pith. sign in

arxiv: 2509.17666 · v2 · pith:RNJD3DRNnew · submitted 2025-09-22 · 💻 cs.RO

Robust and Resilient Soft Robotic Object Insertion with Compliance-Enabled Contact Formation and Failure Recovery

Mimo Shirasaka , Cristian C. Beltran-Hernandez , Masashi Hamaya , Yoshitaka Ushiku This is my paper

Pith reviewed 2026-05-21 22:43 UTC · model grok-4.3

classification 💻 cs.RO
keywords soft roboticsobject insertionfailure recoveryvision-language modelcompliant wristcontact formationrobotic manipulationuncertainty handling
0
0 comments X

The pith

A passively compliant soft wrist structures object insertion into sequential contact formations that enable safe repeated recovery attempts guided by a vision-language model.

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

The paper establishes that a soft robotic wrist's passive compliance absorbs contact forces through large deformations, allowing the insertion process to be broken into compliance-enabled contact formations that progressively constrain the object's degrees of freedom while tolerating initial errors. This structure supports automated failure recovery in which a pre-trained vision-language model evaluates terminal poses and images, identifies what went wrong, and selects the next skill or updates the goal to try again. A sympathetic reader would care because conventional insertion methods often demand precise force sensing, high-speed feedback, or repeated manual tuning when conditions vary, whereas this approach maintains performance across grasp misalignments, position offsets, friction changes, and novel shapes without those requirements.

Core claim

The central claim is that wrist compliance permits safe, repeated recovery attempts by structuring insertion as compliance-enabled contact formations, and that pairing this with a pre-trained vision-language model to assess each execution from terminal poses and images, identify failure modes, and propose recovery actions by selecting skills and updating goals produces resilient insertion under randomized uncertainties.

What carries the argument

compliance-enabled contact formations: sequential contact states that progressively constrain degrees of freedom while the soft wrist absorbs errors through deformation, supported by vision-language model recovery

If this is right

  • The method recovers from grasp misalignments up to 5 degrees and hole-pose errors up to 20 mm through safe contact absorption and repeated attempts.
  • It maintains performance under fivefold friction increases and with previously unseen square or rectangular peg shapes.
  • An 83 percent success rate is achieved in simulation across randomized conditions without high-frequency control or force sensing.
  • The full pipeline is validated on a physical robot, confirming transfer from simulation to hardware.

Where Pith is reading between the lines

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

  • The same compliance-plus-recovery pattern could extend to other contact-rich tasks such as peg-in-hole assembly or tool placement in unstructured settings.
  • Reducing the need for force sensors and precise controllers may allow lower-cost robots to operate reliably in variable factory or household environments.
  • Collecting failure cases from the vision-language model could be used to fine-tune recovery policies for specific robot hardware or object sets.

Load-bearing premise

The pre-trained vision-language model can reliably assess each skill execution from terminal poses and images, correctly identify failure modes, and propose effective recovery actions by selecting skills and updating goals.

What would settle it

A set of trials in which the vision-language model repeatedly misidentifies a failure mode such as excessive friction or proposes an ineffective recovery skill, causing the overall success rate to fall well below 83 percent under the same randomized grasp and hole-pose errors.

Figures

Figures reproduced from arXiv: 2509.17666 by Cristian C. Beltran-Hernandez, Masashi Hamaya, Mimo Shirasaka, Yoshitaka Ushiku.

Figure 1
Figure 1. Figure 1: We propose compliance-enabled contact formations and failure recovery with a soft wrist to achieve more robust and resilient object insertion. denotes reliable execution under uncertainty, and resilience denotes recovery from failures. To enhance robustness, we exploit contact formations [7], discrete contact states that constrain the parts’ relative mo￾tion during assembly. Compliance plays a critical rol… view at source ↗
Figure 2
Figure 2. Figure 2: Compliance-enabled contact formation: We employ the contact formations for peg-in-hole tasks by following the prior studies [8], [20]. It consists of approach, contact, fit, align, and insertion contact formations. We consider applying a failure recovery strategy in fit, align, and insertion contact formations. III. FRAMEWORK OVERVIEW We aim to robustly and resiliently complete peg-in-hole tasks, even when… view at source ↗
Figure 3
Figure 3. Figure 3: First, a VLM judges success or failure based on the termi￾nal robot arm pose pe and the image Ie. If the VLM judges that the skill execution of the contact formation fails, the VLM-based failure recovery is executed. The VLM analyzes the reasons for failure, suggests recovery skill sequences, and updates goal parameters as necessary [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Vision success check: we applied vision-based pose and angle estimation using OpenCV in fit and align, and image com￾parison that is fed to a VLM in insertion. once it enters the hole, making it difficult for OpenCV to determine whether the operation was successful. Therefore, we provide a goal image and success criteria in the prompt and use a VLM to determine vision-based success. The visual success veri… view at source ↗
Figure 3
Figure 3. Figure 3: Overview of our failure recovery strategy: 1) Success check, where a VLM evaluates the skill execution. 2) Failure recovery planning, where the VLM analyzes the failure and generates recov￾ery plans. A. Success check Vision and pose-based success check: To verify success in the Fit and Align CFs, we apply vision-based analysis using OpenCV. The peg and hole are segmented using HSV color thresholding: the p… view at source ↗
Figure 5
Figure 5. Figure 5: Snapshots of a peg-in-hole trial: The robot failed in the Fit and Align contact formations, but it recovered using our method. The contact formations colored with green and purple above each snapshot describe the successful and failed executions, respectively. TABLE II: Performance under different conditions for each peg shape (Sim) Peg Shape Randomized Conditions Fit Align Insert Succ. Avg. # extra execut… view at source ↗
read the original abstract

Object insertion tasks are prone to failure under pose uncertainty and environmental variation, often requiring manual fine-tuning or controller retraining. We present a novel approach for robust and resilient object insertion using a passively compliant soft wrist that enables safe contact absorption through large deformations, without high-frequency control or force sensing. Our method structures insertion as compliance-enabled contact formations, sequential contact states that progressively constrain degrees of freedom, and integrates automated failure recovery strategies. Our key insight is that wrist compliance permits safe, repeated recovery attempts; hence, we refer to it as compliance-enabled failure recovery. We employ a pre-trained vision-language model (VLM) that assesses each skill execution from terminal poses and images, identifies failure modes, and proposes recovery actions by selecting skills and updating goals. In simulation, our method achieved an 83% success rate, recovering from failures induced by randomized conditions, including grasp misalignments up to 5 degrees, hole-pose errors up to 20 mm, fivefold increases in friction, and unseen square/rectangular pegs, and we further validated the approach on a real robot. Project page is available at https://omron-sinicx.github.io/compliance-enabled-failure-recovery/.

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 proposes a compliance-enabled approach for soft robotic object insertion using a passively compliant soft wrist. Insertion is structured as sequential contact formations that progressively constrain degrees of freedom, with automated failure recovery implemented via a pre-trained vision-language model (VLM) that evaluates terminal poses and images, identifies failure modes, selects skills, and updates goals. The central quantitative claim is an 83% success rate in simulation under randomized perturbations (grasp misalignments up to 5°, hole-pose errors up to 20 mm, fivefold friction increases, unseen square/rectangular pegs), plus real-robot validation.

Significance. If the performance claims are supported by adequate controls and statistics, the work could contribute to resilient soft-robotics methods that exploit passive compliance for safe recovery without force sensing or high-frequency control. The combination of contact-formation sequencing with VLM-driven recovery is a plausible direction for handling uncertainty in insertion tasks.

major comments (2)
  1. [Experimental validation] The 83% success rate is presented without baseline comparisons to prior insertion methods, without stating the total number of trials, and without statistical details (variance, confidence intervals, or success-rate breakdown by perturbation type). This directly limits evaluation of the robustness claim under the listed conditions.
  2. [VLM-based failure recovery] The resilience result is attributed to both compliance-enabled contact formations and the VLM recovery loop, yet no accuracy metrics, precision/recall for failure-mode detection, or fraction of VLM-proposed recoveries that succeed on execution are reported. No ablation disabling the VLM component is provided, making it impossible to isolate the contribution of passive compliance from the recovery strategy.
minor comments (1)
  1. [Abstract] The abstract states 'fivefold increases in friction' without specifying the nominal friction coefficient or the exact range used in the randomized trials.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive comments, which help clarify the presentation of our experimental results and the contributions of individual components. We address each major comment below and indicate revisions to the manuscript.

read point-by-point responses

  1. Referee: [Experimental validation] The 83% success rate is presented without baseline comparisons to prior insertion methods, without stating the total number of trials, and without statistical details (variance, confidence intervals, or success-rate breakdown by perturbation type). This directly limits evaluation of the robustness claim under the listed conditions.

    Authors: We agree that additional experimental details would strengthen the robustness evaluation. The manuscript reports the overall 83% success rate across randomized grasp, pose, friction, and shape variations but does not explicitly detail trial counts or breakdowns. In the revised manuscript we will state the total number of simulation trials, provide a per-perturbation success-rate breakdown, and include variance and confidence-interval statistics. For baseline comparisons we will expand the discussion to reference representative prior insertion methods, highlighting differences in hardware assumptions and control requirements while noting that our passive-compliance approach targets a distinct operating regime. revision: yes

  2. Referee: [VLM-based failure recovery] The resilience result is attributed to both compliance-enabled contact formations and the VLM recovery loop, yet no accuracy metrics, precision/recall for failure-mode detection, or fraction of VLM-proposed recoveries that succeed on execution are reported. No ablation disabling the VLM component is provided, making it impossible to isolate the contribution of passive compliance from the recovery strategy.

    Authors: We acknowledge the importance of quantifying the VLM’s isolated contribution. The VLM assesses terminal poses and images to detect failures and propose skill selections or goal updates, while passive compliance permits safe, repeated contact attempts. In the revision we will report VLM failure-detection accuracy and the fraction of VLM-proposed recoveries that succeed on execution, drawn from our existing experimental logs. An explicit ablation that disables the VLM was not performed, because the method is conceived as an integrated pipeline in which compliance enables the recovery loop; we will instead add a qualitative analysis of representative failure cases to illustrate how the two elements interact. revision: partial

Circularity Check

0 steps flagged

No circularity: empirical validation of compliant insertion method

full rationale

The paper describes an engineering method that structures insertion via compliance-enabled contact formations and uses a pre-trained VLM for failure-mode assessment and recovery selection. The central result is an 83% success rate obtained directly from simulation trials and real-robot validation under randomized perturbations (grasp misalignment, hole-pose error, friction, unseen pegs). No derivation chain, equations, or first-principles predictions are presented that reduce by construction to fitted parameters, self-citations, or renamed inputs. The work is self-contained experimental robotics research whose performance claims rest on measured outcomes rather than any tautological mapping from method definition to reported metric.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The central performance claim rests on the assumption that the off-the-shelf VLM will correctly diagnose failures from single terminal images and poses; no free parameters or new physical entities are introduced beyond the conceptual framing of contact formations.

axioms (1)
  • domain assumption The pre-trained vision-language model can reliably assess terminal poses and images to identify failure modes and propose recovery actions.
    Invoked when the method delegates failure detection and skill selection to the VLM without additional fine-tuning or verification steps described.
invented entities (1)
  • compliance-enabled contact formations no independent evidence
    purpose: To decompose insertion into sequential contact states that progressively constrain degrees of freedom using wrist deformation.
    Conceptual structure introduced to organize the control strategy; no independent physical evidence outside the described experiments is supplied.

pith-pipeline@v0.9.0 · 5759 in / 1428 out tokens · 44720 ms · 2026-05-21T22:43:06.389843+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

38 extracted references · 38 canonical work pages · 2 internal anchors

  1. [1]

    Compare Contact Model-based Control and Contact Model-free Learning: A Survey of Robotic Peg-in-hole Assembly Strategies

    J. Xu, Z. Hou, Z. Liu, and H. Qiao, “Compare contact model-based control and contact model-free learning: A survey of robotic peg-in- hole assembly strategies,”arXiv preprint arXiv:1904.05240, 2019

    work page internal anchor Pith review Pith/arXiv arXiv 1904

  2. [2]

    Autonomous robotic assembly: From part singulation to precise assembly,

    K. Ota, D. K. Jha, S. Jain, B. Yerazunis, R. Corcodel, Y . Shukla, A. Bronars, and D. Romeres, “Autonomous robotic assembly: From part singulation to precise assembly,” inIEEE/RSJ International Con- ference on Intelligent Robots and Systems, 2024, pp. 13 525–13 532

    work page 2024

  3. [3]

    Towards gen- eralized robot assembly through compliance-enabled contact forma- tions,

    A. S. Morgan, Q. Bateux, M. Hao, and A. M. Dollar, “Towards gen- eralized robot assembly through compliance-enabled contact forma- tions,” inIEEE International Conference on Robotics and Automation, 2023, pp. 8010–8016

    work page 2023

  4. [4]

    Peg-in-hole under state uncertainties via a passive wrist joint with push-activate- rotation function,

    T. Nishimura, Y . Suzuki, T. Tsuji, and T. Watanabe, “Peg-in-hole under state uncertainties via a passive wrist joint with push-activate- rotation function,” inIEEE-RAS International Conference on Hu- manoid Robotics, 2017, pp. 67–74

    work page 2017

  5. [5]

    Compliant peg-in-hole assembly using a very soft wrist,

    Q. Zhang, Z. Hu, W. Wan, and K. Harada, “Compliant peg-in-hole assembly using a very soft wrist,”IEEE Robotics and Automation Letters, vol. 9, no. 1, pp. 17–24, 2023

    work page 2023

  6. [6]

    Variable compliance control for robotic peg-in-hole as- sembly: A deep-reinforcement-learning approach,

    C. C. Beltran-Hernandez, D. Petit, I. G. Ramirez-Alpizar, and K. Harada, “Variable compliance control for robotic peg-in-hole as- sembly: A deep-reinforcement-learning approach,”Applied Sciences, vol. 10, no. 19, p. 6923, 2020

    work page 2020

  7. [7]

    Identifying contact formations from sensory patterns and its applicability to robot programming by demon- stration,

    M. Skubic and R. A. V olz, “Identifying contact formations from sensory patterns and its applicability to robot programming by demon- stration,” inIEEE/RSJ International Conference on Intelligent Robots and Systems, vol. 2, 1996, pp. 458–464

    work page 1996

  8. [8]

    A framework for robot manipulation: Skill formalism, meta learning and adaptive con- trol,

    L. Johannsmeier, M. Gerchow, and S. Haddadin, “A framework for robot manipulation: Skill formalism, meta learning and adaptive con- trol,” inIEEE International Conference on Robotics and Automation, 2019, pp. 5844–5850

    work page 2019

  9. [9]

    Learning sequences of ma- nipulation primitives for robotic assembly,

    N. Vuong, H. Pham, and Q.-C. Pham, “Learning sequences of ma- nipulation primitives for robotic assembly,” inIEEE International Conference on Robotics and Automation, 2021, pp. 4086–4092

    work page 2021

  10. [10]

    Predictive learning of error recovery with a sensorized passivity-based soft anthropomorphic hand,

    K. Gilday, T. George-Thuruthel, and F. Iida, “Predictive learning of error recovery with a sensorized passivity-based soft anthropomorphic hand,”Advanced Intelligent Systems, vol. 5, no. 7, p. 2200390, 2023

    work page 2023

  11. [11]

    Real-time anomaly detection and reactive planning with large language models,

    R. Sinha, A. Elhafsi, C. Agia, M. Foutter, E. Schmerling, and M. Pavone, “Real-time anomaly detection and reactive planning with large language models,” inRobotics: Science and Systems, 2024

    work page 2024

  12. [12]

    Racer: Rich language-guided failure recovery policies for imitation learning,

    Y . Dai, J. Lee, N. Fazeli, and J. Chai, “Racer: Rich language- guided failure recovery policies for imitation learning,”arXiv preprint arXiv:2409.14674, 2024

    work page arXiv 2024

  13. [13]

    A compact, cable-driven, activatable soft wrist with six degrees of freedom for assembly tasks,

    F. von Drigalski, K. Tanaka, M. Hamaya, R. Lee, C. Nakashima, Y . Shibata, and Y . Ijiri, “A compact, cable-driven, activatable soft wrist with six degrees of freedom for assembly tasks,” inIEEE/RSJ International Conference on Intelligent Robots and Systems, 2020, pp. 8752–8757

    work page 2020

  14. [14]

    Stretchable materials for robust soft actuators towards assistive wearable devices,

    G. Agarwal, N. Besuchet, B. Audergon, and J. Paik, “Stretchable materials for robust soft actuators towards assistive wearable devices,” Scientific Reports, vol. 6, no. 1, p. 34224, 2016

    work page 2016

  15. [15]

    Robust proprioceptive grasping with a soft robot hand,

    B. S. Homberg, R. K. Katzschmann, M. R. Dogar, and D. Rus, “Robust proprioceptive grasping with a soft robot hand,”Autonomous Robots, vol. 43, no. 3, pp. 681–696, 2019

    work page 2019

  16. [16]

    Soft actuators for real-world applications,

    M. Li, A. Pal, A. Aghakhani, A. Pena-Francesch, and M. Sitti, “Soft actuators for real-world applications,”Nature Reviews Materials, vol. 7, no. 3, pp. 235–249, 2022

    work page 2022

  17. [17]

    Self-healing and damage resilience for soft robotics: A review,

    R. A. Bilodeau and R. K. Kramer, “Self-healing and damage resilience for soft robotics: A review,”Frontiers in Robotics and AI, vol. 4, p. 48, 2017

    work page 2017

  18. [18]

    Learning forceful manipulation skills from multi- modal human demonstrations,

    A. T. Le, M. Guo, N. van Duijkeren, L. Rozo, R. Krug, A. G. Kupcsik, and M. B ¨urger, “Learning forceful manipulation skills from multi- modal human demonstrations,” inIEEE/RSJ International Conference on Intelligent Robots and Systems, 2021, pp. 7770–7777

    work page 2021

  19. [19]

    Learning compliant stiffness by impedance control-aware task segmentation and multi-objective bayesian optimization with priors,

    M. Okada, M. Komatsu, R. Okumura, and T. Taniguchi, “Learning compliant stiffness by impedance control-aware task segmentation and multi-objective bayesian optimization with priors,” inIEEE/RSJ International Conference on Intelligent Robots and Systems, 2023, pp. 8155–8162

    work page 2023

  20. [20]

    Learning robotic assembly tasks with lower dimensional systems by leveraging physical softness and environmen- tal constraints,

    M. Hamaya, R. Lee, K. Tanaka, F. V on Drigalski, C. Nakashima, Y . Shibata, and Y . Ijiri, “Learning robotic assembly tasks with lower dimensional systems by leveraging physical softness and environmen- tal constraints,” inIEEE International Conference on Robotics and Automation, 2020, pp. 7747–7753

    work page 2020

  21. [21]

    Pomdp- guided active force-based search for robotic insertion,

    C. Wang, H. Luo, K. Zhang, H. Chen, J. Pan, and W. Zhang, “Pomdp- guided active force-based search for robotic insertion,” inIEEE/RSJ International Conference on Intelligent Robots and Systems, 2023, pp. 10 668–10 675

    work page 2023

  22. [22]

    Learning insertion primitives with discrete-continuous hybrid action space for robotic assembly tasks,

    X. Zhang, S. Jin, C. Wang, X. Zhu, and M. Tomizuka, “Learning insertion primitives with discrete-continuous hybrid action space for robotic assembly tasks,” inInternational Conference on Robotics and Automation, 2022, pp. 9881–9887

    work page 2022

  23. [23]

    Apricot: Action primitives based on contact-state transition for in-hand tool manipulation,

    D. Saito, A. Kanehira, K. Sasabuchi, N. Wake, J. Takamatsu, H. Koike, and K. Ikeuchi, “Apricot: Action primitives based on contact-state transition for in-hand tool manipulation,” inIEEE-RAS International Conference on Humanoid Robots, 2024, pp. 827–834

    work page 2024

  24. [24]

    Active extrinsic contact sensing: Applica- tion to general peg-in-hole insertion,

    S. Kim and A. Rodriguez, “Active extrinsic contact sensing: Applica- tion to general peg-in-hole insertion,” inInternational Conference on Robotics and Automation, 2022, pp. 10 241–10 247

    work page 2022

  25. [25]

    1 khz behavior tree for self-adaptable tactile insertion,

    Y . Wu, F. Wu, L. Chen, K. Chen, S. Schneider, L. Johannsmeier, Z. Bing, F. J. Abu-Dakka, A. Knoll, and S. Haddadin, “1 khz behavior tree for self-adaptable tactile insertion,” inIEEE International Conference on Robotics and Automation, 2024, pp. 16 002–16 008

    work page 2024

  26. [26]

    Anomaly detection for insertion tasks in robotic assembly using gaussian process models,

    D. Romeres, D. K. Jha, W. Yerazunis, D. Nikovski, and H. A. Dau, “Anomaly detection for insertion tasks in robotic assembly using gaussian process models,” inEuropean Control Conference, 2019, pp. 1017–1022

    work page 2019

  27. [27]

    A multimodal anomaly detector for robot-assisted feeding using an LSTM-based variational autoen- coder,

    D. Park, Y . Hoshi, and C. C. Kemp, “A multimodal anomaly detector for robot-assisted feeding using an LSTM-based variational autoen- coder,”IEEE Robotics and Automation Letters, vol. 3, no. 3, pp. 1544– 1551, 2018

    work page 2018

  28. [28]

    Robot task error recovery using petri nets learned from demonstration,

    G. Chang and D. Kuli ´c, “Robot task error recovery using petri nets learned from demonstration,” inInternational Conference on Advanced Robotics, 2013, pp. 1–6

    work page 2013

  29. [29]

    Learning symbolic failure detection for grasping and mobile manip- ulation tasks,

    P. Hegemann, T. Zechmeister, M. Grotz, K. Hitzler, and T. Asfour, “Learning symbolic failure detection for grasping and mobile manip- ulation tasks,” inIEEE/RSJ International Conference on Intelligent Robots and Systems, 2022, pp. 4302–4309

    work page 2022

  30. [30]

    Recovery RL: Safe reinforcement learning with learned recovery zones,

    B. Thananjeyan, A. Balakrishna, S. Nair, M. Luo, K. Srinivasan, M. Hwang, J. E. Gonzalez, J. Ibarz, C. Finn, and K. Goldberg, “Recovery RL: Safe reinforcement learning with learned recovery zones,”IEEE Robotics and Automation Letters, vol. 6, no. 3, pp. 4915– 4922, 2021

    work page 2021

  31. [31]

    Efficient recovery learning using model predictive meta-reasoning,

    S. Vats, M. Likhachev, and O. Kroemer, “Efficient recovery learning using model predictive meta-reasoning,” inIEEE International Con- ference on Robotics and Automation, 2023, pp. 7258–7264

    work page 2023

  32. [32]

    Learning robust failure response for autonomous vision based flight,

    D. M. Saxena, V . Kurtz, and M. Hebert, “Learning robust failure response for autonomous vision based flight,” inIEEE International Conference on Robotics and Automation, 2017, pp. 5824–5829

    work page 2017

  33. [33]

    Recover: A neuro-symbolic framework for failure detection and recovery,

    C. Cornelio and M. Diab, “Recover: A neuro-symbolic framework for failure detection and recovery,” inIEEE/RSJ International Conference on Intelligent Robots and Systems, 2024, pp. 12 435–12 442

    work page 2024

  34. [34]

    Reflect: Summarizing robot experi- ences for failure explanation and correction,

    Z. Liu, A. Bahety, and S. Song, “Reflect: Summarizing robot experi- ences for failure explanation and correction,” inConference on Robot Learning, 2023, pp. 3468–3484

    work page 2023

  35. [35]

    Self- recovery prompting: Promptable general purpose service robot system with foundation models and self-recovery,

    M. Shirasaka, T. Matsushima, S. Tsunashima, Y . Ikeda, A. Horo, S. Ikoma, C. Tsuji, H. Wada, T. Omija, D. Komukaiet al., “Self- recovery prompting: Promptable general purpose service robot system with foundation models and self-recovery,” inIEEE International Conference on Robotics and Automation, 2024, pp. 17 395–17 402

    work page 2024

  36. [36]

    LLM3: Large language model-based task and motion planning with motion failure reasoning,

    S. Wang, M. Han, Z. Jiao, Z. Zhang, Y . N. Wu, S.-C. Zhu, and H. Liu, “LLM3: Large language model-based task and motion planning with motion failure reasoning,” inIEEE/RSJ International Conference on Intelligent Robots and Systems, 2024, pp. 12 086–12 092

    work page 2024

  37. [37]

    Robotic object insertion with a soft wrist through sim-to-real privi- leged training,

    Y . Fuchioka, C. C. Beltran-Hernandez, H. Nguyen, and M. Hamaya, “Robotic object insertion with a soft wrist through sim-to-real privi- leged training,” inIEEE/RSJ International Conference on Intelligent Robots and Systems, 2024, pp. 9159–9166

    work page 2024

  38. [38]

    robosuite: A Modular Simulation Framework and Benchmark for Robot Learning

    Y . Zhu, J. Wong, A. Mandlekar, R. Mart´ın-Mart´ın, A. Joshi, S. Nasiri- any, and Y . Zhu, “robosuite: A modular simulation framework and benchmark for robot learning,”arXiv preprint arXiv:2009.12293

    work page internal anchor Pith review Pith/arXiv arXiv 2009