Pith. sign in

REVIEW 2 major objections 6 minor 47 references

Optimal Interactive Learning on the Job via Facility Location Planning

T0 review · 2 major / 6 minor · reviewed 2026-08-16 · deepseek-v4-flash

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

Pith's one-line read COIL recasts multi-task interactive robot learning as an uncapacitated facility location problem, making query selection a polynomial-time, bounded-suboptimal planning task that cut measured human cost by 12–20% in simulation and 23% on a…

desk verdict New and useful UFL formulation for multi-task interactive learning, but the central equivalence between UFL solutions and executable plans is not proven and can be violated; fixable, deserves review. read the letter →

arxiv 2505.00490 v1 pith:ARH5MEDQ submitted 2025-05-01 cs.RO cs.AI

Shivam Vats , Michelle Zhao , Patrick Callaghan , Mingxi Jia , Maxim Likhachev , Oliver Kroemer , George Konidaris This is my paper · ORCID
classification cs.ROcs.AI
keywords interactiverobotlearningfacilitylocationmulti-taskplanningpreferencequeriesskillhuman-robotcollaborationapproximationalgorithmsbelief-space
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

This paper argues that multi-task interactive robot learning—deciding, for each task in a known sequence, whether to ask the user to teach a skill, answer a preference question, or do the task by hand—can be recast exactly as an uncapacitated facility location (UFL) problem. In that recasting, each task is a demand, each possible query is a facility with an opening cost equal to the human's effort, and serving a demand costs the robot's execution cost minus the predicted reward of the skill. Because UFL has polynomial-time approximation algorithms, COIL can produce bounded-suboptimal interaction plans that in the paper's experiments cut total human burden by 12–20% in simulation and 23% in a physical conveyor task, while keeping task success intact. A one-step belief-space extension decides when a preference query is worth asking before committing to a plan. The appeal is practical: if the reduction holds, robots can plan long sequences of on-the-job learning without exponential search.

What carries the argument

The load-bearing object is the uncapacitated facility location instance $\langle D,F,f_i,c_{ij}\rangle$ built from the task sequence. Demands are the tasks; facilities are the three query/execution actions—human help, skill teaching, and zero-cost use of an already-learned skill—each with opening cost equal to its human effort and service costs computed from a similarity-based safety model $\rho^{\mathrm{safe}}_\pi(\tau',\theta')$ and the current preference belief $b_\Theta$. The UFL objective $\min \sum_{i\in F'} f_i + \sum_{j\in D} \min_{i\in F'} c_{ij}$ is what carries the argument: it converts the combinatorial look-ahead over query types into a problem with known polynomial-time approximation algorithms, and the same facility-location solver is reused as a subroutine inside the one-step belief-space planner for preference requests.

What would settle it

Run COIL on a two-task sequence $\tau_1,\tau_2$ with identical objects, set $c_{\mathrm{hum}}$ high and $c_{\mathrm{skill}}$ low, and inspect the returned plan; if the solver opens the skill facility at $\tau_2$ and serves $\tau_1$ from it without also serving $\tau_2$ by that facility, the plan cannot be executed by the robot's protocol, and the claimed cost guarantee would not transfer. This check directly tests whether the UFL solution set coincides with the executable policy set.

Watch

Extended reading notes

Core claim

COIL's central claim is that optimal interaction can be planned globally rather than myopically: over a fixed sequence of tasks, the choice between learning a skill, requesting a preference, asking the human to do the task, or executing an existing skill is a facility location decision. A skill facility opened at task $i$ costs $c_{\mathrm{skill}}$ and can serve any later task $j$ at service cost $c_{ij}^{\mathrm{skill}}=c_{\mathrm{rob}}-\max_{\theta\in\Theta}\hat{R}_j(\pi_i,\theta)$; a human facility costs $c_{\mathrm{hum}}$ and serves only its own task; a robot facility costs zero and serves any task the already-learned skill can reach. Minimizing total opening plus service cost over all tasks yields the interaction plan, and the paper proves that using a standard primal-dual UFL approximation keeps the runtime polynomial in the number of tasks and preference parameters while guaranteeing bounded suboptimality. When preferences are uncertain, COIL evaluates the expected value of asking the current task's preference by comparing the current plan against the expected cost of replanning under each possible response, and executes the preference request only when the expected improvement outweighs $c_{\mathrm{pref}}$. The paper reports the resulting cost reductions and shows the planner reallocates tasks to the human when a skill repeatedly fails to be learned.

Load-bearing premise

The planning formulation assumes that whenever the robot pays to learn a skill for a task, it is also assigned to perform that same task; if the optimal facility-location solution instead learns a skill and only uses it on later, similar tasks, the resulting plan may not be something the robot can actually execute.

Editorial extensions

If this is right

  • For a known task sequence, COIL can plan skill, preference, and human-help queries in polynomial time, so planning stays fast as the horizon grows; the paper reports large speedups over an optimal mixed-integer formulation with near-identical plan cost.
  • The planner will favor learning skills for objects that repeat and delegate singleton or hard-to-learn tasks to the human, which is the behavior that produces the measured 12–20% (simulation) and 23% (physical) reductions in interaction cost.
  • Online replanning with an updated teaching-success probability $\lambda_{\mathrm{teach}}$ lets the robot stop requesting demonstrations for skills that keep failing and instead assign those tasks to the human.
  • Preference requests are made only when the expected future plan improvement outweighs the user's answer cost, so the user is not asked about preferences for tasks the robot will not perform.
  • The same planner transfers to a physical conveyor sorting task, indicating the cost model and UFL solver behave outside simulation.

Reading between the lines

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

  • The UFL mapping suggests a natural extension to streaming or unordered task sets: because facility location solvers are incremental, one could re-solve or update the plan as new tasks arrive, avoiding the paper's stated need to know the full sequence in advance.
  • If service costs were learned rather than hand-built from task similarity, the same UFL machinery would apply; a testable extension would replace $\rho^{\mathrm{safe}}$ with a learned generalization model and measure whether the bounded-suboptimality guarantee still holds in practice.
  • The equivalence between UFL solutions and executable plans would be tightened by enforcing that a skill facility opened at task $i$ also serves task $i$; until that constraint is explicit, the claimed optimality guarantee applies to the UFL problem, not necessarily to the robot's one-action-per-task execution.
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.

Referee Report

2 major / 6 minor

Summary. The paper proposes COIL, a multi-task interactive learning planner that chooses among skill requests, preference requests, and human help for each task in a known sequence, with the goal of minimizing total human burden. The central idea is to formulate the known-preferences subproblem as an uncapacitated facility location (UFL) problem, where tasks are demands and human, skill, and robot actions are facilities; bounded-suboptimal plans are then obtained in polynomial time via an off-the-shelf approximation algorithm. For preference uncertainty, the planner performs one-step belief-space planning that uses the UFL planner as a subroutine. The paper reports experiments in a Gridworld simulator, a simulated 7-DoF manipulation domain, and a physical conveyor task, claiming 12%--20% cost reductions in simulation and 23% in the physical experiment relative to the best baseline, while maintaining task success.

Significance. If the UFL equivalence is correct, the paper makes a valuable theoretical contribution by connecting multi-task interactive learning to a well-studied operations-research problem, enabling polynomial-time bounded-suboptimal planning and the use of off-the-shelf solvers. The empirical evaluation is comparatively extensive: 30 randomized task sequences, three human cost profiles, several challenging-to-learn settings, online replanning after teaching failures, and a physical robot demonstration with statistical comparisons. These experiments are a useful benchmark among planners under a common cost model. However, the central formal claim is currently not supported: as detailed in the major comments, the UFL formulation admits solutions that cannot be executed under the paper's own interaction protocol, so the optimality and approximation guarantees have been proven only for a relaxed problem. The paper would also benefit from clarifying which approximation guarantee actually applies to the constructed instances.

major comments (2)
  1. [Section IV-A, Algorithm 1] The UFL formulation does not enforce that an opened skill facility serves its own origin task. Under the interaction protocol, exactly one action is executed per task and skill pi_i can be acquired only at task tau_i; therefore a UFL solution that opens i_skill, assigns tau_i to the human facility, and assigns later tasks to i_skill is not executable. As a concrete failure, with c_hum=100, c_skill=10, c_11=200, and c_12=10, the UFL optimum opens skill facility 1 (cost 10), serves task 1 by the human (100), and serves task 2 by skill facility 1 (10), giving total 120, while every executable plan costs at least 200. The manuscript neither states nor verifies the required constraint y_i_skill => x_{i,i}=1, so the claimed cost-optimality and bounded-suboptimality guarantees apply only to a relaxation of the actual interaction problem. This gap is load-bearing for the paper's central contribution and should be fixed by proving that every UFL solution decodes into an executable plan, or by reformulating the optimization with the self-assignment constraint and re-deriving the guarantees.
  2. [Section IV-A, Approximation Algorithms] The paper invokes the 3-approximation guarantee of Jain and Vazirani without establishing that the constructed UFL instance is metric. The service-cost matrix contains infinite entries (human facilities cannot serve non-origin tasks) and facility-dependent offsets derived from a learned reward model, so the instance is not obviously metric. For non-metric UFL, the stated 3-approximation does not follow; the paper mentions a log(n) bound 'otherwise' but does not identify which bound is being used for the experimental instances. The authors should either prove the metric property of the constructed cost structure or explicitly state and rely on the appropriate worst-case approximation bound.
minor comments (6)
  1. [Algorithm 1, line 22] The condition 'if cpref + ¯JPb1 ≤ J then' is not a well-formed expression as printed; it should be rewritten with explicit parentheses or a definition of the averaging operator so that the preference-request decision is unambiguous.
  2. [Equation (2) and Section III] The notation 'ci_query = ci_hum + ci_skill + ci_pref' mixes superscripts and subscripts inconsistently; please clarify whether these are task indices or type labels, and make the equation typographically consistent.
  3. [Appendix E-D and Figure 3] Several reported statistics use 'F = −3.54' where a t-statistic is intended, and p-values are inconsistently reported as p<0.01 and p<0.001 in the text versus the figures; these should be corrected and made uniform.
  4. [Figure 4] The axes are unclear: the caption mentions 'Ratio', 'Suboptimality', and 'Speedup', but the figure does not clearly indicate which curve uses which scale. Please add axis labels and a legend.
  5. [Section V-A and Appendix D-D] The baselines C-ADL and IG use hyperparameters (confidence threshold alpha=0.8 and information-gain scale beta=0.01) that appear to be chosen on the same evaluation domains; please report sensitivity analyses or explain why these choices do not bias the comparison.
  6. [Table III and Section VI-D] The physical experiments use only five task sequences, with one teaching-failure run reported separately; the claim of 'significantly lower cost' should be supported by an appropriate statistical test or explicitly described as a descriptive comparison.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; the UFL derivation is self-contained and the cost benchmark is not a constructed prediction.

full rationale

COIL's central claim is the reduction of multi-task interaction planning to uncapacitated facility location: tasks are demands; human, skill, and robot facilities with opening costs chum/cskill/0 and service costs defined by c_rob and the skill-return model (Eq. 4) are encoded in Algorithm 1; the UFL objective (Eq. 5) is then solved with the external Jain-Vazirani approximation algorithm. Nothing in the UFL objective is defined in terms of the paper's conclusions, and the bounded-suboptimal/polynomial-time claims are imported from an external approximation algorithm with its own metric/non-metric conditions. The preference-request extension (lines 15-23) compares the current plan value to the expected value of replanning after a belief update, which is self-consistent planning rather than circular. The experiments use the same hand-set costs in the planner and in the evaluation metric, but that is a benchmark under a common cost model, not a fitted parameter renamed as a prediction; task success comes from simulator and physical rollouts. The self-citations (HiP-MDP [9], ADL [33]) are used as a modeling formalism and as a baseline, respectively, and are not load-bearing. The skeptic's observation that a UFL solution may open a skill facility at task i without assigning task i to it is a feasibility/correctness gap in the encoding, not a circular reduction; it is therefore outside the circularity score.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The central claim rests on hand-set interaction costs, a known task sequence, discrete stationary preferences, and a hand-specified similarity model for skill transfer. These are inputs to the planner rather than fitted outputs, but the reported cost savings are only meaningful under these assumptions.

free parameters (2)
  • human cost profile parameters = chum=80, cpref=20, crob=10, cskill-fail=100, cpref-fail=100, cskill in {50,100,200}
    Hand-set costs used both for planning and evaluation; reported cost savings depend on these choices.
  • epsilon (similarity threshold) = 0.01
    Threshold in rho_safe for defining exact task-type matches; chosen as a small constant, not tuned.
assumptions (4)
  • domain assumption The task sequence is known in advance.
    COIL plans over the full sequence; the Limitations section states this prior knowledge may be hard to obtain in real settings.
  • domain assumption User preferences are discrete and stationary.
    The HiP-MDP and belief updates model preferences as a fixed discrete set; the Limitations section acknowledges preferences may change.
  • domain assumption The similarity function rho_safe accurately predicts skill generalization.
    The planner's expected return model relies on rho_safe based on task and preference similarity; the Limitations section notes this prior may be inaccurate.
  • ad hoc to paper Each task is completed by exactly one action, and a skill facility must be requested on the task it is opened for.
    The interaction protocol executes one action per task, but the UFL encoding does not enforce that an opened skill facility serves its origin, so solutions can be non-executable.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Optimal Interactive Learning on the Job via Facility Location Planning." pith.science (2026). https://pith.science/paper/ARH5MEDQ

@misc{pith2026250500490,
  author       = {Pith},
  title        = {Pith review of: Optimal Interactive Learning on the Job via Facility Location Planning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ARH5MEDQ}},
  note         = {Machine review of arXiv:2505.00490}
}
read the original abstract

Collaborative robots must continually adapt to novel tasks and user preferences without overburdening the user. While prior interactive robot learning methods aim to reduce human effort, they are typically limited to single-task scenarios and are not well-suited for sustained, multi-task collaboration. We propose COIL (Cost-Optimal Interactive Learning) -- a multi-task interaction planner that minimizes human effort across a sequence of tasks by strategically selecting among three query types (skill, preference, and help). When user preferences are known, we formulate COIL as an uncapacitated facility location (UFL) problem, which enables bounded-suboptimal planning in polynomial time using off-the-shelf approximation algorithms. We extend our formulation to handle uncertainty in user preferences by incorporating one-step belief space planning, which uses these approximation algorithms as subroutines to maintain polynomial-time performance. Simulated and physical experiments on manipulation tasks show that our framework significantly reduces the amount of work allocated to the human while maintaining successful task completion.

Figures

Figures reproduced from arXiv: 2505.00490 by the authors.

Figure 1
Figure 1. A human-robot team working together to pick objects off a conveyor and pack them in bins. The robot queries the human online to learn motor skills and user preference about how each object should be grasped and where it should be placed. We propose a planning algorithm COIL that optimizes how the robot allocates tasks and uses these different types of queries to minimize human effort during the course of its deploym… view at source ↗
Figure 2
Figure 2. Facility location formulation. Tasks τ1, . . . , τ5 are demands to be satisfied. Facilities correspond to interactive actions available for every task. We highlight facilities for τ2: Human facility can only service τ2. Skill facility can service similar future tasks τ2, τ4, τ5. Robot facility cannot service any task as the robot hasn’t learned a skill yet. Furthermore, none of the facilities can service past tasks.… view at source ↗
Figure 3
Figure 3. Under Med- and High-Cost teaching cost profiles, COIL consistently chooses the lowest cost plan compared with baselines. We highlight the qualitative behavior of COIL compared to baselines: COIL under the medium cost teaching profile assigns singleton tasks to the human when the cost of learning is high. Error bars indicate standard error over 30 randomized task sequences and true human preferences in the Gridworld … view at source ↗
Figures from the paper (7 more)
Figure 5
Figure 5. Figure 5: In the Gridworld environment, when half of the objects are challenging-to-learn, COIL achieves the lowest plans, compared to COIL-NoAdapt and other baselines across teaching profiles. In high cost teaching, COIL and COIL-NoAdapt often assign tasks to the human off the …
Figure 6
Figure 6. Figure 6: In the Simulated 7DoF Manipulation domain, COIL achieves significantly lower plan costs than baseline methods. Error bars represent standard error over 10 randomized initialization of the 30- object task sequence. Algorithm #teach #human #pref #robot Cost COIL 1 (1) 12…
Figure 7
Figure 7. Figure 7 [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]
Figure 8
Figure 8. Figure 8: Manipulation domain. Each interaction involves picking 50 objects and placing them in their respective bins. 2) 7DoF Manipulation is a simulated decluttering task where a Franka robot manipulator must pick up and put away 30 total objects of seven different varieties (…
Figure 9
Figure 9. Figure 9: We plot the average plan cost for 10 seeds per all of the three human cost profiles. For IG in our comparisons, we choose the scale factor which achieves the lowest plan cost, β = 0.01. APPENDIX E DOMAIN-SPECIFIC INSTANTIATIONS OF SKILL RETURN MODEL AND PREFERENCE BELI…
Figure 10
Figure 10. Figure 10: In the Gridworld environment, when 10% of the objects are challenging-to-learn, COIL achieves the lowest plans, compared to COIL-NoAdapt and other baselines in all teaching profiles. In high￾cost and med-cost teaching, COIL and COIL-NoAdapt often assign tasks to the h…
Figure 11
Figure 11. Figure 11: In the Gridworld environment, when 90% of the objects are challenging-to-learn, COIL achieves the lowest plans, compared to COIL-NoAdapt and other baselines in all teaching profiles. In high￾cost and med-cost teaching, COIL and COIL-NoAdapt often assign tasks to the h…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

47 extracted references · 29 canonical work pages

  1. [1]

    Robot, organize my shelves! Tidying up objects by predicting user preferences

    Nichola Abdo, Cyrill Stachniss, Luciano Spinello, and Wolfram Burgard. Robot, organize my shelves! Tidying up objects by predicting user preferences. In 2015 IEEE International Conference on Robotics and Automa- tion (ICRA) , pages 1557–1564. doi: 10.1109/ICRA. 2015.7139396. URL https://ieeexplore.ieee.org/abstract/ document/7139396

    arXiv 2015

  2. [2]

    Losey, Marcia K

    Andrea Bajcsy, Dylan P. Losey, Marcia K. O’Malley, and Anca D. Dragan. Learning from Physical Human Correc- tions, One Feature at a Time. In 2018 13th ACM/IEEE International Conference on Human-Robot Interaction (HRI), pages 141–149. URL https://ieeexplore.ieee.org/ document/9473611/?arnumber=9473611

    work page arXiv 2018

  3. [3]

    Learning to optimize autonomy in competence- aware systems

    Connor Basich, Justin Svegliato, Kyle Hollins Wray, Stefan Witwicki, Joy-Deep Biswas, and Shlomo Zilber- stein. Learning to optimize autonomy in competence- aware systems. In Proceedings of the International Joint Conference on Autonomous Agents and Multiagent Systems, AAMAS , volume 2020-May, pages 123–131. ISBN 978-1-4503-7518-4. URL www.ifaamas.org

    work page 2020

  4. [4]

    Learning reward functions from diverse sources of human feedback: Optimally integrating demonstrations and pref- erences

    Erdem Bıyık, Dylan P Losey, Malayandi Palan, Nicholas C Landolfi, Gleb Shevchuk, and Dorsa Sadigh. Learning reward functions from diverse sources of human feedback: Optimally integrating demonstrations and pref- erences. The International Journal of Robotics Research, 41(1):45–67, 2022

    2022

  5. [5]

    Batch active learning of reward functions from human prefer- ences

    Erdem Biyik, Nima Anari, and Dorsa Sadigh. Batch active learning of reward functions from human prefer- ences. ACM Transactions on Human-Robot Interaction , 13(2):1–27, 2024

    work page 2024

  6. [6]

    Confidence- based policy learning from demonstration using Gaussian mixture models

    Sonia Chernova and Manuela Veloso. Confidence- based policy learning from demonstration using Gaussian mixture models. 5:1315–1322. doi: 10.1145/1329125. 1329407

    work page doi:10.1145/1329125

  7. [7]

    Minigrid & miniworld: Modular & customizable rein- forcement learning environments for goal-oriented tasks

    Maxime Chevalier-Boisvert, Bolun Dai, Mark Towers, Rodrigo de Lazcano, Lucas Willems, Salem Lahlou, Suman Pal, Pablo Samuel Castro, and Jordan Terry. Minigrid & miniworld: Modular & customizable rein- forcement learning environments for goal-oriented tasks. CoRR, abs/2306.13831, 2023

    arXiv 2023

  8. [8]

    Deep reinforcement learning from human preferences

    Paul F Christiano, Jan Leike, Tom Brown, Miljan Martic, Shane Legg, and Dario Amodei. Deep reinforcement learning from human preferences. Advances in neural information processing systems , 30, 2017

    2017

Show all 47 references
  1. [9]

    Hidden parameter markov decision processes: A semiparamet- ric regression approach for discovering latent task parametrizations

    Finale Doshi-Velez and George Konidaris. Hidden parameter markov decision processes: A semiparamet- ric regression approach for discovering latent task parametrizations. In IJCAI: proceedings of the confer- ence, volume 2016, page 1432, 2016

    2016

  2. [10]

    INQUIRE: INteractive Querying for User- aware Informative REasoning

    Tesca Fitzgerald, Pallavi Koppol, Patrick Callaghan, Rus- sell Q Wong, Reid Simmons, Oliver Kroemer, and Henny Admoni. INQUIRE: INteractive Querying for User- aware Informative REasoning

  3. [11]

    Gurobi Optimizer Reference Manual, 2024

    Gurobi Optimization, LLC. Gurobi Optimizer Reference Manual, 2024. URL https://www.gurobi.com

    2024

  4. [12]

    Cooperative Inverse Reinforcement Learning

    Dylan Hadfield-Menell, Anca Dragan, Pieter Abbeel, and Stuart Russell. Cooperative Inverse Reinforcement Learning. URL http://arxiv.org/abs/1606.03137

    arXiv

  5. [13]

    Contrastive prefence learning: Learning from human feedback without rl

    Joey Hejna, Rafael Rafailov, Harshit Sikchi, Chelsea Finn, Scott Niekum, W Bradley Knox, and Dorsa Sadigh. Contrastive prefence learning: Learning from human feedback without rl. arXiv preprint arXiv:2310.13639 , 2023

    2023 arXiv

  6. [14]

    Prompt- able behaviors: Personalizing multi-objective rewards from human preferences

    Minyoung Hwang, Luca Weihs, Chanwoo Park, Kimin Lee, Aniruddha Kembhavi, and Kiana Ehsani. Prompt- able behaviors: Personalizing multi-objective rewards from human preferences. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 1...

    2024

  7. [15]

    Adaptive Automation: Sharing and Trading of Control

    Toshiyuki Inagaki. Adaptive Automation: Sharing and Trading of Control. 2001.10:79–84. doi: 10.1299/ jsmetld.2001.10.79

    2001

  8. [16]

    Primal-dual ap- proximation algorithms for metric facility location and k-median problems

    Kamal Jain and Vijay V Vazirani. Primal-dual ap- proximation algorithms for metric facility location and k-median problems. In 40th annual symposium on foundations of computer science (Cat. No. 99CB37039) , pages 2–13. IEEE, 1999

    1999

  9. [17]

    Reward-rational (implicit) choice: A unifying formalism for reward learning

    Hong Jun Jeon, Smitha Milli, and Anca Dragan. Reward-rational (implicit) choice: A unifying formalism for reward learning. In Advances in Neural Information Processing Systems , volume 33, pages 4415–4426. Curran Associates, Inc., 2020. URL https://proceedings.neurips.cc/paper...

    2020

  10. [18]

    Open- vocabulary pick and place via patch-level semantic maps

    Mingxi Jia, Haojie Huang, Zhewen Zhang, Chenghao Wang, Linfeng Zhao, Dian Wang, Jason Xinyu Liu, Robin Walters, Robert Platt, and Stefanie Tellex. Open- vocabulary pick and place via patch-level semantic maps. arXiv preprint arXiv:2406.15677 , 2024

    2024 arXiv

  11. [19]

    T test as a parametric statistic

    Tae Kyun Kim. T test as a parametric statistic. Korean journal of anesthesiology , 68(6):540–546, 2015

    2015

  12. [20]

    Interaction considerations in learning from humans

    Pallavi Koppol, Henny Admoni, and Reid G Simmons. Interaction considerations in learning from humans. In IJCAI, pages 283–291, 2021

    2021

  13. [21]

    Robot Learning on the Job: Human- in-the-Loop Autonomy and Learning During Deploy- ment

    Huihan Liu, Soroush Nasiriany, Lance Zhang, Zhiyao Bao, and Yuke Zhu. Robot Learning on the Job: Human- in-the-Loop Autonomy and Learning During Deploy- ment. URL http://arxiv.org/abs/2211.08416

    arXiv

  14. [22]

    Mehta and Dylan P

    Shaunak A. Mehta and Dylan P. Losey. Unified Learn- ing from Demonstrations, Corrections, and Preferences during Physical Human-Robot Interaction, January 2024. URL http://arxiv.org/abs/2207.03395. arXiv:2207.03395 [cs]

    2024 arXiv

  15. [23]

    Mapping out the space of hu- man feedback for reinforcement learning: A conceptual framework, 2024

    Yannick Metz, David Lindner, Rapha ¨el Baur, and Men- natallah El-Assady. Mapping out the space of hu- man feedback for reinforcement learning: A conceptual framework, 2024. URL https://arxiv.org/abs/2411.11761

    2024 arXiv

  16. [24]

    Learning human preferences over robot be- havior as soft planning constraints

    Austin Narcomey, Nathan Tsoi, Ruta Desai, and Marynel V´azquez. Learning human preferences over robot be- havior as soft planning constraints. arXiv preprint arXiv:2403.19795, 2024

    2024 arXiv

  17. [25]

    Heramb Nemlekar, Robert Ramirez Sanchez, and Dy- lan P. Losey. Pecan: Personalizing robot behaviors through a learned canonical space, 2024. URL https: //arxiv.org/abs/2407.16081

    2024 arXiv

  18. [26]

    Pragmatic feature preferences: Learning reward- relevant preferences from human input

    Andi Peng, Yuying Sun, Tianmin Shu, and David Abel. Pragmatic feature preferences: Learning reward- relevant preferences from human input. arXiv preprint arXiv:2405.14769, 2024

    2024 arXiv

  19. [27]

    Teacher- aware active robot learning

    Mattia Racca, Antti Oulasvirta, and Ville Kyrki. Teacher- aware active robot learning. In 2019 14th ACM/IEEE International Conference on Human-Robot Interaction (HRI), pages 335–343. IEEE, 2019

    2019

  20. [28]

    Direct preference optimization: Your language model is secretly a reward model

    Rafael Rafailov, Archit Sharma, Eric Mitchell, Christo- pher D Manning, Stefano Ermon, and Chelsea Finn. Direct preference optimization: Your language model is secretly a reward model. Advances in Neural Information Processing Systems, 36, 2024

    2024

  21. [29]

    Active Preference-Based Learning of Reward Functions

    Dorsa Sadigh, Anca Dragan, Shankar Sastry, and Sanjit Seshia. Active Preference-Based Learning of Reward Functions. In Robotics: Science and Systems XIII . ISBN 978-0-9923747-3-0. doi: 10.15607/RSS.2017.XIII.053

    2017 doi

  22. [30]

    Human-autonomy teaming using flexible human performance models: An initial pilot study

    Christopher J Shannon, David C Horney, Kimberly F Jackson, and Jonathan P How. Human-autonomy teaming using flexible human performance models: An initial pilot study. Advances in Human Factors in Robots and Unmanned Systems, page 211, 2017

    2017

  23. [31]

    Analysis of variance (anova)

    Lars St, Svante Wold, et al. Analysis of variance (anova). Chemometrics and intelligent laboratory systems , 6(4): 259–272, 1989

    1989

  24. [32]

    Si- multaneously learning intentions and preferences during physical human-robot cooperation

    Linda van der Spaa, Jens Kober, and Michael Gienger. Si- multaneously learning intentions and preferences during physical human-robot cooperation. Autonomous Robots, 48(4):11, 2024

    2024

  25. [33]

    Synergistic scheduling of learning and allocation of tasks in human-robot teams

    Shivam Vats, Oliver Kroemer, and Maxim Likhachev. Synergistic scheduling of learning and allocation of tasks in human-robot teams. In 2022 International Conference on Robotics and Automation (ICRA) , pages 2789–2795. IEEE, 2022

    2022

  26. [34]

    Bonferroni correction

    Eric W Weisstein. Bonferroni correction. https://mathworld. wolfram. com/ , 2004

    2004

  27. [35]

    The design of approximation algorithms

    David P Williamson and David B Shmoys. The design of approximation algorithms. Cambridge university press, 2011

    2011

  28. [36]

    Conformalized interactive imitation learning: Handling expert shift and intermittent feedback, 2024

    Michelle Zhao, Reid Simmons, Henny Admoni, Aaditya Ramdas, and Andrea Bajcsy. Conformalized interactive imitation learning: Handling expert shift and intermittent feedback, 2024. URL https://arxiv.org/abs/2410.08852

    2024 arXiv

  29. [37]

    Learning human contribution preferences in collaborative human-robot tasks

    Michelle D Zhao, Reid Simmons, and Henny Admoni. Learning human contribution preferences in collaborative human-robot tasks. In Jie Tan, Marc Toussaint, and Kourosh Darvish, editors, Proceedings of The 7th Confer- ence on Robot Learning , volume 229 of Proceedings of Machine L...

    2023

  30. [38]

    robosuite: A modular simulation framework and benchmark for robot learning

    Yuke Zhu, Josiah Wong, Ajay Mandlekar, Roberto Mart´ın-Mart´ın, Abhishek Joshi, Soroush Nasiriany, Yifeng Zhu, and Kevin Lin. robosuite: A modular simulation framework and benchmark for robot learning. In arXiv preprint arXiv:2009.12293 , 2020. APPENDIX A BAYESIAN PREFERENCE B...

    2009 arXiv

  31. [39]

    Each object defines a task the agent must execute, and each task is to navigate to an object, pick it up, and transport it to the user’s preferred location for that object

    Gridworld is a discrete, 17x17 grid comprised of a sequence of 15 total objects of nine varieties distin- guished by object type, color, and position. Each object defines a task the agent must execute, and each task is to navigate to an object, pick it up, and transport it to ...

  32. [40]

    The frequency and order of all the objects are randomly sampled to generate a task sequence

    7DoF Manipulation is a simulated decluttering task where a Franka robot manipulator must pick up and put away 30 total objects of seven different varieties (e.g., milk carton, mug) into one of four different shelves. The frequency and order of all the objects are randomly samp...

  33. [41]

    The robot is initialized with an empty skill library L0

    Conveyor is a real-world environment where a Franka tabletop manipulator needs to pick up the object on the conveyor and place it into one of the three colored boxes to which it belongs. The robot is initialized with an empty skill library L0. We randomly sample five task sequ...

  34. [42]

    Low-Cost Teaching (cskill = 50): This profile simulates an experienced teacher for whom providing demonstra- tions of robot skills is not burdensome

  35. [43]

    Med-Cost Teaching (cskill = 100): This profile models teaching to be moderately more burdensome than per- forming the task themselves

  36. [44]

    APPENDIX D BASELINE DETAILS In this section, we describe in detail the baselines used in our experimental evaluation

    High-Cost Teaching (cskill = 200 ): This profile simu- lates a novice teacher for whom providing demonstra- tions of robot skills is highly burdensome. APPENDIX D BASELINE DETAILS In this section, we describe in detail the baselines used in our experimental evaluation. A. Info...

  37. [45]

    c(rob) = ci rob +cskill-fail· (1−ρsafe π,Lt(τj, ˆθj)) +cpref-fail· (1−bθ j(ˆθj))

  38. [46]

    c(skill) =ci skill +ci rob +cskill-fail· (1−ρsafe π,Lt(τj, ˆθj))

  39. [47]

    Confidence-based ADL (C-ADL) Vats et al

    c(hum) =ci hum B. Confidence-based ADL (C-ADL) Vats et al. [33] use a mixed integer program to decide when to learn new skills and when to delegate tasks to the user. While their method accounts for the future utility of learning a task, they do so without also considering the...

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.