pith. sign in

arxiv: 2605.17601 · v1 · pith:AUDQIDRCnew · submitted 2026-05-17 · 💻 cs.RO

From a Single Demonstration to a General Policy for Contact-Rich Manipulation

Xing Li , Oliver Brock This is my paper

Pith reviewed 2026-05-20 12:11 UTC · model grok-4.3

classification 💻 cs.RO
keywords learning from demonstrationcontact-rich manipulationenvironmental constraintsone-shot generalizationrobot policy learningmulti-stage tasksconstraint-based control
0
0 comments X

The pith

Environmental constraints let a robot turn one demonstration into a policy that generalizes across object poses and contact variations in multi-stage tasks.

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

The paper introduces a learning-from-demonstration method that extracts task structure from a single example by identifying the environmental constraints the robot must follow at each stage. It then uses exploration to resolve ambiguities, targeted corrections for new variations, and compliant control to recover details during execution. This produces a policy that succeeds on new object positions, local geometries, and unmodeled dynamics instead of copying the exact demonstrated path. A reader should care because contact-rich tasks like assembly usually demand many demonstrations or detailed models, whereas this approach claims to achieve reliable performance after one showing plus light interaction.

Core claim

By representing a demonstration as a sequence of behaviors that exploit environmental constraints, the robot separates task-general structure—the constraint types and their transitions—from instance-specific details such as exact trajectories, poses, and local geometries. The four-stage pipeline first abstracts the demonstration into environmental-constraint primitives, then disambiguates them through self-guided exploration, assimilates human corrections for out-of-distribution cases, and finally recovers abstracted details online through compliant interaction. Because the resulting policy follows constraints rather than mimics trajectories, it generalizes across object poses, local geometr

What carries the argument

Environmental-constraint primitives: a sequence of constraint types and transitions extracted from the demonstration that encode the task-general structure while discarding instance-specific trajectory details.

If this is right

  • The policy succeeds on new object poses and local geometries without retraining because it follows constraints rather than exact paths.
  • Unmodeled contact dynamics are tolerated because the policy relies on compliant interaction instead of precise trajectory replay.
  • The same framework applies across seven distinct real-world multi-stage tasks with over 90 percent success.
  • Task-general structure is isolated from instance-specific details, reducing the need for task-specific modeling.

Where Pith is reading between the lines

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

  • The approach may reduce the number of human demonstrations required for new contact-rich tasks by shifting effort to online correction and exploration.
  • It could extend to tasks with greater variability if the constraint primitive vocabulary is expanded beyond the current set.
  • Integration with vision or tactile sensing might automate the abstraction stage more reliably for unseen geometries.

Load-bearing premise

A single demonstration can be reliably turned into a sequence of environmental-constraint primitives whose types and transitions contain all task-general structure, so that later exploration and corrections can handle remaining variations.

What would settle it

A new multi-stage contact task where the initial abstraction from the single demonstration produces incorrect constraint types or transitions, causing the policy to fail even after exploration and corrections.

Figures

Figures reproduced from arXiv: 2605.17601 by Oliver Brock, Xing Li.

Figure 1
Figure 1. Figure 1: Our LfD approach generalizes from a single demonstration and a few corrections more strongly than other approaches in the literature. This [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 3
Figure 3. Figure 3: Overview of the four-stage LfD pipeline. A single human demonstration is segmented into a sequence of motion phases, where the exact demonstration [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Segmentation result of a demonstration for a single-object insertion [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Overview of the four EC primitives. The free-space primitive aligns the end-effector to the target object through visual servoing using a wrist￾mounted camera, enabling adaptation to changes in object pose. It then replays the transferred demonstrated trajectory to reproduce contact-free motions. The plane primitive first replays the transferred demonstration trajectory along a surface and then refines its… view at source ↗
Figure 6
Figure 6. Figure 6: Vision-based augmentation to filter unstable features. By actively [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗
Figure 8
Figure 8. Figure 8: Observed z-axis forces as the robot approaches the lock surface [PITH_FULL_IMAGE:figures/full_fig_p009_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: The three insertion puzzles used in our experiments. While each [PITH_FULL_IMAGE:figures/full_fig_p010_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Illustration of success and failure modes in the puzzle insertion [PITH_FULL_IMAGE:figures/full_fig_p010_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Solving a green puzzle by following a sequence of ECs. [PITH_FULL_IMAGE:figures/full_fig_p011_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Experimental setup for the door-lock manipulation task. The policy [PITH_FULL_IMAGE:figures/full_fig_p012_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Policy refinement for door lock manipulation. The policy learned [PITH_FULL_IMAGE:figures/full_fig_p012_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Execution of the coffee machine manipulation policy consisting [PITH_FULL_IMAGE:figures/full_fig_p013_14.png] view at source ↗
Figure 16
Figure 16. Figure 16: EC-based policy execution across different latches. The top row [PITH_FULL_IMAGE:figures/full_fig_p013_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: Execution of the policy of opening two drawers. By generating [PITH_FULL_IMAGE:figures/full_fig_p014_17.png] view at source ↗
Figure 18
Figure 18. Figure 18: Failures and corrections in the FMB assembly task. The top row [PITH_FULL_IMAGE:figures/full_fig_p014_18.png] view at source ↗
Figure 19
Figure 19. Figure 19: Two failure cases observed in the FMB and FB tasks due to incorrect [PITH_FULL_IMAGE:figures/full_fig_p014_19.png] view at source ↗
Figure 19
Figure 19. Figure 19: J. Lessons Learned From Real-World Contact-Rich Manip￾ulation Based on our extensive real-world experiments, we share several hands-on lessons that reveal the fundamental chal￾lenges in contact-rich manipulation. First, exact task trajec￾tories, like the angle needed to turn a handle or open a latch, are nearly impossible to infer from visual observa￾tions alone. Second, when manipulating tightly constrai… view at source ↗
Figure 20
Figure 20. Figure 20: Complex multi-stage contact-rich manipulation tasks can be decomposed into sequential motions. Our EC primitives are able to reproduce these [PITH_FULL_IMAGE:figures/full_fig_p016_20.png] view at source ↗
Figure 21
Figure 21. Figure 21: Experimental setup for evaluating the impact of vision-based [PITH_FULL_IMAGE:figures/full_fig_p021_21.png] view at source ↗
Figure 22
Figure 22. Figure 22: Effect of vision-based augmentation on the insertion puzzle task. [PITH_FULL_IMAGE:figures/full_fig_p021_22.png] view at source ↗
read the original abstract

We present a Learning from Demonstration (LfD) framework that achieves one-shot generalization in multi-stage, contact-rich manipulation tasks. Central to our approach is the utilization of environmental constraints as the inductive bias. By representing a demonstration as a sequence of behaviors that exploit environmental constraints, the robot separates task-general structure -- the constraint types and their transitions -- from instance-specific details such as exact demonstration trajectories, poses, and local geometries. Our four-stage pipeline builds a complete policy on this representation: the robot first abstracts a single demonstration into environmental-constraint primitives, then disambiguates them through self-guided exploration, next assimilates targeted human corrections that handle out-of-distribution variations, and finally recovers the abstracted-away details online through compliant interaction. Because the resulting policy follows constraints rather than mimics trajectories, it generalizes across object poses, local geometries, and unmodeled contact dynamics. We validate our approach on seven real-world multi-stage contact-rich manipulation tasks and achieve over 90% success. These extensive experimental results establish environmental constraints as fundamental building blocks for efficient generalization in learning from demonstration.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The manuscript presents a Learning from Demonstration (LfD) framework for one-shot generalization in multi-stage, contact-rich manipulation tasks. It represents a single demonstration as a sequence of environmental-constraint primitives to separate task-general structure (constraint types and transitions) from instance-specific details (trajectories, poses, geometries). A four-stage pipeline follows: abstraction into primitives, self-guided exploration for disambiguation, assimilation of targeted human corrections, and online recovery of details via compliant interaction. The resulting policy is claimed to generalize across object poses, local geometries, and unmodeled contact dynamics. Validation is reported on seven real-world tasks with over 90% success.

Significance. If the central claims hold, the work would be significant for LfD in robotics by demonstrating that environmental constraints can serve as an effective inductive bias for generalization from a single demonstration, reducing reliance on trajectory imitation or extensive task-specific modeling. The real-world experimental scope across seven multi-stage tasks is a positive aspect that could support practical adoption if the generalization mechanism is rigorously validated.

major comments (2)
  1. [Abstract / Pipeline Description] The generalization argument rests on the first stage successfully abstracting a single demonstration into a sequence of constraint primitives whose types and transitions capture all task-general structure (see Abstract and pipeline description). The manuscript provides no analysis or experiments demonstrating that this abstraction step is invariant to the choice of demonstration trajectory or local geometry; if instance-specific details leak into the primitive sequence, later stages of exploration and correction cannot remain free of task-specific modeling, undermining the claims of generalization across poses, geometries, and unmodeled dynamics.
  2. [Experimental Validation] The experimental results claim over 90% success on seven tasks (Abstract), yet the provided description includes no details on task definitions, failure modes, statistical tests, controls, or how success was measured across variations in object poses and geometries. Without these, it is not possible to evaluate whether the data support the one-shot generalization claim or to rule out that performance depends on particular demonstration choices.
minor comments (1)
  1. [Abstract] The abstract would be clearer if it briefly listed the types of environmental constraints or the specific manipulation tasks used, to help readers assess the scope of the claimed generalization.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. We address each major comment below and will revise the manuscript to strengthen the presentation of the abstraction process and experimental details.

read point-by-point responses

  1. Referee: [Abstract / Pipeline Description] The generalization argument rests on the first stage successfully abstracting a single demonstration into a sequence of constraint primitives whose types and transitions capture all task-general structure (see Abstract and pipeline description). The manuscript provides no analysis or experiments demonstrating that this abstraction step is invariant to the choice of demonstration trajectory or local geometry; if instance-specific details leak into the primitive sequence, later stages of exploration and correction cannot remain free of task-specific modeling, undermining the claims of generalization across poses, geometries, and unmodeled dynamics.

    Authors: We agree that explicit validation of the abstraction step's invariance is important for supporting the generalization claims. The abstraction is defined to extract only constraint types and transition sequences while discarding instance-specific trajectory and geometry details, but the current manuscript does not include dedicated analysis or multi-demonstration experiments to demonstrate this property. In the revised version we will add a new subsection with theoretical justification based on the constraint representation and empirical results applying the abstraction to multiple demonstrations per task with varied trajectories and local geometries, confirming that the resulting primitive sequences remain consistent. revision: yes

  2. Referee: [Experimental Validation] The experimental results claim over 90% success on seven tasks (Abstract), yet the provided description includes no details on task definitions, failure modes, statistical tests, controls, or how success was measured across variations in object poses and geometries. Without these, it is not possible to evaluate whether the data support the one-shot generalization claim or to rule out that performance depends on particular demonstration choices.

    Authors: We acknowledge that the experimental section requires more detailed reporting to allow full evaluation of the results. While the manuscript describes the seven tasks at a high level and reports aggregate success rates, it does not provide the requested specifics on definitions, variations, failure modes, measurement criteria, or statistical analysis. In the revision we will substantially expand the experimental evaluation to include explicit task definitions, the ranges of pose and geometry variations tested, per-task success rates with breakdowns, observed failure modes and their frequencies, precise success criteria for each stage, and statistical measures such as standard deviations across trials. This will better substantiate the one-shot generalization performance. revision: yes

Circularity Check

0 steps flagged

No circularity: forward pipeline from demonstration to constraint-based policy remains self-contained

full rationale

The paper presents a four-stage pipeline that begins with abstracting a single demonstration into environmental-constraint primitives, followed by exploration, human corrections, and online recovery of details. Generalization is claimed to arise because the policy follows constraints rather than mimicking trajectories, with validation on seven real-world tasks. No equations, fitted parameters, or self-citations are exhibited that reduce any claimed prediction or uniqueness result back to the input demonstration by construction. The separation of task-general structure from instance-specific details is treated as an empirical outcome of the abstraction step rather than a definitional equivalence, leaving the derivation chain independent of its own outputs.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the domain assumption that environmental constraints form a sufficient inductive bias to separate general task structure from instance-specific details; no free parameters or invented entities are mentioned in the abstract.

axioms (1)
  • domain assumption A demonstration can be represented as a sequence of behaviors that exploit environmental constraints, separating task-general structure from instance-specific details.
    This representation is the foundation for the entire pipeline and generalization claim.

pith-pipeline@v0.9.0 · 5712 in / 1343 out tokens · 48900 ms · 2026-05-20T12:11:13.688229+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

141 extracted references · 141 canonical work pages · 8 internal anchors

  1. [1]

    Survey: Robot programming by demonstration,

    A. Billard, S. Calinon, R. Dillmann, and S. Schaal, “Survey: Robot programming by demonstration,”Springer Handbook of Robotics, pp. 1371–1394, 2008

    work page 2008

  2. [2]

    Exploitation of environmental constraints in human and robotic grasping,

    C. Eppner, R. Deimel, J. Alvarez-Ruiz, M. Maertens, and O. Brock, “Exploitation of environmental constraints in human and robotic grasping,”The International Journal of Robotics Research, vol. 34, no. 7, pp. 1021–1038, 2015

    work page 2015

  3. [3]

    Human- inspired force compliant grasping primitives,

    M. Kazemi, J.-S. Valois, J. A. Bagnell, and N. Pollard, “Human- inspired force compliant grasping primitives,”Autonomous Robots, vol. 37, pp. 209–225, 2014

    work page 2014

  4. [4]

    Experimental analysis of human control strategies in contact manipulation tasks,

    E. Klingbeil, S. Menon, and O. Khatib, “Experimental analysis of human control strategies in contact manipulation tasks,” inInterna- tional Symposium on Experimental Robotics. Springer, 2017, pp. 275–286

    work page 2017

  5. [5]

    Automatic syn- thesis of fine-motion strategies for robots,

    T. Lozano-P ´erez, M. T. Mason, and R. H. Taylor, “Automatic syn- thesis of fine-motion strategies for robots,”The International Journal of Robotics Research, vol. 3, no. 1, pp. 3–24, 1984

    work page 1984

  6. [6]

    Planning grasp strategies that exploit environmental constraints,

    C. Eppner and O. Brock, “Planning grasp strategies that exploit environmental constraints,” inIEEE International Conference on Robotics and Automation (ICRA), 2015, pp. 4947–4952

    work page 2015

  7. [7]

    Grasping with soft hands,

    M. Bonilla, E. Farnioli, C. Piazza, M. Catalano, G. Grioli, M. Gara- bini, M. Gabiccini, and A. Bicchi, “Grasping with soft hands,” in IEEE-RAS International Conference on Humanoid Robots, 2014, pp. 581–587

    work page 2014

  8. [8]

    Exploiting robot hand compliance and environmental constraints for edge grasps,

    J. Bimbo, E. Turco, M. Ghazaei Ardakani, M. Pozzi, G. Salvietti, V . Bo, M. Malvezzi, and D. Prattichizzo, “Exploiting robot hand compliance and environmental constraints for edge grasps,”Frontiers in Robotics and AI, vol. 6, p. 135, 2019

    work page 2019

  9. [9]

    Analysis of open-loop grasping from piles,

    E. P ´all and O. Brock, “Analysis of open-loop grasping from piles,” in IEEE International Conference on Robotics and Automation (ICRA), 2021, pp. 2591–2597

    work page 2021

  10. [10]

    Interleaving Motion in Contact and in Free Space for Planning Under Uncer- tainty,

    A. Sieverling, C. Eppner, F. Wolff, and O. Brock, “Interleaving Motion in Contact and in Free Space for Planning Under Uncer- tainty,” inIEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2017, pp. 4011–4017

    work page 2017

  11. [11]

    Dual execution of optimized contact interaction trajec- tories,

    M. Toussaint, N. Ratliff, J. Bohg, L. Righetti, P. Englert, and S. Schaal, “Dual execution of optimized contact interaction trajec- tories,” inIEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2014, pp. 47–54

    work page 2014

  12. [12]

    Differentiable physics and stable modes for tool-use and manipula- tion planning,

    M. A. Toussaint, K. R. Allen, K. A. Smith, and J. B. Tenenbaum, “Differentiable physics and stable modes for tool-use and manipula- tion planning,” inRobotics: Science and Systems (RSS), 2018

    work page 2018

  13. [13]

    Learning reusable manipulation strategies,

    J. Mao, T. Lozano-P ´erez, J. B. Tenenbaum, and L. P. Kaelbling, “Learning reusable manipulation strategies,” inConference on Robot Learning (CoRL). PMLR, 2023, pp. 1467–1483

    work page 2023

  14. [14]

    Automated co- design of soft hand morphology and control strategy for grasping,

    R. Deimel, P. Irmisch, V . Wall, and O. Brock, “Automated co- design of soft hand morphology and control strategy for grasping,” inIEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2017, pp. 1213–1218

    work page 2017

  15. [15]

    Co-designing manipulation systems using task-relevant constraints,

    A. Vaish and O. Brock, “Co-designing manipulation systems using task-relevant constraints,” in2024 IEEE International Conference on Robotics and Automation (ICRA), 2024, pp. 4177–4183

    work page 2024

  16. [16]

    Adaptive synergies for the design and control of the pisa/iit softhand,

    M. G. Catalano, G. Grioli, E. Farnioli, A. Serio, C. Piazza, and A. Bicchi, “Adaptive synergies for the design and control of the pisa/iit softhand,”The International Journal of Robotics Research, vol. 33, no. 5, pp. 768–782, 2014

    work page 2014

  17. [17]

    A compliant, underactuated hand for robust manipulation,

    L. U. Odhner, L. P. Jentoft, M. R. Claffee, N. Corson, Y . Tenzer, R. R. Ma, M. Buehler, R. Kohout, R. D. Howe, and A. M. Dollar, “A compliant, underactuated hand for robust manipulation,”The International Journal of Robotics Research, vol. 33, no. 5, pp. 736– 752, 2014

    work page 2014

  18. [18]

    A novel type of compliant and underactu- ated robotic hand for dexterous grasping,

    R. Deimel and O. Brock, “A novel type of compliant and underactu- ated robotic hand for dexterous grasping,”The International Journal of Robotics Research, vol. 35, no. 1-3, pp. 161–185, 2016

    work page 2016

  19. [19]

    Rbo hand 3: A platform for soft dexterous manipulation,

    S. Puhlmann, J. Harris, and O. Brock, “Rbo hand 3: A platform for soft dexterous manipulation,”IEEE Transactions on Robotics, vol. 38, no. 6, pp. 3434–3449, 2022

    work page 2022

  20. [20]

    An exploration of sensorless manipulation,

    M. A. Erdmann and M. T. Mason, “An exploration of sensorless manipulation,”IEEE Journal on Robotics and Automation, vol. 4, no. 4, pp. 369–379, 2002

    work page 2002

  21. [21]

    Manipulation with shared grasping,

    Y . Hou, Z. Jia, and M. Mason, “Manipulation with shared grasping,” inRobotics: Science and Systems, 2020

    work page 2020

  22. [22]

    Collision-inclusive manipulation planning for occluded object grasping via compliant robot motions,

    K. Ren, G. Wang, A. S. Morgan, and K. Hang, “Collision-inclusive manipulation planning for occluded object grasping via compliant robot motions,”arXiv preprint arXiv:2412.06983, 2024

    work page arXiv 2024

  23. [23]

    Extrinsic dexterity: In-hand manipulation with external forces,

    N. C. Dafle, A. Rodriguez, R. Paolini, B. Tang, S. S. Srinivasa, M. Erdmann, M. T. Mason, I. Lundberg, H. Staab, and T. Fuhlbrigge, “Extrinsic dexterity: In-hand manipulation with external forces,” in IEEE International Conference on Robotics and Automation (ICRA), 2014, pp. 1578–1585

    work page 2014

  24. [24]

    Surprisingly robust in-hand manipulation: An empirical study,

    A. Bhatt, A. Sieler, S. Puhlmann, and O. Brock, “Surprisingly robust in-hand manipulation: An empirical study,”Robotics: Science and Systems (RSS), 2021

    work page 2021

  25. [25]

    In-hand cube reconfiguration: Simplified,

    S. Patidar, A. Sieler, and O. Brock, “In-hand cube reconfiguration: Simplified,” inIEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2023, pp. 8751–8756

    work page 2023

  26. [26]

    RT-1: Robotics transformer for real-world control at scale,

    A. Brohan, N. Brown, J. Carbajal, Y . Chebotar, J. Dabis, C. Finn, K. Gopalakrishnan, K. Hausman, A. Herzog, J. Hsu,et al., “RT-1: Robotics transformer for real-world control at scale,” inRobotics: Science and Systems (RSS), 2023

    work page 2023

  27. [27]

    RT-2: Vision-language-action models transfer web knowledge to robotic control,

    B. Zitkovich, T. Yu, S. Xu, P. Xu, T. Xiao, F. Xia, J. Wu, P. Wohlhart, S. Welker, A. Wahid,et al., “RT-2: Vision-language-action models transfer web knowledge to robotic control,” inConference on Robot Learning (CoRL). PMLR, 2023, pp. 2165–2183

    work page 2023

  28. [28]

    Open X-Embodiment: Robotic learning datasets and RT-X models: Open X-Embodiment collaboration,

    A. O’Neill, A. Rehman, A. Maddukuri, A. Gupta, A. Padalkar, A. Lee, A. Pooley, A. Gupta, A. Mandlekar, A. Jain,et al., “Open X-Embodiment: Robotic learning datasets and RT-X models: Open X-Embodiment collaboration,” inIEEE International Conference on Robotics and Automation (ICRA), 2024, pp. 6892–6903

    work page 2024

  29. [29]

    Droid: A large-scale in-the-wild robot manipulation dataset,

    A. Khazatsky, K. Pertsch, S. Nair, A. Balakrishna, S. Dasari, S. Karamcheti, S. Nasiriany, M. K. Srirama, L. Y . Chen, K. Ellis, et al., “Droid: A large-scale in-the-wild robot manipulation dataset,” inRobotics: Science and Systems (RSS), 2024

    work page 2024

  30. [30]

    GR00T N1: An Open Foundation Model for Generalist Humanoid Robots

    J. Bjorck, F. Casta ˜neda, N. Cherniadev, X. Da, R. Ding, L. Fan, Y . Fang, D. Fox, F. Hu, S. Huang,et al., “Gr00t n1: An open foundation model for generalist humanoid robots,”arXiv preprint arXiv:2503.14734, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  31. [31]

    Gemini Robotics: Bringing AI into the Physical World

    G. R. Team, S. Abeyruwan, J. Ainslie, J.-B. Alayrac, M. G. Arenas, T. Armstrong, A. Balakrishna, R. Baruch, M. Bauza, M. Blokzijl, et al., “Gemini robotics: Bringing ai into the physical world,”arXiv preprint arXiv:2503.20020, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  32. [32]

    A Careful Examination of Large Behavior Models for Multitask Dexterous Manipulation

    T. L. Team, “A careful examination of large behavior models for multitask dexterous manipulation,”arXiv preprint arXiv:2507.05331, 2025

    work page internal anchor Pith review arXiv 2025

  33. [33]

    AgiBot World Colosseo: A Large-scale Manipulation Platform for Scalable and Intelligent Embodied Systems

    Q. Bu, J. Cai, L. Chen, X. Cui, Y . Ding, S. Feng, S. Gao, X. He, X. Hu, X. Huang,et al., “Agibot world colosseo: A large-scale ma- nipulation platform for scalable and intelligent embodied systems,” arXiv preprint arXiv:2503.06669, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  34. [34]

    Useek: Unsupervised se (3)-equivariant 3d keypoints for generalizable ma- nipulation,

    Z. Xue, Z. Yuan, J. Wang, X. Wang, Y . Gao, and H. Xu, “Useek: Unsupervised se (3)-equivariant 3d keypoints for generalizable ma- nipulation,”arXiv preprint arXiv:2209.13864, 2022

    work page arXiv 2022

  35. [35]

    Visual imitation learning of task-oriented object grasping and rearrangement,

    Y . Cai, J. Gao, C. Pohl, and T. Asfour, “Visual imitation learning of task-oriented object grasping and rearrangement,” in2024 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2024, pp. 364–371

    work page 2024

  36. [36]

    Dinobot: Robot manipulation via retrieval and alignment with vision foundation models,

    N. Di Palo and E. Johns, “Dinobot: Robot manipulation via retrieval and alignment with vision foundation models,” inIEEE International Conference on Robotics and Automation (ICRA), 2024, pp. 2798– 2805

    work page 2024

  37. [37]

    Hrt1: One-shot human-to-robot trajectory transfer for mobile ma- nipulation,

    S. H. Allu, N. Khargonkar, T. Summers, J. Yao, Y . Xiang,et al., “Hrt1: One-shot human-to-robot trajectory transfer for mobile ma- nipulation,”arXiv preprint arXiv:2510.21026, 2025

    work page arXiv 2025

  38. [38]

    Tref-6: Inferring task-relevant frames from a single demonstration for one-shot skill generalization,

    Y . Ding, S. Wang, and T. Fitzgerald, “Tref-6: Inferring task-relevant frames from a single demonstration for one-shot skill generalization,” arXiv preprint arXiv:2509.00310, 2025

    work page arXiv 2025

  39. [39]

    A backbone for long-horizon robot task understanding,

    X. Chen, W. Chen, D. Lee, Y . Ge, N. Rojas, and P. Kormushev, “A backbone for long-horizon robot task understanding,”IEEE Robotics and Automation Letters, 2025

    work page 2025

  40. [40]

    Annotation-free one-shot imitation learning for multi-step manip- ulation tasks,

    V . Wichitwechkarn, E. Williams, C. Fox, and R. Choudhary, “Annotation-free one-shot imitation learning for multi-step manip- ulation tasks,”arXiv preprint arXiv:2509.24972, 2025

    work page arXiv 2025

  41. [41]

    Functo: Function-centric one-shot imitation learning for tool manipulation,

    C. Tang, A. Xiao, Y . Deng, T. Hu, W. Dong, H. Zhang, D. Hsu, and H. Zhang, “Functo: Function-centric one-shot imitation learning for tool manipulation,”arXiv preprint arXiv:2502.11744, 2025

    work page arXiv 2025

  42. [42]

    Guided uncertainty-aware policy optimization: Combin- ing learning and model-based strategies for sample-efficient policy learning,

    M. A. Lee, C. Florensa, J. Tremblay, N. Ratliff, A. Garg, F. Ramos, and D. Fox, “Guided uncertainty-aware policy optimization: Combin- ing learning and model-based strategies for sample-efficient policy learning,” inIEEE international conference on robotics and automa- tion (ICRA), 2020, pp. 7505–7512

    work page 2020

  43. [43]

    One-shot imitation learning: A pose estimation perspective,

    P. Vitiello, K. Dreczkowski, and E. Johns, “One-shot imitation learning: A pose estimation perspective,” inConference on Robot Learning (CoRL). PMLR, 2023, pp. 943–970

    work page 2023

  44. [44]

    Ditto: Demonstration imitation by trajectory transformation,

    N. Heppert, M. Argus, T. Welschehold, T. Brox, and A. Valada, “Ditto: Demonstration imitation by trajectory transformation,” in IEEE/RSJ International Conference on Intelligent Robots and Sys- tems (IROS), 2024, pp. 7565–7572

    work page 2024

  45. [45]

    Learning a thousand tasks in a day,

    K. Dreczkowski, P. Vitiello, V . V osylius, and E. Johns, “Learning a thousand tasks in a day,”Science Robotics, vol. 10, no. 108, p. eadv7594, 2025

    work page 2025

  46. [46]

    Manipulation primitives: A paradigm for abstraction and execution of grasping and manipulation tasks,

    J. Felip, J. Laaksonen, A. Morales, and V . Kyrki, “Manipulation primitives: A paradigm for abstraction and execution of grasping and manipulation tasks,”Robotics and Autonomous Systems, vol. 61, no. 3, pp. 283–296, 2013

    work page 2013

  47. [47]

    A review of robot learn- ing for manipulation: Challenges, representations, and algorithms,

    O. Kroemer, S. Niekum, and G. Konidaris, “A review of robot learn- ing for manipulation: Challenges, representations, and algorithms,” Journal of Machine Learning Research, vol. 22, no. 30, pp. 1–82, 2021

    work page 2021

  48. [48]

    Acquiring robust, force-based assembly skills from human demonstration,

    M. Skubic and R. A. V olz, “Acquiring robust, force-based assembly skills from human demonstration,”IEEE Transactions on Robotics and Automation, vol. 16, no. 6, pp. 772–781, 2000

    work page 2000

  49. [49]

    Imitation learning- based framework for learning 6-d linear compliant motions,

    M. Suomalainen, F. J. Abu-Dakka, and V . Kyrki, “Imitation learning- based framework for learning 6-d linear compliant motions,”Au- tonomous Robots, vol. 45, no. 3, pp. 389–405, 2021

    work page 2021

  50. [50]

    Robot learning from demon- stration for assembly with sequential assembly movement primitives,

    H. Hu, H. Yan, X. Yang, and Y . Lou, “Robot learning from demon- stration for assembly with sequential assembly movement primitives,” IEEE/ASME Transactions on Mechatronics, 2023

    work page 2023

  51. [51]

    One- shot transfer of long-horizon extrinsic manipulation through contact retargeting,

    A. Wu, R. Wang, S. Chen, C. Eppner, and C. K. Liu, “One- shot transfer of long-horizon extrinsic manipulation through contact retargeting,” inIEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2024, pp. 13 891–13 898

    work page 2024

  52. [52]

    Robot learning of everyday object manipulations via human demonstration,

    H. Dang and P. K. Allen, “Robot learning of everyday object manipulations via human demonstration,” inIEEE/RSJ International Conference on Intelligent Robots and Systems, 2010, pp. 1284–1289

    work page 2010

  53. [53]

    Online bayesian changepoint detection for articulated motion models,

    S. Niekum, S. Osentoski, C. G. Atkeson, and A. G. Barto, “Online bayesian changepoint detection for articulated motion models,” in IEEE international conference on robotics and automation (ICRA), 2015, pp. 1468–1475

    work page 2015

  54. [54]

    C-learn: Learning geometric constraints from demonstrations for multi-step manipulation in shared autonomy,

    C. P ´erez-D’Arpino and J. A. Shah, “C-learn: Learning geometric constraints from demonstrations for multi-step manipulation in shared autonomy,” inIEEE International Conference on Robotics and Au- tomation (ICRA), 2017, pp. 4058–4065

    work page 2017

  55. [55]

    Learning articulated constraints from a one-shot demonstration for robot manipulation planning,

    Y . Liu, F. Zha, L. Sun, J. Li, M. Li, and X. Wang, “Learning articulated constraints from a one-shot demonstration for robot manipulation planning,”IEEE Access, vol. 7, pp. 172 584–172 596, 2019

    work page 2019

  56. [56]

    Learning hybrid object kinematics for efficient hierarchical planning under uncertainty,

    A. Jain and S. Niekum, “Learning hybrid object kinematics for efficient hierarchical planning under uncertainty,” inIEEE/RSJ In- ternational Conference on Intelligent Robots and Systems (IROS), 2020, pp. 5253–5260

    work page 2020

  57. [57]

    Inferring geometric con- straints in human demonstrations,

    G. Subramani, M. Zinn, and M. Gleicher, “Inferring geometric con- straints in human demonstrations,” inConference on Robot Learning (CoRL). PMLR, 2018, pp. 223–236

    work page 2018

  58. [58]

    A learning-from-observation frame- work: One-shot robot teaching for grasp-manipulation-release house- hold operations,

    N. Wake, R. Arakawa, I. Yanokura, T. Kiyokawa, K. Sasabuchi, J. Takamatsu, and K. Ikeuchi, “A learning-from-observation frame- work: One-shot robot teaching for grasp-manipulation-release house- hold operations,” inIEEE/SICE International Symposium on System Integration (SII), 2021, pp. 461–466

    work page 2021

  59. [59]

    Learning from demonstration based on environ- mental constraints,

    X. Li and O. Brock, “Learning from demonstration based on environ- mental constraints,”IEEE Robotics and Automation Letters, vol. 7, no. 4, pp. 10 938–10 945, 2022

    work page 2022

  60. [60]

    Teaching contact- rich tasks from visual demonstrations by constraint extraction,

    C. Hegeler, F. Rozzi, L. Roveda, and K. Haninger, “Teaching contact- rich tasks from visual demonstrations by constraint extraction,”arXiv preprint arXiv:2303.17481, 2023

    work page arXiv 2023

  61. [61]

    K-VIL: Keypoints-based visual imitation learning,

    J. Gao, Z. Tao, N. Jaquier, and T. Asfour, “K-VIL: Keypoints-based visual imitation learning,”IEEE Transactions on Robotics, pp. 1–21, 2023

    work page 2023

  62. [62]

    A helping (human) hand in kinematic structure estimation,

    A. Pfisterer, X. Li, V . Mengers, and O. Brock, “A helping (human) hand in kinematic structure estimation,” inIEEE International Con- ference on Robotics and Automation (ICRA), 2025

    work page 2025

  63. [63]

    A generic task model and control strategy to support learning, robust control, and generalization of contact-rich manipulation tasks,

    A. M. Mohammadi, M. V ochten, E. Aertbeli ¨en, and J. De Schutter, “A generic task model and control strategy to support learning, robust control, and generalization of contact-rich manipulation tasks,” Robotics and Autonomous Systems, p. 105270, 2025

    work page 2025

  64. [64]

    Versatile kinematics-based constraint identification applied to robot task reproduction,

    A. H. Overbeek, D. Dresscher, H. van der Kooij, and M. Vlutters, “Versatile kinematics-based constraint identification applied to robot task reproduction,”Frontiers in robotics and AI, 2025

    work page 2025

  65. [65]

    Identifying physical interactions in contact-based robot ma- nipulation for learning from demonstration,

    A. H. G.-J. Overbeek, H. van der Kooij, D. Dresscher, and M. Vlut- ters, “Identifying physical interactions in contact-based robot ma- nipulation for learning from demonstration,”Advanced Robotics Research, p. e202500109, 2025

    work page 2025

  66. [66]

    Augmentation enables one-shot generalization in learning from demonstration for contact-rich manip- ulation,

    X. Li, M. Baum, and O. Brock, “Augmentation enables one-shot generalization in learning from demonstration for contact-rich manip- ulation,” inIEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2023, pp. 3656–3663

    work page 2023

  67. [67]

    A survey of robot manipulation in contact,

    M. Suomalainen, Y . Karayiannidis, and V . Kyrki, “A survey of robot manipulation in contact,”Robotics and Autonomous Systems, vol. 156, p. 104224, 2022

    work page 2022

  68. [68]

    Active articulation model estimation through interactive perception,

    K. Hausman, S. Niekum, S. Osentoski, and G. S. Sukhatme, “Active articulation model estimation through interactive perception,” inIEEE International Conference on Robotics and Automation (ICRA), 2015, pp. 3305–3312

    work page 2015

  69. [69]

    Kinematics-based estimation of contact constraints using only proprioception,

    V . Ortenzi, H.-C. Lin, M. Azad, R. Stolkin, J. A. Kuo, and M. Mis- try, “Kinematics-based estimation of contact constraints using only proprioception,” inIEEE-RAS 16th International Conference on Humanoid Robots (Humanoids), 2016, pp. 1304–1311

    work page 2016

  70. [70]

    Contact surface estimation via haptic perception,

    H.-C. Lin and M. Mistry, “Contact surface estimation via haptic perception,” inIEEE International Conference on Robotics and Automation (ICRA), 2020, pp. 5087–5093

    work page 2020

  71. [71]

    A reduction of imitation learning and structured prediction to no-regret online learning,

    S. Ross, G. Gordon, and D. Bagnell, “A reduction of imitation learning and structured prediction to no-regret online learning,” in Proceedings of the fourteenth international conference on artificial intelligence and statistics. JMLR Workshop and Conference Pro- ceedings, 2011, pp. 627–635

    work page 2011

  72. [72]

    Interactive imitation learning in robotics: A survey,

    C. Celemin, R. P ´erez-Dattari, E. Chisari, G. Franzese, L. de Souza Rosa, R. Prakash, Z. Ajanovi ´c, M. Ferraz, A. Valada, J. Kober,et al., “Interactive imitation learning in robotics: A survey,” Foundations and Trends® in Robotics, vol. 10, no. 1-2, pp. 1–197, 2022

    work page 2022

  73. [73]

    Learning of exception strategies in assembly tasks,

    B. Nemec, M. Simoni ˇc, and A. Ude, “Learning of exception strategies in assembly tasks,” inIEEE International Conference on Robotics and Automation (ICRA), 2020, pp. 6521–6527

    work page 2020

  74. [74]

    Motion generation based on contact state estimation using two-stage clustering,

    K. Takeuchi, S. Sakaino, and T. Tsuji, “Motion generation based on contact state estimation using two-stage clustering,”IEEJ Journal of Industry Applications, p. 22012635, 2023

    work page 2023

  75. [75]

    Model-based runtime monitoring with interactive imitation learning,

    H. Liu, S. Dass, R. Mart ´ın-Mart´ın, and Y . Zhu, “Model-based runtime monitoring with interactive imitation learning,” inIEEE International Conference on Robotics and Automation (ICRA), 2024

    work page 2024

  76. [76]

    Ilesia: Interactive learning of robot situa- tional awareness from camera input,

    P. Vanc, G. Franzese, J. K. Behrens, C. Della Santina, K. Stepanova, J. Kober, and R. Babuska, “Ilesia: Interactive learning of robot situa- tional awareness from camera input,”IEEE Robotics and Automation Letters, 2025

    work page 2025

  77. [77]

    Data-driven online decision making for autonomous manipulation,

    D. Kappler, P. Pastor, M. Kalakrishnan, M. W ¨uthrich, and S. Schaal, “Data-driven online decision making for autonomous manipulation,” inRobotics: Science and Systems (RSS), vol. 11, 2015

    work page 2015

  78. [78]

    Learning grounded finite-state representations from unstructured demonstrations,

    S. Niekum, S. Osentoski, G. Konidaris, S. Chitta, B. Marthi, and A. G. Barto, “Learning grounded finite-state representations from unstructured demonstrations,”The International Journal of Robotics Research, vol. 34, no. 2, pp. 131–157, 2015

    work page 2015

  79. [79]

    Sc-airl: Share-critic in adversarial inverse reinforcement learning for long-horizon task,

    G. Xiang, S. Li, F. Shuang, F. Gao, and X. Yuan, “Sc-airl: Share-critic in adversarial inverse reinforcement learning for long-horizon task,” IEEE Robotics and Automation Letters, vol. 9, no. 4, pp. 3179–3186, 2024

    work page 2024

  80. [80]

    The need for biases in learning generalizations (technical report no. cbm-tr-117),

    T. M. Mitchell, “The need for biases in learning generalizations (technical report no. cbm-tr-117),”New Brunswick, New Jersey: Department of Computer Science, Rutgers University, 1980

    work page 1980

Showing first 80 references.