pith. sign in

arxiv: 2606.03385 · v1 · pith:ZUIOUPIWnew · submitted 2026-06-02 · 💻 cs.RO · cs.AI

Grasp-Then-Plan with Failure Attribution: A Closed Two-Stage Framework for Precise and Generalizable Robotic Manipulation

Jiahao Xu , Peiyuan Wang , Hanzhuo Zhang , Zihao Yu , Tianyu Fu , Hao Chen , Xuanhao Xiang , Jianbo Yu
show 2 more authors
Chenchen Fu Wanyuan Wang
This is my paper

Pith reviewed 2026-06-28 09:35 UTC · model grok-4.3

classification 💻 cs.RO cs.AI
keywords robotic manipulationgrasp planningmotion planningfailure attributiontwo-stage frameworktask success ratediagnosis-guided optimization
0
0 comments X

The pith

A grasp-then-plan framework with failure attribution raises robotic manipulation success by diagnosing errors separately in each stage.

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

The paper establishes that the tight coupling of grasping and motion planning hides the real cause of failures in robotic tasks, leading to wasteful trial-and-error. It proposes a two-stage GTP-FA method that first produces grasp candidates and then runs motion planning conditioned on a chosen grasp. A failure attribution model is trained on failed trajectories; it generalizes to new grasps and outputs a stable distribution over failure modes. This distribution drives separate fixes: task priors and risk penalties are added to grasping, while planning receives data focused on high-risk starting states. When applied to RL, imitation learning, diffusion policies, and vision-language-action models, the method produces substantially higher task success rates in both simulation and real-robot tests.

Core claim

GTP-FA generates grasp candidates and performs downstream motion planning conditioned on the selected grasp; given failed trajectories it learns a failure attribution model that generalizes to unseen grasps and yields a stable distribution over failure modes, enabling diagnosis-driven optimization that injects task-level priors and risk penalties into grasping while targeting high-risk initial states in planning, thereby improving overall task success rates across multiple base learners.

What carries the argument

The failure attribution model that generalizes to unseen grasps and produces a stable distribution over failure modes to guide separate optimization of grasping and planning stages.

If this is right

  • Grasping modules receive injected task-level priors and risk penalties that suppress unstable or task-incompatible candidates.
  • Planning modules are fine-tuned on data collected from high-risk initial states that expose genuine planning bottlenecks.
  • Base learners from RL, imitation learning, diffusion-policy, and VLA settings each record substantially higher overall task success rates.
  • The same two-stage structure and attribution process operates in both simulation and real-robot experiments.

Where Pith is reading between the lines

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

  • The same attribution-driven separation could be tested on other sequential robotic skills such as pushing followed by grasping.
  • If failure-mode distributions prove stable across environments, the method might reduce the need for full retraining when only one stage changes.
  • Extending the framework to three or more coupled stages would require checking whether attribution remains reliable when more than two modules interact.

Load-bearing premise

The failure attribution model generalizes to unseen grasps and produces a stable distribution over failure modes for diagnosis-guided optimization.

What would settle it

Apply the trained failure attribution model to a set of previously unseen but similar grasps in a held-out manipulation task and check whether the assigned failure-mode distributions remain consistent and correctly predict which stage caused the observed failure; large inconsistency or poor predictive match would falsify the claim that attribution enables reliable separate optimization.

Figures

Figures reproduced from arXiv: 2606.03385 by Chenchen Fu, Hanzhuo Zhang, Hao Chen, Jiahao Xu, Jianbo Yu, Peiyuan Wang, Tianyu Fu, Wanyuan Wang, Xuanhao Xiang, Zihao Yu.

Figure 1
Figure 1. Figure 1: Overview of GTP-FA. Given a task description, GTP-FA performs task-aware grasp [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Failure attribution and diagnostic signal generation. Given a task instruction, visual obser [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Sensitivity analysis of weak failure-attribution thresholds. The left figure reports the [PITH_FULL_IMAGE:figures/full_fig_p013_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Visualization of VLM/LLM-guided task-prior grasp filtering. Each row shows one task, [PITH_FULL_IMAGE:figures/full_fig_p019_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Real-world SoM-guided task-prior grasp selection. For each task, we show SoM Stage [PITH_FULL_IMAGE:figures/full_fig_p020_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: summarizes the final success_at_end performance across the five downstream policy learners and eight ManiSkill3 tasks. The grouped bars compare the original policies, module-level ablations, and the corresponding GTP-FA variants. Overall, GTP-FA achieves the best or competitive final performance in most task–algorithm combinations. The gains are particularly pronounced for data-driven learners such as BC a… view at source ↗
Figure 7
Figure 7. Figure 7: Full success_at_end learning curves across all simulation tasks. Each subfigure corre￾sponds to one downstream learner and reports the performance of five settings: the original policy (00), planning-side-only optimization (01), grasp-side-only optimization (10), naive grasp–plan optimization without attribution (11), and the full GTP-FA variant. Overall, GTP-FA consistently im￾proves terminal success, con… view at source ↗
Figure 8
Figure 8. Figure 8: Full success_once learning curves across all simulation tasks. This metric measures whether the policy reaches a successful state at least once during an episode. Across PPO, SAC, BC, and DP, GTP-FA generally improves or stabilizes success-reaching behavior, and in conjunction with [PITH_FULL_IMAGE:figures/full_fig_p022_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: GTP-FA closed-loop iteration curves across downstream learners. Each subfigure shows [PITH_FULL_IMAGE:figures/full_fig_p024_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: VLA (π0.5) training-loss diagnostics across all simulation tasks. The curves show that VLA fine-tuning is stable across settings. GTP-FA tends to reach lower or faster-converging training loss, while the final evaluation is still determined by task success rather than loss alone. 24 [PITH_FULL_IMAGE:figures/full_fig_p024_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Real-world Franka system used in our experiments. The figure shows the Franka Research [PITH_FULL_IMAGE:figures/full_fig_p025_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Real-robot execution interfaces for the two representative manipulation tasks. Each [PITH_FULL_IMAGE:figures/full_fig_p026_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Raw GraspNet candidates versus the selected unique execution grasp on real observations. [PITH_FULL_IMAGE:figures/full_fig_p027_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Representative successful real-robot rollout of GTP-FA- [PITH_FULL_IMAGE:figures/full_fig_p027_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Representative failure rollout of the original [PITH_FULL_IMAGE:figures/full_fig_p028_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: Representative successful real-robot rollout of GTP-FA- [PITH_FULL_IMAGE:figures/full_fig_p028_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: Representative failure rollout of the original [PITH_FULL_IMAGE:figures/full_fig_p028_17.png] view at source ↗
read the original abstract

In robotic manipulation, the tight coupling between grasping and motion planning often obscures the true source of failure, leading to inefficient trial-and-error. To enable efficient long-horizon manipulation, we propose GTP-FA (Grasp-Then-Plan with Failure Attribution), a task-oriented two-stage grasp-then-plan framework that generates grasp candidates and performs downstream motion planning conditioned on the selected grasp. Given a failed manipulation trajectory, we learn a failure attribution model that generalizes to unseen grasps and produces a stable distribution over failure modes for diagnosis-guided optimization. Based on these attribution results, we then optimize both modules in a diagnosis-driven manner: on the grasping side, we inject task-level priors and risk penalties into grasp candidate scoring and optimization to suppress unstable or task-incompatible grasps; on the planning side, we target high-risk initial states through data collection and fine-tuning to address genuine planning bottlenecks. We evaluate the proposed framework in both simulation and real-robot experiments, and show that GTP-FA improves the corresponding base learners across RL, IL, diffusion-policy, and VLA-based settings, achieving substantially higher overall task success rates.

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 GTP-FA, a closed two-stage grasp-then-plan framework for robotic manipulation. It generates grasp candidates, performs motion planning conditioned on the selected grasp, and learns a failure attribution model from failed trajectories that generalizes to unseen grasps and outputs a distribution over failure modes. This attribution then drives targeted optimization: task-level priors and risk penalties on the grasp side, and data collection/fine-tuning on high-risk initial states for the planner. The central empirical claim is that GTP-FA improves base learners across RL, IL, diffusion-policy, and VLA settings, yielding substantially higher task success rates in both simulation and real-robot experiments.

Significance. If the failure attribution model reliably disentangles grasp-induced versus planning-induced failures, the framework offers a principled alternative to undifferentiated trial-and-error in long-horizon manipulation and could generalize across multiple learning paradigms. The closed-loop diagnosis-driven optimization is a conceptual strength, but its practical value hinges on the empirical demonstration that the attribution step produces stable, actionable distributions without introducing new failure modes.

major comments (2)
  1. [Abstract] Abstract: The central claim that the failure attribution model 'generalizes to unseen grasps and produces a stable distribution over failure modes for diagnosis-guided optimization' is load-bearing for the subsequent grasp-scoring penalties and planner fine-tuning. However, because grasp selection directly determines the initial configuration seen by the planner, grasp-induced instabilities can manifest as apparent planning failures (and vice versa), creating an identifiability risk that is not addressed in the provided description of the attribution model.
  2. [Abstract] The manuscript states that GTP-FA 'improves the corresponding base learners across RL, IL, diffusion-policy, and VLA-based settings, achieving substantially higher overall task success rates,' yet the abstract supplies no quantitative deltas, error bars, number of trials, or ablation controls that would allow verification of the data-to-claim link for this multi-learner improvement.
minor comments (1)
  1. [Abstract] The abstract would be strengthened by a single sentence reporting the magnitude of the reported success-rate gains (e.g., 'from X% to Y%') to give readers an immediate sense of effect size.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments on our manuscript. We address each major comment point by point below and will revise the paper accordingly.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The central claim that the failure attribution model 'generalizes to unseen grasps and produces a stable distribution over failure modes for diagnosis-guided optimization' is load-bearing for the subsequent grasp-scoring penalties and planner fine-tuning. However, because grasp selection directly determines the initial configuration seen by the planner, grasp-induced instabilities can manifest as apparent planning failures (and vice versa), creating an identifiability risk that is not addressed in the provided description of the attribution model.

    Authors: We acknowledge the identifiability concern. The attribution model is trained using labeled trajectories that distinguish grasp failures (via pre-execution metrics such as grasp quality scores and force thresholds) from planning failures (via post-grasp execution deviations). The model's reported generalization to unseen grasps is intended to support this separation. We agree an explicit discussion of the disentanglement approach, its assumptions, and limitations is warranted and will add this to the revised manuscript. revision: yes

  2. Referee: [Abstract] The manuscript states that GTP-FA 'improves the corresponding base learners across RL, IL, diffusion-policy, and VLA-based settings, achieving substantially higher overall task success rates,' yet the abstract supplies no quantitative deltas, error bars, number of trials, or ablation controls that would allow verification of the data-to-claim link for this multi-learner improvement.

    Authors: We agree that quantitative support would strengthen the abstract. In the revision we will incorporate concise numerical results, including average success-rate improvements across the four learning paradigms, trial counts, and error-bar references, subject to abstract length limits. revision: yes

Circularity Check

0 steps flagged

No circularity detected; empirical framework with independent evaluation

full rationale

The paper describes an empirical two-stage robotic manipulation framework (GTP-FA) that learns a failure attribution model from failed trajectories and uses its outputs for diagnosis-guided optimization of grasp scoring and planner fine-tuning. No equations, parameter fits, or derivation steps are presented that reduce any claimed result to its own inputs by construction. No self-citation load-bearing steps, uniqueness theorems, or ansatzes are invoked. Improvements are demonstrated via simulation and real-robot experiments across RL, IL, diffusion, and VLA baselines, rendering the central claims externally falsifiable and self-contained against benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review supplies no equations, datasets, or modeling choices, so the ledger is empty.

pith-pipeline@v0.9.1-grok · 5766 in / 1099 out tokens · 18716 ms · 2026-06-28T09:35:29.518423+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

57 extracted references · 24 canonical work pages · 7 internal anchors

  1. [1]

    Langley , title =

    P. Langley , title =. Proceedings of the 17th International Conference on Machine Learning (ICML 2000) , address =. 2000 , pages =

    2000

  2. [2]

    T. M. Mitchell. The Need for Biases in Learning Generalizations. 1980

    1980

  3. [3]

    M. J. Kearns , title =

  4. [4]

    Machine Learning: An Artificial Intelligence Approach, Vol. I. 1983

    1983

  5. [5]

    R. O. Duda and P. E. Hart and D. G. Stork. Pattern Classification. 2000

    2000

  6. [6]

    Suppressed for Anonymity , author=

  7. [7]

    Newell and P

    A. Newell and P. S. Rosenbloom. Mechanisms of Skill Acquisition and the Law of Practice. Cognitive Skills and Their Acquisition. 1981

    1981

  8. [8]

    A. L. Samuel. Some Studies in Machine Learning Using the Game of Checkers. IBM Journal of Research and Development. 1959

    1959

  9. [10]

    Efficient online reinforcement learning with offline data

    Philip J Ball, Laura Smith, Ilya Kostrikov, and Sergey Levine. Efficient online reinforcement learning with offline data. In International Conference on Machine Learning, pages 1577--1594. PMLR, 2023

    2023

  10. [15]

    Graspnet-1billion: A large-scale benchmark for general object grasping

    Hao-Shu Fang, Chenxi Wang, Minghao Gou, and Cewu Lu. Graspnet-1billion: A large-scale benchmark for general object grasping. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 11444--11453, 2020

    2020

  11. [17]

    Copa: General robotic manipulation through spatial constraints of parts with foundation models

    Haoxu Huang, Fanqi Lin, Yingdong Hu, Shengjie Wang, and Yang Gao. Copa: General robotic manipulation through spatial constraints of parts with foundation models. In 2024 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 9488--9495. IEEE, 2024

    2024

  12. [20]

    $\pi^{*}_{0.6}$: a VLA That Learns From Experience

    Physical Intelligence, Ali Amin, Raichelle Aniceto, Ashwin Balakrishna, Kevin Black, Ken Conley, Grace Connors, James Darpinian, Karan Dhabalia, Jared DiCarlo, et al. ^ * _ 0.6 : A VLA that learns from experience. arXiv preprint arXiv:2511.14759, 2025 a

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  13. [21]

    $\pi_{0.5}$: a Vision-Language-Action Model with Open-World Generalization

    Physical Intelligence, Kevin Black, Noah Brown, James Darpinian, Karan Dhabalia, Danny Driess, Adnan Esmail, Michael Equi, Chelsea Finn, Niccolo Fusai, et al. _ 0.5 : A vision--language--action model with open-world generalization. arXiv preprint arXiv:2504.16054, 2025 b

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  14. [27]

    Generalizing 6-dof grasp detection via domain prior knowledge

    Haoxiang Ma, Modi Shi, Boyang Gao, and Di Huang. Generalizing 6-dof grasp detection via domain prior knowledge. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 18102--18111, 2024

    2024

  15. [28]

    Liv: Language-image representations and rewards for robotic control

    Yecheng Jason Ma, Vikash Kumar, Amy Zhang, Osbert Bastani, and Dinesh Jayaraman. Liv: Language-image representations and rewards for robotic control. In International Conference on Machine Learning, pages 23301--23320. PMLR, 2023

    2023

  16. [30]

    6-dof graspnet: Variational grasp generation for object manipulation

    Arsalan Mousavian, Clemens Eppner, and Dieter Fox. 6-dof graspnet: Variational grasp generation for object manipulation. In Proceedings of the IEEE/CVF international conference on computer vision, pages 2901--2910, 2019

    2019

  17. [33]

    Roboclip: One demonstration is enough to learn robot policies

    Sumedh Sontakke, Jesse Zhang, S \'e b Arnold, Karl Pertsch, Erdem B y k, Dorsa Sadigh, Chelsea Finn, and Laurent Itti. Roboclip: One demonstration is enough to learn robot policies. Advances in Neural Information Processing Systems, 36: 0 55681--55693, 2023

    2023

  18. [34]

    Task-oriented grasp prediction with visual-language inputs

    Chao Tang, Dehao Huang, Lingxiao Meng, Weiyu Liu, and Hong Zhang. Task-oriented grasp prediction with visual-language inputs. In 2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 4881--4888. IEEE, 2023

    2023

  19. [35]

    Foundationgrasp: Generalizable task-oriented grasping with foundation models

    Chao Tang, Dehao Huang, Wenlong Dong, Ruinian Xu, and Hong Zhang. Foundationgrasp: Generalizable task-oriented grasping with foundation models. IEEE Transactions on Automation Science and Engineering, 2025

    2025

  20. [38]

    Grasp as you say: Language-guided dexterous grasp generation

    Yi-Lin Wei, Jian-Jian Jiang, Chengyi Xing, Xian-Tuo Tan, Xiao-Ming Wu, Hao Li, Mark Cutkosky, and Wei-Shi Zheng. Grasp as you say: Language-guided dexterous grasp generation. Advances in Neural Information Processing Systems, 37: 0 46881--46907, 2024

    2024

  21. [39]

    Catgrasp: Learning category-level task-relevant grasping in clutter from simulation

    Bowen Wen, Wenzhao Lian, Kostas Bekris, and Stefan Schaal. Catgrasp: Learning category-level task-relevant grasping in clutter from simulation. In 2022 International Conference on Robotics and Automation (ICRA), pages 6401--6408. IEEE, 2022

    2022

  22. [44]

    arXiv preprint arXiv:2512.13380 , year=

    Universal Dexterous Functional Grasping via Demonstration-Editing Reinforcement Learning , author=. arXiv preprint arXiv:2512.13380 , year=

    work page arXiv

  23. [45]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Generalizing 6-dof grasp detection via domain prior knowledge , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  24. [46]

    IEEE Transactions on Automation Science and Engineering , year=

    Foundationgrasp: Generalizable task-oriented grasping with foundation models , author=. IEEE Transactions on Automation Science and Engineering , year=

  25. [47]

    Advances in Neural Information Processing Systems , volume=

    Grasp as you say: Language-guided dexterous grasp generation , author=. Advances in Neural Information Processing Systems , volume=

  26. [48]

    2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) , pages=

    Task-oriented grasp prediction with visual-language inputs , author=. 2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) , pages=. 2023 , organization=

    2023

  27. [49]

    arXiv preprint arXiv:2509.01746 , year=

    Fail2Progress: Learning from Real-World Robot Failures with Stein Variational Inference , author=. arXiv preprint arXiv:2509.01746 , year=

    work page arXiv

  28. [50]

    Aha: A vision-language-model for detecting and reasoning over failures in robotic manipulation.arXiv preprint arXiv:2410.00371, 2024

    Aha: A vision-language-model for detecting and reasoning over failures in robotic manipulation , author=. arXiv preprint arXiv:2410.00371 , year=

    work page arXiv

  29. [51]

    FailSafe: Reasoning and Recovery from Failures in Vision-Language-Action Models

    Failsafe: Reasoning and recovery from failures in vision-language-action models , author=. arXiv preprint arXiv:2510.01642 , year=

    work page internal anchor Pith review arXiv

  30. [52]

    2022 International Conference on Robotics and Automation (ICRA) , pages=

    Catgrasp: Learning category-level task-relevant grasping in clutter from simulation , author=. 2022 International Conference on Robotics and Automation (ICRA) , pages=. 2022 , organization=

    2022

  31. [53]

    Proceedings of the IEEE/CVF international conference on computer vision , pages=

    6-dof graspnet: Variational grasp generation for object manipulation , author=. Proceedings of the IEEE/CVF international conference on computer vision , pages=

  32. [54]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Graspnet-1billion: A large-scale benchmark for general object grasping , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  33. [55]

    arXiv preprint arXiv:2211.02647 , year=

    Neural grasp distance fields for robot manipulation , author=. arXiv preprint arXiv:2211.02647 , year=

    work page arXiv

  34. [56]

    arXiv preprint arXiv:2510.25889 , year=

    RL: Online rl fine-tuning for flow-based visionlanguage-action models , author=. arXiv preprint arXiv:2510.25889 , year=

    work page arXiv

  35. [57]

    arXiv preprint arXiv:2512.05107 , year=

    STARE-VLA: Progressive Stage-Aware Reinforcement for Fine-Tuning Vision-Language-Action Models , author=. arXiv preprint arXiv:2512.05107 , year=

    work page arXiv

  36. [58]

    $\pi_0$: A Vision-Language-Action Flow Model for General Robot Control

    _0 : A Vision-Language-Action Flow Model for General Robot Control , author=. arXiv preprint arXiv:2410.24164 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  37. [59]

    Data Scaling Laws in Imitation Learning for Robotic Manipulation

    Data scaling laws in imitation learning for robotic manipulation , author=. arXiv preprint arXiv:2410.18647 , year=

    work page internal anchor Pith review arXiv

  38. [60]

    arXiv preprint arXiv:2503.01837 , year=

    Multi-Stage Manipulation with Demonstration-Augmented Reward, Policy, and World Model Learning , author=. arXiv preprint arXiv:2503.01837 , year=

    work page arXiv

  39. [61]

    Intelligence, Physical and Amin, Ali and Aniceto, Raichelle and Balakrishna, Ashwin and Black, Kevin and Conley, Ken and Connors, Grace and Darpinian, James and Dhabalia, Karan and DiCarlo, Jared and others , journal=. ^

  40. [62]

    Intelligence, Physical and Black, Kevin and Brown, Noah and Darpinian, James and Dhabalia, Karan and Driess, Danny and Esmail, Adnan and Equi, Michael and Finn, Chelsea and Fusai, Niccolo and others , journal=. _

  41. [63]

    arXiv preprint arXiv:2405.03379 , year=

    Reverse forward curriculum learning for extreme sample and demonstration efficiency in reinforcement learning , author=. arXiv preprint arXiv:2405.03379 , year=

    work page arXiv

  42. [64]

    International Conference on Machine Learning , pages=

    Efficient online reinforcement learning with offline data , author=. International Conference on Machine Learning , pages=. 2023 , organization=

    2023

  43. [65]

    IEEE Robotics and Automation Letters , volume=

    Centergrasp: Object-aware implicit representation learning for simultaneous shape reconstruction and 6-dof grasp estimation , author=. IEEE Robotics and Automation Letters , volume=. 2024 , publisher=

    2024

  44. [66]

    2024 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) , pages=

    Copa: General robotic manipulation through spatial constraints of parts with foundation models , author=. 2024 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) , pages=. 2024 , organization=

    2024

  45. [67]

    arXiv preprint arXiv:2503.15035 , year=

    GraspCorrect: Robotic Grasp Correction via Vision-Language Model-Guided Feedback , author=. arXiv preprint arXiv:2503.15035 , year=

    work page arXiv

  46. [68]

    arXiv preprint arXiv:2305.04639 , year=

    Multimodal Detection and Identification of Robot Manipulation Failures , author=. arXiv preprint arXiv:2305.04639 , year=

    work page arXiv

  47. [69]

    RoboMD: Uncovering Robot Vulnerabilities through Semantic Potential Fields

    From Mystery to Mastery: Failure Diagnosis for Improving Manipulation Policies , author=. arXiv preprint arXiv:2412.02818 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  48. [70]

    Unpacking Failure Modes of Generative Policies: Runtime Monitoring of Consistency and Progress, 2024

    Unpacking failure modes of generative policies: Runtime monitoring of consistency and progress , author=. arXiv preprint arXiv:2410.04640 , year=

    work page arXiv

  49. [71]

    arXiv preprint arXiv:2209.03855 , year=

    Se (3)-diffusionfields: Learning smooth cost functions for joint grasp and motion optimization through diffusion , author=. arXiv preprint arXiv:2209.03855 , year=

    work page arXiv

  50. [72]

    Advances in Neural Information Processing Systems , volume=

    Roboclip: One demonstration is enough to learn robot policies , author=. Advances in Neural Information Processing Systems , volume=

  51. [73]

    arXiv preprint arXiv:2502.20630 , year=

    Subtask-Aware Visual Reward Learning from Segmented Demonstrations , author=. arXiv preprint arXiv:2502.20630 , year=

    work page arXiv

  52. [74]

    International Conference on Machine Learning , pages=

    Liv: Language-image representations and rewards for robotic control , author=. International Conference on Machine Learning , pages=. 2023 , organization=

    2023

  53. [75]

    arXiv preprint arXiv:2406.11815 , year=

    Llarva: Vision-action instruction tuning enhances robot learning , author=. arXiv preprint arXiv:2406.11815 , year=

    work page arXiv

  54. [76]

    arXiv preprint arXiv:2503.01616 , year=

    Robodexvlm: Visual language model-enabled task planning and motion control for dexterous robot manipulation , author=. arXiv preprint arXiv:2503.01616 , year=

    work page arXiv

  55. [77]

    arXiv preprint arXiv:2412.13630 , year=

    Policy decorator: Model-agnostic online refinement for large policy model , author=. arXiv preprint arXiv:2412.13630 , year=

    work page arXiv

  56. [78]

    arXiv preprint arXiv:2408.02912 , year=

    Koi: Accelerating online imitation learning via hybrid key-state guidance , author=. arXiv preprint arXiv:2408.02912 , year=

    work page arXiv

  57. [79]

    Set-of-Mark Prompting Unleashes Extraordinary Visual Grounding in GPT-4V

    Set-of-mark prompting unleashes extraordinary visual grounding in gpt-4v , author=. arXiv preprint arXiv:2310.11441 , year=

    work page internal anchor Pith review Pith/arXiv arXiv