-play” for zero-shot coor- dination, in: H. D. III, A. Singh (Eds.), Proceedings of the 37th International Conference on Machine Learning, Vol. 119 of Proceedings of Machine Learning Research, PMLR, 2020, pp. 4399–4410

    2020

  6. [7]

    A. Shih, A. Sawhney, J. Kondic, S. Ermon, D. Sadigh, On the Critical Role of Conventions in Adaptive Human-AI Collaboration, ICLR 2021 (2021).arXiv: 2104.02871. URLhttp://arxiv.org/abs/2104.02871

    Pith/arXiv arXiv 2021

  7. [8]

    Heinrich, D

    J. Heinrich, D. Silver, Deep reinforcement learning from self-play in imperfect- information games, arXiv preprint arXiv:1603.01121 (2016)

    Pith/arXiv arXiv 2016

  8. [9]

    Strouse, K

    D. Strouse, K. R. McKee, M. Botvinick, E. Hughes, R. Everett, Collaborating with humans without human data, in: Proceedings of the 35th International Conference on Neural Information Processing Systems, NIPS ’21, Curran Asso- ciates Inc., Red Hook, NY, USA, 2021, pp. 14502–14515

    2021

  9. [10]

    A. Lupu, B. Cui, H. Hu, J. Foerster, Trajectory diversity for zero-shot coordi- nation, in: M. Meila, T. Zhang (Eds.), Proceedings of the 38th International Conference on Machine Learning, Vol. 139 of Proceedings of Machine Learning Research, PMLR, 2021, pp. 7204–7213

    2021

  10. [11]

    A. Szot, U. Jain, D. Batra, Z. Kira, R. Desai, A. Rai, Adaptive coordination in social embodied rearrangement, in: Proceedings of the 40th International Conference on Machine Learning, ICML’23, JMLR.org, 2023, pp. 33365–33380

    2023

  11. [12]

    D. Xing, Q. Liu, Q. Zheng, G. Pan, Learning with Generated Teammates to Achieve Type-Free Ad-Hoc Teamwork, in: Proceedings of the Thirtieth International Joint Conference on Artificial Intelligence, International Joint Conferences on Artificial Intelligence Organization, 2021, pp. 472–478.doi: 10.24963/ijcai.2021/66

    work page doi:10.24963/ijcai.2021/66 2021

  12. [13]

    Sarkar, A

    B. Sarkar, A. Shih, D. Sadigh, Diverse conventions for human-ai collaboration, in: Proceedings of the 37th International Conference on Neural Information Processing Systems, NIPS ’23, Curran Associates Inc., Red Hook, NY, USA, 2023, pp. 23115–23139

    2023

  13. [14]

    C. Yu, J. Gao, W. Liu, B. Xu, H. Tang, J. Yang, Y. Wang, Y. Wu, Learn- ing zero-shot cooperation with humans, assuming humans are biased, in: The 39 Eleventh International Conference on Learning Representations, ICLR 2023, Ki- gali, Rwanda, May 1-5, 2023, OpenReview.net, 2023, pp. 1–10. URLhttps://openreview.net/forum?id=TrwE8l9aJzs

    2023

  14. [15]

    X. Lou, J. Guo, J. Zhang, J. Wang, K. Huang, Y. Du, Pecan: Leveraging policy ensemble for context-aware zero-shot human-ai coordination, in: Proceedings of the 2023 International Conference on Autonomous Agents and Multiagent Systems, AAMAS ’23, International Foundation for Autonomous Agents and Multiagent Systems, Richland, SC, 2023, p. 679–688

    2023

  15. [17]

    O. C. Görür, B. Rosman, S. Albayrak, Anticipatory bayesian policy selection for online adaptation of collaborative robots to unknown human types, in: Pro- ceedings of the 18th International Conference on Autonomous Agents and Mul- tiAgent Systems, 2019, pp. 77–85

    2019

  16. [18]

    Barrett, P

    S. Barrett, P. Stone, Cooperating with Unknown Teammates in Complex Do- mains: A Robot Soccer Case Study of Ad Hoc Teamwork, Proceedings of the AAAI Conference on Artificial Intelligence 29 (1) (2015).doi:10.1609/aaai. v29i1.9428

    doi:10.1609/aaai 2015

  17. [19]

    H.Li, T.Ni, S.Agrawal, F.Jia, S.Raja, Y.Gui, D.Hughes, M.Lewis, K.Sycara, Individualized Mutual Adaptation in Human-Agent Teams, IEEE Transactions on Human-Machine Systems 51 (6) (2021) 706–714.doi:10.1109/THMS.2021. 3107675

    doi:10.1109/thms.2021 2021

  18. [20]

    Ghosh, S

    A. Ghosh, S. Tschiatschek, H. Mahdavi, A. Singla, Towards Deployment of Ro- bust Cooperative AI Agents: An Algorithmic Framework for Learning Adaptive Policies, New Zealand (2020)

    2020

  19. [21]

    C. Wang, C. Pérez-D’Arpino, D. Xu, L. Fei-Fei, K. Liu, S. Savarese, Co-gail: Learning diverse strategies for human-robot collaboration, in: Conference on Robot Learning, PMLR, 2022, pp. 1279–1290

    2022

  20. [22]

    S. V. Albrecht, P. Stone, Autonomous agents modelling other agents: A com- prehensive survey and open problems, Artificial Intelligence 258 (2018) 66–95. doi:10.1016/j.artint.2018.01.002. 40

    doi:10.1016/j.artint.2018.01.002 2018

  21. [23]

    Carroll, R

    M. Carroll, R. Shah, M. K. Ho, T. Griffiths, S. Seshia, P. Abbeel, A. Dragan, On the utility of learning about humans for human-ai coordination, Advances in neural information processing systems 32 (2019)

    2019

  22. [24]

    L. M. Hiatt, C. Narber, E. Bekele, S. S. Khemlani, J. G. Trafton, Human modeling for human–robot collaboration, The International Journal of Robotics Research 36 (5-7) (2017) 580–596.doi:10.1177/0278364917690592

    work page doi:10.1177/0278364917690592 2017

  23. [25]

    Raileanu, E

    R. Raileanu, E. L. Denton, A. Szlam, R. Fergus, Modeling others using oneself in multi-agent reinforcement learning, in: International Conference on Machine Learning, 2018, pp. 4257–4266. URLhttps://api.semanticscholar.org/CorpusID:3622509

    2018

  24. [26]

    S. V. Albrecht, J. W. Crandall, S. Ramamoorthy, Belief and truth in hypoth- esised behaviours, Artificial Intelligence 235 (2016) 63–94.doi:10.1016/j. artint.2016.02.004

    doi:10.1016/j 2016

  25. [27]

    Barrett, A

    S. Barrett, A. Rosenfeld, S. Kraus, P. Stone, Making friends on the fly: Cooperating with new teammates, Artificial Intelligence 242 (2017) 132–171. doi:10.1016/j.artint.2016.10.005

    work page doi:10.1016/j.artint.2016.10.005 2017

  26. [28]

    S. V. Albrecht, P. Stone, Reasoning about Hypothetical Agent Behaviours and theirParameters, in: Proceedingsofthe16thConferenceonAutonomousAgents andMultiAgentSystems, 2019, p.547–555.arXiv:1906.11064,doi:10.48550/ arXiv.1906.11064

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1906.11064 2019

  27. [29]

    Shafipour Yourdshahi, M

    E. Shafipour Yourdshahi, M. A. Do Carmo Alves, A. Varma, L. Soriano Mar- colino, J. Ueyama, P. Angelov, On-line estimators for ad-hoc task execu- tion: Learning types and parameters of teammates for effective teamwork, Au- tonomous Agents and Multi-Agent Systems 36 (2) (2022) 45.doi:10.1007/ s10458-022-09571-9

    2022

  28. [30]

    Czechowski, F

    A. Czechowski, F. A. Oliehoek, Decentralized MCTS via Learned Teammate Models, in: Proceedings of the Twenty-Ninth International Joint Conference on Artificial Intelligence, International Joint Conferences on Artificial Intelligence Organization, 2020, pp. 81–88.doi:10.24963/ijcai.2020/12

    work page doi:10.24963/ijcai.2020/12 2020

  29. [31]

    F. Wu, S. Zilberstein, X. Chen, Online planning for ad hoc autonomous agent teams, in: Proceedings of the Twenty-Second International Joint Conference on Artificial Intelligence - Volume One, IJCAI’11, AAAI Press, 2011, p. 439–445. 41

    2011

  30. [32]

    Felip, D

    J. Felip, D. Gonzalez-Aguirre, L. Nachman, Intuitive & efficient human-robot collaboration via real-time approximate bayesian inference, in: 2022 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2022, pp. 3093–3099.doi:10.1109/IROS47612.2022.9982251

    arXiv 2022

  31. [33]

    J. G. Ribeiro, M. Faria, A. Sardinha, F. S. Melo, Helping People on the Fly: Ad Hoc Teamwork for Human-Robot Teams, in: G. Marreiros, F. S. Melo, N. Lau, H. Lopes Cardoso, L. P. Reis (Eds.), Progress in Artificial Intelligence, Vol. 12981, Springer International Publishing, 2021, pp. 635–647.doi:10.1007/ 978-3-030-86230-5_50

    2021

  32. [34]

    J. G. Ribeiro, L. M. Henriques, S. Colcher, J. C. Duarte, F. S. Melo, R. L. Mi- lidiú, A. Sardinha, HOTSPOT: An ad hoc teamwork platform for mixed human- robot teams, PLOS ONE 19 (6) (2024) e0305705.doi:10.1371/journal.pone. 0305705

    doi:10.1371/journal.pone 2024

  33. [35]

    Huang, B

    C.-M. Huang, B. Mutlu, Anticipatory robot control for efficient human-robot collaboration, in: 2016 11th ACM/IEEE International Conference on Human- Robot Interaction (HRI), IEEE, 2016, pp. 83–90.doi:10.1109/HRI.2016. 7451737

    work page doi:10.1109/hri.2016 2016

  34. [36]

    Laidlaw, E

    C. Laidlaw, E. Bronstein, T. Guo, D. Feng, L. Berglund, J. Svegliato, S. Russell, A. Dragan, Assistancezero: Scalably solving assistance games, in: Forty-second International Conference on Machine Learning, 2025, pp. 1–10. URLhttps://openreview.net/forum?id=b9hVMJi0t2

    2025

  35. [37]

    Nikolaidis, R

    S. Nikolaidis, R. Ramakrishnan, K. Gu, J. Shah, Efficient Model Learning from Joint-Action Demonstrations for Human-Robot Collaborative Tasks, in: Pro- ceedings of the Tenth Annual ACM/IEEE International Conference on Human- Robot Interaction, HRI ’15, Association for Computing Machinery, 2015, pp. 189–196.doi:10.1145/2696454.2696455

    arXiv 2015

  36. [38]

    Bestick, R

    A. Bestick, R. Pandya, R. Bajcsy, A. D. Dragan, Learning Human Ergonomic Preferences for Handovers, in: 2018 IEEE International Conference on Robotics and Automation (ICRA), IEEE, 2018, pp. 3257–3264.doi:10.1109/ICRA. 2018.8461216

    arXiv 2018

  37. [39]

    M. D. Zhao, R. Simmons, H. Admoni, Learning human contribution preferences in collaborative human-robot tasks, in: J. Tan, M. Toussaint, K. Darvish (Eds.), Proceedings of The 7th Conference on Robot Learning, Vol. 229 of Proceedings of Machine Learning Research, PMLR, 2023, pp. 3597–3618. 42

    2023

  38. [40]

    Trivedi, P

    M. Trivedi, P. Doshi, Inverse Learning of Robot Behavior for Collaborative Planning, in: 2018 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2018, pp. 1–9.doi:10.1109/IROS.2018.8593745

    arXiv 2018

  39. [41]

    Nikolaidis, J

    S. Nikolaidis, J. Shah, Human-robot cross-training: Computational formula- tion, modeling and evaluation of a human team training strategy, in: 2013 8th ACM/IEEE International Conference on Human-Robot Interaction (HRI), 2013, pp. 33–40.doi:10.1109/HRI.2013.6483499

    arXiv 2013

  40. [42]

    Nikolaidis, P

    S. Nikolaidis, P. Lasota, R. Ramakrishnan, J. Shah, Improved human–robot team performance through cross-training, an approach inspired by human team training practices, The International Journal of Robotics Research 34 (14) (2015) 1711–1730.doi:10.1177/0278364915609673

    work page doi:10.1177/0278364915609673 2015

  41. [43]

    Narcomey, N

    A. Narcomey, N. Tsoi, R. Desai, M. Vázquez, Learning Human Preferences Over Robot Behavior as Soft Planning Constraints, arXiv:2403.19795 [cs.RO] (2024). arXiv:2403.19795. URLhttp://arxiv.org/abs/2403.19795

    Pith/arXiv arXiv 2024

  42. [44]

    Canal, G

    G. Canal, G. Alenyà, C. Torras, Adapting robot task planning to user prefer- ences: An assistive shoe dressing example, Autonomous Robots 43 (6) (2019) 1343–1356.doi:10.1007/s10514-018-9737-2

    work page doi:10.1007/s10514-018-9737-2 2019

  43. [45]

    T. A. Nguyen, M. Do, A. E. Gerevini, I. Serina, B. Srivastava, S. Kambhampati, Generating diverse plans to handle unknown and partially known user prefer- ences, Artificial Intelligence 190 (2012) 1–31.doi:10.1016/j.artint.2012. 05.005

    work page doi:10.1016/j.artint.2012 2012

  44. [46]

    Y. Emam, S. Mayya, G. Notomista, A. Bohannon, M. Egerstedt, Adaptive Task Allocation for Heterogeneous Multi-Robot Teams with Evolving and Unknown Robot Capabilities, in: 2020 IEEE International Conference on Robotics and Automation (ICRA), IEEE, 2020, pp. 7719–7725.doi:10.1109/ICRA40945. 2020.9197283

    arXiv 2020

  45. [47]

    doi:10.1038/s41598-022-19140-5

    A.Ali, H.Azevedo-Sa, D.M.Tilbury, L.P.Robert, Heterogeneoushuman–robot task allocation based on artificial trust, Scientific Reports 12 (1) (2022) 15304. doi:10.1038/s41598-022-19140-5

    work page doi:10.1038/s41598-022-19140-5 2022

  46. [48]

    R. Liu, M. Natarajan, M. C. Gombolay, Coordinating Human-Robot Teams with Dynamic and Stochastic Task Proficiencies, ACM Transactions on Human- Robot Interaction 11 (1) (2022) 1–42.doi:10.1145/3477391. 43

    work page doi:10.1145/3477391 2022

  47. [49]

    Izquierdo-Badiola, G

    S. Izquierdo-Badiola, G. Canal, C. Rizzo, G. Alenyà, Improved task planning through failure anticipation in human-robot collaboration, in: 2022 Interna- tional Conference on Robotics and Automation (ICRA), 2022, pp. 7875–7880

    2022

  48. [50]

    Zhang, Y

    S. Zhang, Y. Chen, J. Zhang, Y. Jia, Real-Time Adaptive Assembly Scheduling in Human-Multi-Robot Collaboration According to Human Capability, in: 2020 IEEE International Conference on Robotics and Automation (ICRA), IEEE, 2020, pp. 3860–3866.doi:10.1109/ICRA40945.2020.9196618

    arXiv 2020

  49. [51]

    B. Fu, W. Smith, D. M. Rizzo, M. Castanier, M. Ghaffari, K. Barton, Robust Task Scheduling for Heterogeneous Robot Teams Under Capability Uncertainty, IEEE Transactions on Robotics 39 (2) (2023) 1087–1105.doi:10.1109/TRO. 2022.3216068

    arXiv 2023

  50. [52]

    Orlov-Savko, A

    L. Orlov-Savko, A. Jain, G. M. Gremillion, C. E. Neubauer, J. D. Canady, V. Unhelkar, Factorial agent markov model: Modeling other agents’ behavior in presence of dynamic latent decision factors, in: Proceedings of the 21st Inter- national Conference on Autonomous Agents and Multiagent Systems, AAMAS ’22, International Foundation for Autonomous Agents and...

    2022

  51. [53]

    Unhelkar, S

    V. Unhelkar, S. Li, J. A. Shah, Semi-supervised learning of decision-making models for human-robot collaboration, in: Conference on Robot Learning, 2019, pp. 192–203. URLhttps://api.semanticscholar.org/CorpusID:208175027

    2019

  52. [54]

    Hallak, D

    A. Hallak, D. Di Castro, S. Mannor, Contextual Markov Decision Processes, arXiv:1502.02259 [stat.ML] (2015).arXiv:1502.02259. URLhttp://arxiv.org/abs/1502.02259

    Pith/arXiv arXiv 2015

  53. [55]

    D. V. Pynadath, M. Tambe, The communicative multiagent team decision prob- lem: Analyzing teamwork theories and models, Journal of artificial intelligence research 16 (2002) 389–423

    2002

  54. [56]

    M. A. Beaumont, Approximate bayesian computation, Annual review of statis- tics and its application 6 (1) (2019) 379–403

    2019

  55. [57]

    Benke, T

    L. Benke, T. Miller, M. Papasimeon, N. Lipovetzky, Diverse, Top-k, and Top-Quality Planning Over Simulators, in: K. Gal, A. Nowé, G. J. Nalepa, R. Fairstein, R. Rădulescu (Eds.), Frontiers in Artificial Intelligence and Appli- cations, IOS Press, 2023, pp. 1–10.doi:10.3233/FAIA230275. 44

    work page doi:10.3233/faia230275 2023

  56. [58]

    Srivastava, T

    B. Srivastava, T. A. Nguyen, A. Gerevini, S. Kambhampati, M. B. Do, I. Serina, Domain independent approaches for finding diverse plans, in: Proceedings of the 20th International Joint Conference on Artifical Intelligence, IJCAI’07, Morgan Kaufmann Publishers Inc., San Francisco, CA, USA, 2007, p. 2016–2022

    2007

  57. [59]

    M. Fox, D. Long, Pddl2. 1: An extension to pddl for expressing temporal plan- ning domains, Journal of artificial intelligence research 20 (2003) 61–124

    2003

  58. [60]

    S. A. Wu, R. E. Wang, J. A. Evans, J. B. Tenenbaum, D. C. Parkes, M. Kleiman- Weiner, Too Many Cooks: Bayesian Inference for Coordinating Multi-Agent Collaboration, Topics in Cognitive Science 13 (2) (2021) 414–432.doi:10. 1111/tops.12525

    2021

  59. [61]

    Y. Xiao, W. Tan, J. Hoffman, T. Xia, C. Amato, Asynchronous multi-agent deep reinforcement learning under partial observability, The International Journal of Robotics Research 44 (8) (2025) 1257–1286.doi:10.1177/02783649241306124

    work page doi:10.1177/02783649241306124 2025

  60. [62]

    Coles, A

    A. Coles, A. Coles, M. Fox, D. Long, Forward-chaining partial-order planning, in: Proceedings of the International Conference on Automated Planning and Scheduling, Vol. 20, 2010, pp. 42–49

    2010

  61. [63]

    Silver, J

    D. Silver, J. Schrittwieser, K. Simonyan, I. Antonoglou, A. Huang, A. Guez, T. Hubert, L. baker, M. Lai, A. Bolton, Y. Chen, T. P. Lillicrap, F. Hui, L. Sifre, G.vandenDriessche, T.Graepel, D.Hassabis, Masteringthegameofgowithout human knowledge, Nature 550 (2017) 354–359. URLhttps://api.semanticscholar.org/CorpusID:205261034 45

    2017