pith. sign in

arxiv: 2606.18328 · v1 · pith:BBTU4FWPnew · submitted 2026-06-16 · 💻 cs.RO

Recover, Discover, Plan: Learning Skills and Concepts from Robot Failures

Bowen Li , Mayank Mishra , Y. Isabel Liu , Stone Tao , Nishanth Kumar , Alexander G. Gray , Ruwan Wickramarachchi , Jonathan Francis
show 2 more authors
Sebastian Scherer Tom Silver
This is my paper

Pith reviewed 2026-06-27 00:11 UTC · model grok-4.3

classification 💻 cs.RO
keywords robot learningfailure recoveryrelational predicatesabstract planningreinforcement learningsim-to-real transferstate abstraction
0
0 comments X

The pith

ReSYNC turns robot failure recoveries into relational predicates that support abstract planning for unseen long-horizon 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 method that lets a robot first use reinforcement learning to recover from failures encountered in training tasks. It then analyzes those recoveries to discover and refine relational predicates that abstract the state space. These predicates are added to a planning model so the robot can avoid similar failures in new, longer problems without retraining separate policies for each failure mode. The approach alternates between skill learning and concept learning in an incremental loop. This process converts specific recovery behaviors into general failure-avoidance strategies that transfer from simulation to real robots.

Core claim

ReSYNC performs an incremental dual-learning process in which reinforcement learning acquires recovery skills from observed failures and a subsequent concept-learning stage discovers new relational predicates that explain and generalize those recoveries. The discovered predicates are incorporated into an abstract planning model that the robot uses at test time to solve previously unseen long-horizon tasks. Across four simulated domains the method solves more problems than strong baselines and achieves over 50 percent higher success rates; the same learned abstractions also support sim-to-real transfer on non-prehensile manipulation tasks.

What carries the argument

ReSYNC's incremental dual-learning loop that alternates RL-based recovery skill acquisition with predicate discovery and refinement to expand an abstraction library for abstract planning.

If this is right

  • The robot solves long-horizon problems that were never seen in training by using the expanded predicate library for abstract planning.
  • No separate recovery policy needs to be trained for each distinct failure mode.
  • The same learned predicates enable generalization to unseen scenarios in both simulation and real-robot non-prehensile manipulation.
  • Continual expansion of the abstraction library improves performance over time without restarting from scratch.

Where Pith is reading between the lines

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

  • Manual engineering of state abstractions for planning could be reduced if the discovery process scales to richer sensor data.
  • The same loop might be applied to domains where failures are expensive, by first learning recoveries in simulation and then transferring the resulting predicates.
  • Incorrect predicates could propagate planning errors unless an explicit verification step is added later.

Load-bearing premise

The discovered relational predicates will be sufficiently accurate and complete to support reliable planning on tasks outside the specific failures seen during training.

What would settle it

A new long-horizon test task whose solution requires a predicate that was never discovered or refined during the training failures, causing the abstract planner to produce an incorrect or incomplete plan.

Figures

Figures reproduced from arXiv: 2606.18328 by Alexander G. Gray, Bowen Li, Jonathan Francis, Mayank Mishra, Nishanth Kumar, Ruwan Wickramarachchi, Sebastian Scherer, Stone Tao, Tom Silver, Y. Isabel Liu.

Figure 1
Figure 1. Figure 1: Overview of ReSYNC. The learning in ReSYNC is driven by failures in training tasks, [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overview of the running example (Clut￾tered Drawer). We show the initial knowledge avail￾able to the robot, each stage of the curriculum, and example generalization tests. Training Curriculum: A user-specified cur￾riculum presents related tasks in stages to elicit failures. Let T train i be the i th stage with Ni = 10 tasks in our experiments. We assume the robot could detect failure states (e.g., simulato… view at source ↗
Figure 3
Figure 3. Figure 3: The ReSYNC framework. Top: ReSYNC alternates between recovery skill learning and concept discovery in each stage. Bottom left: Skill learning uses successful replanning as the recovery reward. Bottom right: Dreaming generates trajectories for predicate and operator learning. object before opening the drawer in a new context. Thus, beyond skills, the robot must also learn concepts that enable reasoning over… view at source ↗
Figure 4
Figure 4. Figure 4: Visualization of the three simulated domains (excluding Cluttered Drawer) and the two [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Hardware setup and ReSYNC pipeline in the real-world experiments. During training [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Learning efficiency comparison. Thanks to relational abstractions and planning, ReSYNC reduces training episodes by ∼ 22×. Finetuning Across Scenarios: As demon￾strated in [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Example VLM-OS prompt for the Blocked Stacking domain. The prompt includes the [PITH_FULL_IMAGE:figures/full_fig_p018_7.png] view at source ↗
Figure 9
Figure 9. Figure 9: Representative failures in our real-robot experiments. [PITH_FULL_IMAGE:figures/full_fig_p023_9.png] view at source ↗
Figure 8
Figure 8. Figure 8: Planning efficiency compari￾son. ReSYNC learns compact relational concepts that minimize the planning ob￾jective and yield single-plan execution, while DSG-M [5] relies solely on ter￾minal predicates, resulting in inefficient planning. ReSYNC achieves substantially higher planning efficiency than prior work such as DSG-M [5]. As shown in Fig￾ure 8, DSG-M learns predicates that primarily character￾ize termi… view at source ↗
read the original abstract

Intelligent robots should not only recover from failures, but also acquire the abstract knowledge needed to avoid them in the future. While reinforcement learning (RL) can learn reactive recovery behaviors, training a separate policy for every distinct failure mode is highly inefficient. We introduce Recovery-Driven Synthesis of Relational Concepts (ReSYNC), the first approach that progressively discovers and refines state abstractions (relational predicates) from failure-recovery experience to support abstract planning. Unlike purely reactive methods, ReSYNC jointly learns skills and concepts through an incremental dual-learning process. In the skill-learning phase, the robot uses RL to learn to recover from failures seen in training tasks. In the concept-learning phase, the robot discovers new relational predicates and refines its abstract planning model to explain and generalize the learned recovery behaviors. This interaction enables ReSYNC to convert local recoveries seen during training into global failure avoidance at test time. Across four simulated domains, we show that ReSYNC's ability to continually expand and refine its abstraction library allows it to solve long-horizon, previously unseen problems, outperforming strong baselines by over 50%. Additionally, we demonstrate sim-to-real transfer of ReSYNC, where it performs real-world non-prehensile manipulation skills and generalizes to unseen scenarios through abstract planning. Overall, ReSYNC represents a significant step toward robots that autonomously acquire abstractions for scalable, failure-aware planning in the physical world.

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

0 major / 2 minor

Summary. The paper introduces ReSYNC, an incremental dual-learning framework in which RL is used to acquire recovery skills from observed failures and a concept-learning phase discovers and refines relational predicates that support abstract planning. The interaction between the two phases is claimed to convert local recoveries into global failure-avoidance policies that solve long-horizon, previously unseen tasks. Empirical results are reported across four simulated domains (outperforming strong baselines by more than 50 %) together with a sim-to-real demonstration of non-prehensile manipulation and generalization to unseen scenarios.

Significance. If the empirical claims are substantiated, the work would constitute a meaningful contribution to integrated learning and planning in robotics by showing how failure-recovery experience can be turned into reusable relational abstractions without hand-crafted predicates. The reported performance margin and the sim-to-real transfer are concrete strengths that would be of interest to the robotics community.

minor comments (2)
  1. [Abstract] The abstract states that ReSYNC 'outperforms strong baselines by over 50 %' but does not name the baselines, the performance metric, or the statistical test used; the results section of the full manuscript should supply these details.
  2. [Abstract] The description of the concept-learning phase does not indicate how predicate accuracy or sufficiency is verified before the predicates are added to the planning model; if this validation procedure is described later in the paper it should be cross-referenced from the abstract.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for their careful summary of ReSYNC and for noting the potential contribution of turning failure-recovery experience into reusable relational abstractions. The recommendation is listed as uncertain, but the report contains no enumerated major comments. We therefore provide no point-by-point responses below. If the referee has additional specific concerns, we are happy to address them in a revision.

Circularity Check

0 steps flagged

No significant circularity

full rationale

The paper presents an empirical method (ReSYNC) combining RL-based skill recovery with predicate discovery for abstract planning. No equations, derivations, or fitted parameters are described in the provided text that reduce a claimed prediction or result to its own inputs by construction. The performance claims rest on experimental outcomes across domains rather than any self-referential theoretical step. No self-citation load-bearing arguments or uniqueness theorems are invoked in the abstract or description. The derivation chain is self-contained as an algorithmic process whose validity is tested externally via simulation and real-world transfer.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

No technical details are supplied in the abstract, so no free parameters, axioms, or invented entities can be identified.

pith-pipeline@v0.9.1-grok · 5807 in / 1222 out tokens · 36265 ms · 2026-06-27T00:11:52.731326+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

59 extracted references · 23 canonical work pages · 11 internal anchors

  1. [1]

    Javed and R

    K. Javed and R. S. Sutton. The Big World Hypothesis and Its Ramifications for Artificial Intelligence. InFinding the Frame: An RLC Workshop for Examining Conceptual Frameworks, 2024

    2024

  2. [2]

    S. Vats, M. Likhachev, and O. Kroemer. Efficient Recovery Learning using Model Predictive Meta-Reasoning. InProceedings of the International Conference on Robotics and Automation (ICRA), pages 7258–7264, 2023

    2023

  3. [3]

    S. Vats, D. K. Jha, M. Likhachev, O. Kroemer, and D. Romeres. Recoverychaining: Learning Local Recovery Policies for Robust Manipulation.arXiv preprint arXiv:2410.13979, 2024

    work page arXiv 2024

  4. [4]

    Bagaria, J

    A. Bagaria, J. K. Senthil, and G. Konidaris. Skill Discovery for Exploration and Planning using Deep Skill Graphs. InProceedings of International Conference on Machine Learning (ICML), pages 521–531. PMLR, 2021

    2021

  5. [5]

    Bagaria, A

    A. Bagaria, A. D. M. Koch, R. Rodriguez-Sanchez, S. Lobel, and G. Konidaris. Intrinsically Motivated Discovery of Temporally Abstract Graph-based Models of the World. InProceedings of Reinforcement Learning Conference, 2025

    2025

  6. [6]

    A. Li, N. Kumar, T. Lozano-P´erez, and L. P. Kaelbling. Learning to Bridge the Gap: Efficient Novelty Recovery with Planning and Reinforcement Learning. InNeurIPS 2024 Workshop on Open-World Agents, 2024

    2024

  7. [7]

    Silver, R

    T. Silver, R. Chitnis, N. Kumar, W. McClinton, T. Lozano-P ´erez, L. Kaelbling, and J. B. Tenenbaum. Predicate Invention for Bilevel Planning. InProceedings of The AAAI Conference on Artificial Intelligence (AAAI), volume 37, pages 12120–12129, 2023

    2023

  8. [8]

    N. Shah, J. Nagpal, P. Verma, and S. Srivastava. From Reals to Logic and Back: Inventing Symbolic V ocabularies, Actions and Models for Planning from Raw Data.arXiv preprint arXiv:2402.11871, 2024. URLhttps://arxiv.org/pdf/2402.11871v2

    work page arXiv 2024

  9. [9]

    B. Li, T. Silver, S. Scherer, and A. Gray. Bilevel Learning for Bilevel Planning. InProceedings of the Robotics: Science And Systems (RSS), 2025

    2025

  10. [10]

    Silver, A

    T. Silver, A. Athalye, J. B. Tenenbaum, T. Lozano-P´erez, and L. P. Kaelbling. Learning Neuro- Symbolic Skills for Bilevel Planning. InProceedings of the Conference on Robot Learning (CoRL), 2022

    2022

  11. [11]

    Y . I. Liu, B. Li, B. Eysenbach, and T. Silver. SLAP: Shortcut Learning for Abstract Planning. arXiv preprint arXiv:2511.01107, 2025

    work page arXiv 2025

  12. [12]

    Y . S. Shao, Y . Zheng, S. Sun, P. Chaudhari, V . Kumar, and N. Figueroa. Symskill: Symbol and Skill Co-Invention for Data-Efficient and Real-Time Long-Horizon Manipulation.arXiv preprint arXiv:2510.01661, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  13. [13]

    D. Abel, D. Arumugam, L. Lehnert, and M. Littman. State abstractions for lifelong reinforce- ment learning. InInternational Conference on Machine Learning. PMLR, 2018

    2018

  14. [14]

    Eysenbach, A

    B. Eysenbach, A. Gupta, J. Ibarz, and S. Levine. Diversity is all you need: Learning skills without a reward function. InInternational Conference on Learning Representations, 2019

    2019

  15. [15]

    R. K. Nayyar and S. Srivastava. Autonomous Option Invention for Continual Hierarchical Reinforcement Learning and Planning. InProceedings of the AAAI Conference on Artificial Intelligence (AAAI), volume 39, pages 19642–19650, 2025

    2025

  16. [16]

    Shukla, S

    A. Shukla, S. Tao, and H. Su. Maniskill-hab: A benchmark for low-level manipulation in home rearrangement tasks. InProceedings of the International Conference on Learning Representations (ICLR), 2025. 10

    2025

  17. [17]

    P. W. Battaglia, J. B. Hamrick, V . Bapst, A. Sanchez-Gonzalez, V . Zambaldi, M. Malinowski, A. Tacchetti, D. Raposo, A. Santoro, R. Faulkner, et al. Relational Inductive Biases, Deep Learning, and Graph Networks.arXiv preprint arXiv:1806.01261, 2018

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  18. [18]

    Vaswani, N

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, L. Kaiser, and I. Polosukhin. Attention is All You Need. InProceedings of the Advances in Neural Information Processing Systems (NeurIPS), volume 30, 2017

    2017

  19. [19]

    GPT-5.4 Thinking System Card, Mar

    OpenAI. GPT-5.4 Thinking System Card, Mar. 2026. URL https://openai.com/index/ gpt-5-4-thinking-system-card/

    2026

  20. [20]

    L. P. Kaelbling and T. Lozano-P ´erez. Hierarchical Task and Motion Planning in the Now. InProceedings of the International Conference on Robotics and Automation (ICRA), pages 1470–1477. IEEE, 2011

    2011

  21. [21]

    C. R. Garrett, T. Lozano-P´erez, and L. P. Kaelbling. Pddlstream: Integrating Symbolic Planners and Blackbox Samplers via Optimistic Adaptive Planning. InProceedings of the International Conference on Automated Planning and Scheduling (ICAPS), volume 30, pages 440–448, 2020

    2020

  22. [22]

    Chitnis, T

    R. Chitnis, T. Silver, J. B. Tenenbaum, T. Lozano-P´erez, and L. P. Kaelbling. Learning Neuro- Symbolic Relational Transition Models for Bilevel Planning. InProceedings of the IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 4166–4173, 2022

    2022

  23. [23]

    Kumar, T

    N. Kumar, T. Silver, W. McClinton, L. Zhao, S. Proulx, T. Lozano-P´erez, L. P. Kaelbling, and J. Barry. Practice Makes Perfect: Planning To Learn Skill Parameter Policies. InProceedings of the Robotics: Science And Systems (RSS), 2024

    2024

  24. [24]

    Konidaris, L

    G. Konidaris, L. P. Kaelbling, and T. Lozano-P´erez. From skills to symbols: Learning symbolic representations for abstract high-level planning.Journal of Artificial Intelligence Research (JAIR), 2018

    2018

  25. [25]

    Q. Wang, B. Li, Z. Luo, Y . Xu, A. Gray, T. Silver, S. Scherer, K. Sycara, and Y . Xie. Unifying Deep Predicate Invention with Pre-trained Foundation Models.arXiv preprint arXiv:2512.17992, 2025

    work page arXiv 2025

  26. [26]

    R. S. Sutton, D. Precup, and S. Singh. Between MDPs and Semi-MDPs: A Framework for Temporal Abstraction in Reinforcement Learning.Artificial intelligence, 1999

    1999

  27. [27]

    Bacon, J

    P.-L. Bacon, J. Harb, and D. Precup. The option-critic architecture. InProceedings of the AAAI Conference on Artificial Intelligence (AAAI), volume 31, 2017

    2017

  28. [28]

    Zhang and S

    S. Zhang and S. Whiteson. DAC: The Double Actor-Critic Architecture for Learning Options. In Proceedings of the Advances in Neural Information Processing Systems (NeurIPS), volume 32, 2019

    2019

  29. [29]

    S. Moon, J. Yeom, B. Park, and H. O. Song. Discovering Hierarchical Achievements in Reinforcement Learning via Contrastive Learning. InProceedings of the Advances in Neural Information Processing Systems (NeurIPS), volume 36, pages 63674–63686, 2023

    2023

  30. [30]

    Eysenbach, T

    B. Eysenbach, T. Zhang, S. Levine, and R. R. Salakhutdinov. Contrastive Learning as Goal- Conditioned Reinforcement Learning. InProceedings of the Advances in Neural Information Processing Systems (NeurIPS), volume 35, pages 35603–35620, 2022

    2022

  31. [31]

    Eysenbach, R

    B. Eysenbach, R. Salakhutdinov, and S. Levine. Search on the Replay Buffer: Bridging Planning and Reinforcement Learning. InProceedings of the Advances in Neural Information Processing Systems (NeurIPS), volume 32, 2019

    2019

  32. [32]

    Semi-parametric Topological Memory for Navigation

    N. Savinov, A. Dosovitskiy, and V . Koltun. Semi-Parametric Topological Memory for Naviga- tion.arXiv preprint arXiv:1803.00653, 2018. 11

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  33. [33]

    F. Yang, D. Lyu, B. Liu, and S. Gustafson. PEORL: Integrating Symbolic Planning and Hierarchical Reinforcement Learning for Robust Decision-Making. InProceedings of the International Joint Conference on Artificial Intelligence (IJCAI), pages 4860–4866, 2018

    2018

  34. [34]

    Integrating Task-Motion Planning with Reinforcement Learning for Robust Decision Making in Mobile Robots

    Y . Jiang, F. Yang, S. Zhang, and P. Stone. Integrating Task-Motion Planning with Reinforcement Learning for Robust Decision Making in Mobile Robots.arXiv preprint arXiv:1811.08955, 2018

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  35. [35]

    Sarathy, D

    V . Sarathy, D. Kasenberg, S. Goel, J. Sinapov, and M. Scheutz. Spotter: Extending Symbolic Planning Operators through Targeted Reinforcement Learning.arXiv preprint arXiv:2012.13037, 2020

    work page arXiv 2012

  36. [36]

    S. Goel, Y . Shukla, V . Sarathy, M. Scheutz, and J. Sinapov. Rapid-learn: A Framework for Learning to Recover for Handling Novelties in Open-world Environments. In2022 IEEE International Conference on Development and Learning (ICDL), pages 15–22. IEEE, 2022

    2022

  37. [37]

    Aeronautiques, A

    C. Aeronautiques, A. Howe, C. Knoblock, I. D. McDermott, A. Ram, M. Veloso, D. Weld, D. W. Sri, A. Barrett, D. Christianson, et al. Pddl—the planning domain definition language. Technical Report, Tech. Rep., 1998

    1998

  38. [38]

    M. Helmert. The Fast Downward Planning System.Journal of Artificial Intelligence Research, 26:191–246, 2006

    2006

  39. [39]

    Proximal Policy Optimization Algorithms

    J. Schulman, F. Wolski, P. Dhariwal, A. Radford, and O. Klimov. Proximal Policy Optimization Algorithms.arXiv preprint arXiv:1707.06347, 2017

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  40. [40]

    KinDER: A Physical Reasoning Benchmark for Robot Learning and Planning

    Y . Huang, B. Li, V . Saxena, Y . Liang, U. A. Mishra, L. Ji, L. Zha, J. Wu, N. Kumar, S. Scherer, et al. KinDER: A Physical Reasoning Benchmark for Robot Learning and Planning.arXiv preprint arXiv:2604.25788, 2026

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  41. [41]

    P. Yin, T. Westenbroek, Z. Zhang, J. Tran, I. Dagnino, E. Shilamkar, N. Mbiziwo-Tiapo, S. Bagaria, X. Liu, G. Mullins, A. Kolobov, and A. Gupta. Emergent Dexterity via Diverse Resets and Large-Scale Reinforcement Learning. InThe Fourteenth International Conference on Learning Representations, 2026

    2026

  42. [42]

    Bjelonic, F

    F. Bjelonic, F. Tischhauser, and M. Hutter. Towards Bridging the Gap: Systematic Sim-to-Real Transfer for Diverse Legged Robots.arXiv preprint arXiv:2509.06342, 2025

    work page arXiv 2025

  43. [43]

    Solving Rubik's Cube with a Robot Hand

    I. Akkaya, M. Andrychowicz, M. Chociej, M. Litwin, B. McGrew, A. Petron, A. Paino, M. Plappert, G. Powell, R. Ribas, et al. Solving Rubik’s Cube with a Robot Hand.arXiv preprint arXiv:1910.07113, 2019

    work page internal anchor Pith review Pith/arXiv arXiv 1910

  44. [44]

    C. Chi, S. Feng, Y . Du, Z. Xu, E. Cousineau, B. Burchfiel, and S. Song. Diffusion Policy: Visuomotor Policy Learning via Action Diffusion. InRobotics: Science and Systems (RSS), 2023

    2023

  45. [45]

    SAM 3: Segment Anything with Concepts

    N. Carion, L. Gustafson, Y .-T. Hu, S. Debnath, R. Hu, D. Suris, C. Ryali, K. V . Alwala, H. Khedr, A. Huang, et al. SAM 3: Segment Anything with Concepts.arXiv preprint arXiv:2511.16719, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  46. [46]

    P. Liu, L. P. Kaelbling, J. B. Tenenbaum, and J. Mao. Lifelong Experience Abstraction and Planning. InICML 2025 Workshop on Programmatic Representations for Agent Learning, 2025

    2025

  47. [47]

    M. Fu, J. Yu, K. El-Refai, E. Kou, H. Xue, H. Huang, W. Xiao, G. Wang, F.-F. Li, G. Shi, et al. CaP-X: A Framework for Benchmarking and Improving Coding Agents for Robot Manipulation. arXiv preprint arXiv:2603.22435, 2026. 12

    work page internal anchor Pith review arXiv 2026

  48. [48]

    Zabounidis, Y

    R. Zabounidis, Y . Wu, S. Stepputtis, W. Kim, Y . Li, T. Mitchell, and K. Sycara. SCALAR: Learn- ing and Composing Skills through LLM Guided Symbolic Planning and Deep RL Grounding. arXiv preprint arXiv:2603.09036, 2026

    work page arXiv 2026

  49. [49]

    J. Luo, T. Ding, K. H. R. Chan, H. Min, C. Callison-Burch, and R. Vidal. Concept Lancet: Image Editing with Compositional Representation Transplant. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 28502–28512, 2025

    2025

  50. [50]

    J. Luo, J. Yang, T. Neiman, L. Fan, B. Yin, S. Tran, M. Shah, and R. Vidal. Dictionary-Aligned Concept Control for Safeguarding Multimodal LLMs.arXiv preprint arXiv:2604.08846, 2026

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  51. [51]

    Athalye, N

    A. Athalye, N. Kumar, T. Silver, Y . Liang, T. Lozano-P´erez, and L. P. Kaelbling. Predicate Invention from Pixels via Pretrained Vision-Language Models.arXiv preprint arXiv:2501.00296, 2024

    work page arXiv 2024

  52. [52]

    Kumar, W

    N. Kumar, W. Shen, F. Ramos, D. Fox, T. Lozano-P´erez, L. P. Kaelbling, and C. R. Garrett. Open-World Task and Motion Planning via Vision-Language Model Generated Constraints. IEEE Robotics and Automation Letters (RA-L), 11:3366–3373, 2026. URL https://arxiv. org/abs/2411.08253

    work page arXiv 2026

  53. [53]

    J. Duan, W. Pumacay, N. Kumar, Y . R. Wang, S. Tian, W. Yuan, R. Krishna, D. Fox, A. Man- dlekar, and Y . Guo. AHA: A Vision-Language-Model for Detecting and Reasoning Over Failures in Robotic Manipulation. InProceedings of the International Conference on Learning Representations (ICLR), 2025. URLhttps://arxiv.org/abs/2410.00371

    work page arXiv 2025

  54. [54]

    S. Wang, M. Han, Z. Jiao, Z. Zhang, Y . N. Wu, S.-C. Zhu, and H. Liu. Llm 3:large lan- guage model-based task and motion planning with motion failure reasoning.arXiv preprint arXiv:2403.11552, 2024. URLhttps://arxiv.org/pdf/2403.11552

    work page arXiv 2024

  55. [55]

    C. Xu, T. K. Nguyen, E. Dixon, C. Rodriguez, P. Miller, R. Lee, P. Shah, R. Ambrus, H. Nishimura, and M. Itkina. Can we detect failures without failure data? uncertainty-aware runtime failure detection for imitation learning policies. InProceedings of the Robotics: Science And Systems (RSS), 2025

    2025

  56. [56]

    Z. Yang, B. Hedegaard, A. Jaafar, Y . Wei, S. Thompson, S. S. Raman, H. Fu, S. Tellex, G. Konidaris, D. Paulius, et al. SkillWrapper: Generative Predicate Invention for Skill Abstrac- tion.arXiv preprint arXiv:2511.18203, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  57. [57]

    Z. Yang, J. Mao, Y . Du, J. Wu, J. B. Tenenbaum, T. Lozano-P ´erez, and L. P. Kaelbling. Compositional Diffusion-Based Continuous Constraint Solvers. InConference on Robot Learning (CoRL), 2023. URLhttps://arxiv.org/pdf/2309.00966.pdf. 13 Table 4: Important notations used in this work. Symbol Description x∈ XContinuous world state; state space xo Feature ...

    work page arXiv 2023

  58. [58]

    One block is initially placed near a box corner, causing the nominal insertion behavior to fail

    Cornered Insertion-S.A UR7e arm with a Robotiq 2F-140 gripper must insert one block onto another to form a tower. One block is initially placed near a box corner, causing the nominal insertion behavior to fail. Train and test tasks differ in goal specification. • State and actions:States are SE(3) object poses, estimated from wrist-camera scans using SAM3...

  59. [59]

    overfitting

    Cornered Insertion-L.This domain uses the same hardware setup but adds a lid and a drawer, requiring the robot to place the completed tower into the drawer. Although it shares the recovery skill with Cornered Insertion-S, its task structure leads to different discovered concepts and planning operators. •State and actions:Same as Cornered Insertion-S. •Obj...