Pith. sign in

REVIEW 5 major objections 6 minor 57 references

Few-Shot Neuro-Symbolic Imitation Learning for Long-Horizon Planning and Acting

T0 review · 5 major / 6 minor · reviewed 2026-08-05 · deepseek-v4-flash

Pith's one-line read The paper claims a robot can jointly learn continuous control policies and the symbolic planning domain of a task from as few as five raw skill demonstrations, with no predefined states, predicates, or hand-coded domain knowledge.

desk verdict A genuinely useful integration of ASP-based symbolic learning with diffusion-policy skills, and the zero-shot Hanoi transfer is the right kind of evidence, but the thin quantitative reporting and the 'five demonstrations' claim make the current version a strong workshop paper rather than a finished contribution. read the letter →

arxiv 2508.21501 v1 pith:L5PCTEIK submitted 2025-08-29 cs.RO

classification cs.RO
keywords neuro-symbolicimitationlearningtaskandmotionplanningsymbolicabstractionPDDLanswersetprogrammingdiffusionpolicylong-horizonmanipulationfew-shot
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper's central claim is that long-horizon robotic tasks can be learned end-to-end from a handful of raw skill demonstrations without any predefined symbolic vocabulary: the system jointly discovers low-level continuous controllers and the high-level PDDL domain needed to plan them. Each demonstration is treated as an edge between two black-box high-level states that a human matches by looking at before/after images, and an Answer Set Programming solver finds the minimal symbolic domain consistent with the resulting graph. A learned oracle filters each controller's observation down to the objects the symbolic operator actually changes, expressed relative to the end-effector, so diffusion policies can be trained from very few trajectories. The authors report that with about five demonstrations per skill the approach solves Robosuite tasks (Stacking, Kitchen, Nut Assembly), a Towers of Hanoi implementation, and forklift pallet tasks, and that the learned Hanoi domain transfers zero-shot from 3x3 to 7x5 instances. The manuscript's Section 7.1 acknowledges that success depends on demonstration quality, accurate grasp-pose estimation, and the oracle preserving the Markov property.

What carries the argument

The load-bearing object is the labeled transition graph of black-box states: nodes are high-level states identified only by visual snapshots, edges are human-labeled skill transitions. Bisimulation minimization removes redundant nodes, then an ASP solver treats the graph as a specification and searches for the minimal PDDL domain whose state graph is isomorphic to it. Around that core, the oracle phi = alpha composed with gamma filters each skill's observation space to the objects its symbolic operator changes and re-expresses their coordinates relative to the end-effector, and diffusion policies provide the continuous low-level controllers. The PDDL domain connects the two levels: a classic

What would settle it

Take the same Towers of Hanoi demonstrations and deliberately corrupt one state snapshot match (or remove one node from the graph), rerun the full pipeline, and test on the 7x5 configuration: if the planner still solves all instances, the framework tolerates annotation noise; if it fails or produces an invalid plan, the completeness/noise-free assumption is doing the load-bearing work.

Watch

Extended reading notes

Core claim

The discovery the paper is trying to establish is that symbolic abstraction and neural control can be learned jointly from sparse demonstrations, removing the usual requirement for manually designed predicates, object types, or lexicons. The pipeline maps each raw skill trajectory to a labeled node transition (n, l, n') with visual snapshots; humans only decide whether two snapshots show the same high-level state. A bisimulation quotient compacts the graph, and an ASP solver recovers the simplest first-order planning domain (expressed in PDDL) whose state graph is isomorphic to the observed one. Each discovered operator is then paired with a diffusion policy trained on the corresponding demo

Load-bearing premise

The load-bearing premise is that the human-annotated graph of state transitions is complete and noise-free enough that the ASP solver's minimal symbolic domain is the true generalizable domain; miss a key state or mismatch a snapshot and the learned PDDL model will be wrong for unseen configurations.

Editorial extensions

If this is right

  • Long-horizon manipulation no longer needs hand-coded symbolic models: a few demonstrations plus visual state matching can produce a planning domain and the controllers to execute it.
  • The learned symbolic abstraction supports zero-shot generalization to larger instances and spatial shifts, since planning happens in compact PDDL space rather than in raw observation space.
  • Adding new skills, object types, or predicates can be done by extending the graph with a few nodes and demonstrations, enabling curriculum-style fine-tuning.
  • Because the plans are symbolic, a human can inspect and correct the robot's high-level reasoning before any execution.
  • The approach is domain-agnostic: the same pipeline works for Cartesian manipulator arms and for an articulated forklift with rear-wheel steering.

Reading between the lines

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

  • Editorial inference: the human visual-matching bottleneck is the main remaining supervision cost; the paper names vision-language models as a possible automation, but the data-efficiency claim depends on how cheap this matching is in practice.
  • Editorial inference: 'five demonstrations' counts skill demonstrations, not full task traces, so the total human effort for a long-horizon task scales with the number of distinct skills and states the graph needs.
  • Editorial inference: the ASP solver's minimality criterion is a simplicity prior; if the same transition graph admits two equally simple PDDL theories with different generalization behavior, the method would silently pick the wrong one, so broader zero-shot tests across multiple domains would stress this.
  • Editorial inference: the oracle assumes the symbolic operator identifies all task-relevant objects; tasks where relevance depends on continuous or unobserved quantities (e.g., force, temperature, occlusion) would need a richer symbolic vocabulary than a sparse graph can invent.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

5 major / 6 minor

Summary. The paper proposes a neuro-symbolic imitation learning framework that learns, from a small number of skill demonstrations, both a PDDL-style symbolic domain and the low-level neural policies needed to execute plans. Demonstrations are first converted into a labeled transition graph via human state matching and skill labels; the graph is minimized by bisimulation and passed to an ASP solver to obtain a minimal symbolic domain. A learned planner produces abstract plans, and each operator is realized by a diffusion policy trained on demonstration segments, with an oracle filtering observations and expressing them relative to the end effector. The authors claim state-of-the-art data efficiency (as few as five demonstrations), zero-shot transfer to larger Hanoi instances and new forklift tasks, and interpretable symbolic plans. The method is evaluated in six simulated domains.

Significance. If the empirical claims are substantiated, the framework is a meaningful step toward joint learning of symbolic abstractions and continuous controllers from few demonstrations, without hand-designed predicates or lexicons. The paper's concrete strengths are the explicit pipeline from raw trajectories to bisimulation-minimized graphs to ASP-discovered PDDL domains, the oracle-based observation filtering, and the interpretable generated domains shown in Appendix A.1. The use of a classical planner and modular skills also gives a credible route to long-horizon generalization. However, the manuscript does not provide the quantitative evidence needed to support its central claims, and the symbolic-generalization argument rests on an identifiability assumption that is stated but not justified. The paper ships no code or machine-checked proofs, so the reported results are not independently reproducible from the text alone.

major comments (5)
  1. [Section 6, Figs. 4–5] The central empirical claim is that N-S achieves high data efficiency and strong zero-/few-shot generalization. However, Section 5 states that each agent is evaluated over 30 episodes with 5 seeds, but Section 6 and Figs. 4–5 report no numerical success rates, standard deviations, confidence intervals, or per-seed breakdowns. Phrases such as 'near 100% success' and 'baselines completely fail' are qualitative. Without quantitative tables or error bars, the comparisons and the generalization results cannot be assessed. This is load-bearing because the paper's contribution is primarily empirical.
  2. [Section 5, Baselines] The comparison to baselines is not apples-to-apples. The N-S method uses a high-level symbolic planner and an oracle that filters observations and provides relative coordinates; the H-IL and IL baselines do not have access to this planner or to the oracle filtering γ. The paragraph on 'fair access' only describes observation-coordinate differences, not the presence of the planner/oracle. Consequently, the reported success of N-S may reflect the privileged planner/oracle rather than the jointly learned symbolic abstraction. To support the claim that both layers are learned, the paper should include an ablation in which H-IL receives the same ground-truth subgoals, or a version of N-S without the symbolic planner, so that each component's contribution can be isolated.
  3. [Section 4.1, last paragraph] The identifiability of the ASP-minimal PDDL domain is assumed but not established. The paragraph states 'We assume graphs are complete and noise-free.' Since the graph is built from human visual snapshot matching and label assignment, completeness is not guaranteed by construction. Any finite labeled graph is consistent with infinitely many PDDL domains that disagree on unseen states; the ASP solver returns the simplest one. The only evidence that the minimal model matches the true Hanoi domain is the single zero-shot transfer in Fig. 5 (3×3 to 7×5), which is presented without error bars. The limitation statement in Section 7.1 invokes prior work [14] on ASP robustness to noise; that robustness concerns a different pipeline and does not cover errors introduced by snapshot matching. The paper should provide an identifiability argument or, at minimum, an additional validation (e.g., held-o
  4. [Problem Formulation and Appendix A.5] The paper frames the input as 'raw skill demonstrations' and says 'Human input is limited to two forms' (Section 3), but the experiments use demonstrations generated by a 'hand-coded automated script' with injected Gaussian noise (Appendix A.5), not human demonstrations. This is a mismatch with the claimed human-taught setting. The manuscript should either clarify that the reported results are on scripted stand-ins and discuss how human demonstrations would affect snapshot matching and graph completeness, or report experiments with actual human demonstrations. This matters because the human annotation step is one of the paper's stated contributions.
  5. [Introduction, Contributions; Related Work] The statement 'no prior work jointly learns low-level control policies and high-level planning models from a few demonstrations without relying on predefined symbolic states...' is not substantiated relative to the cited reference [50], whose title is 'Neuro-symbolic imitation learning: Discovering symbolic abstractions for skill learning.' Unless the authors show that [50] requires predefined symbolic states or differs in a concrete technical way, the 'first' claim is unverified. Please add an explicit comparison of assumptions, inputs, and mechanisms with [50], or soften the claim.
minor comments (6)
  1. [Fig. 4 caption] There is a typo: 'Nut Assemly' should be 'Nut Assembly'.
  2. [Appendix A.5] The phrase '0 secondes' should be '0 seconds'.
  3. [Algorithm 1, line 5] The notation Doi is used in 'Cluster Action Steps(πi, Doi)' but is not formally defined before this line. Please define it in Section 4.
  4. [Section 4.2, Eq. (1)–(2)] The oracle function ϕ is clear in principle, but the text does not state how the relevant-object set E_oi is computed from the ASP-discovered operator. Is it derived from predicate changes in the operator's effects? Please specify the procedure.
  5. [Section 5] The action step clustering threshold is never given a value or a sensitivity analysis. Since it is a free parameter, the paper should report the chosen threshold and, ideally, an ablation.
  6. [General reproducibility] No code, dataset, or detailed hyperparameter tables beyond Appendix A.5 are provided. A public release of the demonstration-generation scripts, learned PDDL domains, and evaluation harness would be necessary to reproduce the reported results.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the symbolic model is a fit to the demonstration graph, but every predictive claim is tested on held-out configurations, and the core ASP method is externally cited.

full rationale

The paper's derivation chain is: raw demonstrations -> human-matched transition graph -> minimal bisimulation -> ASP solver produces a PDDL domain whose state graph is isomorphic to the input graph -> oracle filtering derived from the learned operators -> diffusion policies -> planning/execution. The only step that could superficially look circular is the ASP extraction: by construction the learned domain is required to reproduce the input graph, so the 'discovered rules' are a compression of the demonstrations rather than an independent derivation. However, the paper does not present that reproduction as a prediction; it presents it as the learning step. The actual predictive claims are the zero-shot Hanoi transfers from 3x3 training to 4x3/5x4/7x5, spatial peg shifts, and new forklift pallet configurations, all of which are evaluated on instances not used to build the graph. These held-out results give external grounding and break any fitted-input/prediction equivalence. The oracle is derived from the same learned symbols used to train the controllers, but it is a feature-selection/preprocessing mechanism; whether it helps generalization is an empirical matter, and the paper tests it on novel configurations rather than only on the training graph. Self-citations ([33,34,35,43]) appear only in related-work attribution and do not supply the load-bearing machinery; the ASP framework is credited to the external works [13,14]. The paper explicitly states the completeness/noise-free assumption in Sec. 4.1 and, in Sec. 7.1, points to external robustness results [14] rather than to its own prior work. No equation or definition in the paper has the form 'X is defined in terms of Y' such that a claimed prediction reduces to its own input. Thus, under the quoted-equation standard required here, no circular step can be exhibited, and the appropriate score is 0.

Assumptions & free parameters 1 free parameters · 4 assumptions · 0 invented entities

The method introduces no new physical entities; the 'oracle' is a filtering function, and the 'black-box nodes' are graph abstractions. The main assumptions are domain-level rather than invented entities.

free parameters (1)
  • action_step_clustering_threshold
    The number of action steps per skill is determined by 'consistent sequential action space patterns' (Sec. 4.1), with no quantitative threshold given; this choice affects policy learning and is not derived from first principles.
assumptions (4)
  • domain assumption The input graph G of state transitions, built from human-annotated demonstrations, is complete and noise-free.
    Assumed in Sec. 4.1 ('We assume graphs are complete and noise-free') and needed for the ASP solver to return a valid model; violations are not handled and would invalidate the learned domain.
  • domain assumption The ASP solver (Bonet et al. 2020, Rodriguez et al. 2021) always yields a minimal first-order representation when one exists, and that representation is correct for unseen instances.
    The paper relies entirely on [13,14] for symbolic abstraction and generalizes to larger Hanoi instances under this expectation, without proving that the learned 'greater' constraint generalizes.
  • domain assumption The oracle filter phi(s_tilde, o_i) = alpha ∘ gamma(s_tilde, o_i) preserves a Markovian state for the low-level policies.
    Stated in Sec. 7.1 as a limitation: 'our method relies on an oracle to simplify skill learning, assuming the symbolic solver abstracts information at a sufficiently low level to maintain the Markov property.' If violated, policies fail.
  • domain assumption Diffusion policies can represent the necessary action distributions from few demonstrations.
    Adopted from Chi et al. [55]; the few-shot performance depends on this sample efficiency, which is not proven for all domains.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Few-Shot Neuro-Symbolic Imitation Learning for Long-Horizon Planning and Acting." pith.science (2026). https://pith.science/paper/L5PCTEIK

@misc{pith2026250821501,
  author       = {Pith},
  title        = {Pith review of: Few-Shot Neuro-Symbolic Imitation Learning for Long-Horizon Planning and Acting},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/L5PCTEIK}},
  note         = {Machine review of arXiv:2508.21501}
}
read the original abstract

Imitation learning enables intelligent systems to acquire complex behaviors with minimal supervision. However, existing methods often focus on short-horizon skills, require large datasets, and struggle to solve long-horizon tasks or generalize across task variations and distribution shifts. We propose a novel neuro-symbolic framework that jointly learns continuous control policies and symbolic domain abstractions from a few skill demonstrations. Our method abstracts high-level task structures into a graph, discovers symbolic rules via an Answer Set Programming solver, and trains low-level controllers using diffusion policy imitation learning. A high-level oracle filters task-relevant information to focus each controller on a minimal observation and action space. Our graph-based neuro-symbolic framework enables capturing complex state transitions, including non-spatial and temporal relations, that data-driven learning or clustering techniques often fail to discover in limited demonstration datasets. We validate our approach in six domains that involve four robotic arms, Stacking, Kitchen, Assembly, and Towers of Hanoi environments, and a distinct Automated Forklift domain with two environments. The results demonstrate high data efficiency with as few as five skill demonstrations, strong zero- and few-shot generalizations, and interpretable decision making.

Figures

Figures reproduced from arXiv: 2508.21501 by the authors.

Figure 1
Figure 1. Our neuro-symbolic framework integrates graph construction, symbolic abstraction, plan [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Shown here is an example demonstration of the MOVE operator in the Towers of Hanoi [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Illustrations of some of the simulation domains used for evaluation. [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Performance comparison between our Neuro-Symbolic (N-S) framework and baseline methods. Our approach achieves high success rates on short—Stacking & Forklift Pallet Load￾ing/Unloading—and long-horizon tasks—including Towers of Hanoi, Multiple Pallets Storage, Nut Assem…
Figure 5
Figure 5. Figure 5: Zero- and few-shot generalization results on Different Hanoi Towers configurations. [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]
Figure 6
Figure 6. Figure 6: Illustration of the Nut Assembly envi￾ronment in Robosuite [PITH_FULL_IMAGE:figures/full_fig_p015_6.png]
Figure 8
Figure 8. Figure 8: Task graphs for different environments. Stacking and Forklift Load/Unload Pallets graphs consist only of two nodes and a connecting edge, as they consider one-skill tasks and do not require planning. 18 [PITH_FULL_IMAGE:figures/full_fig_p018_8.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

57 extracted references · 38 canonical work pages

  1. [14]

    I. D. Rodriguez, B. Bonet, J. Romero, and H. Geffner. Learning First-Order Representations for Planning from Black Box States: New Results. In Proceedings of the 18th International Conference on Principles of Knowledge Representation and Reasoning , pages 539–548, 11

  2. [50]

    Keller, D

    L. Keller, D. Tanneberg, and J. Peters. Neuro-symbolic imitation learning: Discovering sym- bolic abstractions for skill learning. (arXiv:2503.21406), Mar. 2025. doi:10.48550/arXiv.2503. 21406. URL http://arxiv.org/abs/2503.21406. arXiv:2503.21406 [cs]

  3. [1]

    Hussein, M

    A. Hussein, M. M. Gaber, E. Elyan, and C. Jayne. Imitation learning: A survey of learning methods. ACM Comput. Surv., 50(2):21:1–21:35, Apr. 2017. ISSN 0360-0300. doi:10.1145/ 3054912

  4. [2]

    T. Osa, J. Pajarinen, G. Neumann, J. Bagnell, P. Abbeel, and J. Peters. An algorithmic per- spective on imitation learning. Foundations and Trends in Robotics , 7:1–179, Nov. 2018. doi:10.1561/2300000053

  5. [3]

    B. Fang, S. Jia, D. Guo, M. Xu, S. Wen, and F. Sun. Survey of imitation learning for robotic manipulation. International Journal of Intelligent Robotics and Applications , 3, Dec. 2019. doi:10.1007/s41315-019-00103-5

  6. [4]

    Imitation Learning: Progress, Taxonomies and Challenges

    B. Zheng, S. Verma, J. Zhou, I. Tsang, and F. Chen. Imitation learning: Progress, taxonomies and challenges. (arXiv:2106.12177), Oct. 2022. doi:10.48550/arXiv.2106.12177. URL http: //arxiv.org/abs/2106.12177. arXiv:2106.12177 [cs]

  7. [5]

    M. Zare, P. M. Kebria, A. Khosravi, and S. Nahavandi. A survey of imitation learning: Al- gorithms, recent developments, and challenges. IEEE Transactions on Cybernetics , 54(12): 7173–7186, Dec. 2024. ISSN 2168-2275. doi:10.1109/TCYB.2024.3395626

  8. [6]

    Toward the Fundamental Limits of Imitation Learning

    N. Rajaraman, L. F. Yang, J. Jiao, and K. Ramachandran. Toward the fundamental limits of imitation learning. (arXiv:2009.05990), Sept. 2020. doi:10.48550/arXiv.2009.05990. URL http://arxiv.org/abs/2009.05990. arXiv:2009.05990 [cs]

Show all 57 references
  1. [7]

    T. Xu, Z. Li, and Y . Yu. Error bounds of imitating policies and environments. (arXiv:2010.11876), Oct. 2020. doi:10.48550/arXiv.2010.11876. URL http://arxiv.org/ abs/2010.11876. arXiv:2010.11876 [cs]

  2. [8]

    Wolfe, B

    J. Wolfe, B. Marthi, and S. Russell. Combined task and motion planning for mobile manipu- lation. Proceedings of the International Conference on Automated Planning and Scheduling , 20:254–257, May 2010. ISSN 2334-0843. doi:10.1609/icaps.v20i1.13436

  3. [9]

    L. P. Kaelbling and T. Lozano-P ´erez. Hierarchical task and motion planning in the now. In 2011 IEEE International Conference on Robotics and Automation , page 1470–1477, May 2011. doi:10.1109/ICRA.2011.5980391. URL https://ieeexplore.ieee.org/ abstract/document/5980391/

  4. [10]

    C. R. Garrett, R. Chitnis, R. Holladay, B. Kim, T. Silver, L. P. Kaelbling, and T. Lozano- P´erez. Integrated task and motion planning. Annual Review of Control, Robotics, and Autonomous Systems , 4(V olume 4, 2021):265–293, May 2021. ISSN 2573-5144. doi: 10.1146/annurev-contr...

  5. [11]

    Arora, H

    A. Arora, H. Fiorino, D. Pellier, M. M ´etivier, and S. Pesty. A review of learning planning action models. The Knowledge Engineering Review , 33:e20, 2018. ISSN 0269-8889, 1469-

  6. [12]

    Konidaris, L

    G. Konidaris, L. P. Kaelbling, and T. Lozano-Perez. From skills to symbols: Learning symbolic representations for abstract high-level planning.Journal of Artificial Intelligence Research, 61: 215–289, Jan. 2018. ISSN 1076-9757. doi:10.1613/jair.5575

  7. [13]

    Blai and G

    B. Blai and G. Hector. Learning First-Order Symbolic Representations for Planning from the Structure of the State Space . IOS Press, 2020. doi:10.3233/FAIA200361. URL https://www.medra.org/servlet/aliasResolver?alias=iospressISBN&isbn= 978-1-64368-100-9&spage=2322&doi=10.3233/...

  8. [15]

    Silver, R

    T. Silver, R. Chitnis, J. Tenenbaum, L. P. Kaelbling, and T. Lozano-Perez. Learning sym- bolic operators for task and motion planning. In 2021 IEEE/RSJ International Confer- ence on Intelligent Robots and Systems (IROS) , page 3182–3189, Prague, Czech Republic, Sept. 2021. IEE...

  9. [16]

    Verma, S

    P. Verma, S. R. Marpally, and S. Srivastava. Discovering user-interpretable capabilities of black-box planning agents. (arXiv:2107.13668), May 2022. URL http://arxiv.org/abs/ 2107.13668. arXiv:2107.13668 [cs]

  10. [18]

    Ahmetoglu, M

    A. Ahmetoglu, M. Y . Seker, J. Piater, E. Oztop, and E. Ugur. Deepsym: Deep symbol generation and rule learning from unsupervised continuous robot interaction for planning. Journal of Artificial Intelligence Research , 75:709–745, Nov. 2022. ISSN 1076-9757. doi: 10.1613/jair.1...

  11. [19]

    Silver, R

    T. Silver, R. Chitnis, N. Kumar, W. McClinton, T. Lozano-Perez, L. P. Kaelbling, and J. Tenen- baum. Predicate invention for bilevel planning. (arXiv:2203.09634), Nov. 2022. URL http://arxiv.org/abs/2203.09634. arXiv:2203.09634 [cs]

  12. [20]

    Li and T

    A. Li and T. Silver. Embodied active learning of relational state abstractions for bilevel planning. (arXiv:2303.04912), June 2023. doi:10.48550/arXiv.2303.04912. URL http: //arxiv.org/abs/2303.04912. arXiv:2303.04912 [cs]

  13. [21]

    Kumar, W

    N. Kumar, W. McClinton, R. Chitnis, T. Silver, T. Lozano-P´erez, and L. P. Kaelbling. Learning efficient abstract planning models that choose what to predict. (arXiv:2208.07737), Sept. 2023. URL http://arxiv.org/abs/2208.07737. arXiv:2208.07737 [cs]

  14. [22]

    Iii and W

    M. Iii and W. B. Learning Compositional Abstract Models Incrementally for Efficient Bilevel Task and Motion Planning. Thesis, Massachusetts Institute of Technology, Feb. 2024. URL https://dspace.mit.edu/handle/1721.1/153869. Accepted: 2024-03-21T19:12:11Z

  15. [23]

    N. Shah, J. Nagpal, P. Verma, and S. Srivastava. From reals to logic and back: Inventing symbolic vocabularies, actions, and models for planning from raw data. (arXiv:2402.11871), Mar. 2024. doi:10.48550/arXiv.2402.11871. URL http://arxiv.org/abs/2402.11871. arXiv:2402.11871 [cs]

  16. [24]

    Umili, E

    E. Umili, E. Antonioni, F. Riccio, R. Capobianco, D. Nardi, and G. D. Giacomo. Learning a symbolic planning domain through the interaction with continuous environments. 2021

  17. [25]

    F. Yang, D. Lyu, B. Liu, and S. Gustafson. Peorl: Integrating symbolic planning and hierarchi- cal reinforcement learning for robust decision-making. (arXiv:1804.07779), June 2018. URL http://arxiv.org/abs/1804.07779. arXiv:1804.07779 [cs, stat]

  18. [26]

    Illanes, X

    L. Illanes, X. Yan, R. T. Icarte, and S. A. McIlraith. Symbolic plans as high-level instruc- tions for reinforcement learning. Proceedings of the International Conference on Automated Planning and Scheduling, 30:540–550, June 2020. ISSN 2334-0843. doi:10.1609/icaps.v30i1. 6750. 11

  19. [27]

    Kokel, A

    H. Kokel, A. Manoharan, S. Natarajan, B. Ravindran, and P. Tadepalli. Reprel: Integrating relational planning and reinforcement learning for effective abstraction. Proceedings of the International Conference on Automated Planning and Scheduling , 31:533–541, May 2021. ISSN 233...

  20. [28]

    L. Guan, S. Sreedharan, and S. Kambhampati. Leveraging approximate symbolic models for reinforcement learning via skill diversity. 2022

  21. [30]

    Cheng and D

    S. Cheng and D. Xu. League: Guided skill learning and abstraction for long-horizon manipu- lation. IEEE Robotics and Automation Letters, 8(10):6451–6458, Oct. 2023. ISSN 2377-3766. doi:10.1109/LRA.2023.3308061

  22. [31]

    Silver, A

    T. Silver, A. Athalye, J. B. Tenenbaum, T. Lozano-P´erez, and L. P. Kaelbling. Learning neuro- symbolic skills for bilevel planning. In Proceedings of The 6th Conference on Robot Learn- ing, page 701–714. PMLR, Mar. 2023. URL https://proceedings.mlr.press/v205/ silver23a.html

  23. [32]

    Acharya, W

    K. Acharya, W. Raza, C. Dourado, A. Velasquez, and H. Song. Neurosymbolic reinforcement learning and planning: A survey. IEEE Transactions on Artificial Intelligence, PP:1–14, Sept

  24. [33]

    Lorang, S

    P. Lorang, S. Goel, Y . Shukla, P. Zips, and M. Scheutz. A framework for neurosymbolic goal- conditioned continual learning in open world environments. In 2024 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) , page 12070–12077, Oct. 2024. doi: 10.110...

  25. [34]

    open world

    P. Lorang, H. Horvath, T. Kietreiber, P. Zips, C. Heitzinger, and M. Scheutz. Adapting to the “open world”: The utility of hybrid hierarchical reinforcement learning and symbolic planning. In 2024 IEEE International Conference on Robotics and Automation (ICRA) , page 508–514, ...

  26. [35]

    S. Goel, P. Lymperopoulos, R. Thielstrom, E. Krause, P. Feeney, P. Lorang, S. Schneider, Y . Wei, E. Kildebeck, S. Goss, M. C. Hughes, L. Liu, J. Sinapov, and M. Scheutz. A neu- rosymbolic cognitive architecture framework for handling novelties in open worlds. Artificial Intel...

  27. [36]

    doi:10.1109/TAI.2023.3311428

  28. [37]

    Chitnis, T

    R. Chitnis, T. Silver, J. B. Tenenbaum, T. Lozano-Perez, and L. P. Kaelbling. Learning neuro- symbolic relational transition models for bilevel planning. arXiv preprint arXiv:2105.14074, 2021

  29. [38]

    R. T. Icarte, T. Q. Klassen, R. A. Valenzano, and S. A. McIlraith. Reward machines: Exploiting reward function structure in reinforcement learning. JAIR, 73:173–208, 2020

  30. [39]

    Sarathy, D

    V . Sarathy, D. Kasenberg, S. Goel, J. Sinapov, and M. Scheutz. Spotter: Extending symbolic planning operators through targeted reinforcement learning. In AAMAS, 2021

  31. [40]

    McDermott, M

    D. McDermott, M. Ghallab, A. Howe, C. Knoblock, A. Ram, M. Veloso, D. Weld, and D. Wilkins. PDDL - The Planning Domain Definition Language, 1998

  32. [41]

    Gehring, M

    C. Gehring, M. Asai, R. Chitnis, T. Silver, L. Kaelbling, S. Sohrabi, and M. Katz. Reinforce- ment learning for classical planning: Viewing heuristics as dense reward generators. Proceed- ings of the International Conference on Automated Planning and Scheduling, 32(1):588–596,...

  33. [42]

    F. Yang, D. Lyu, B. Liu, and S. Gustafson. Peorl: Integrating symbolic planning and hi- erarchical reinforcement learning for robust decision-making. pages 4860–4866, 07 2018. doi:10.24963/ijcai.2018/675

  34. [43]

    Lorang, H

    P. Lorang, H. Lu, and M. Scheutz. Curiosity-driven imagination: Discovering plan operators and learning associated policies for open-world adaptation. Number arXiv:2503.04931. IEEE International Conference on Robotics and Automation (ICRA), Mar. 2025. doi:10.48550/ arXiv.2503....

  35. [44]

    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. In IEEE ICDL, 2022. 12

  36. [45]

    H. Le, N. Jiang, A. Agarwal, M. Dudik, Y . Yue, and I. I. I. Hal Daum´e. Hierarchical imitation and reinforcement learning. In Proceedings of the 35th International Conference on Machine Learning, page 2917–2926. PMLR, July 2018. URL https://proceedings.mlr.press/ v80/le18a.html

  37. [46]

    Pertsch, Y

    K. Pertsch, Y . Lee, Y . Wu, and J. J. Lim. Demonstration-guided reinforcement learning with learned skills. July 2021

  38. [47]

    A. K. Tanwani, A. Yan, J. Lee, S. Calinon, and K. Goldberg. Sequential robot imitation learn- ing from observations.The International Journal of Robotics Research, 40(10–11):1306–1325, Sept. 2021. ISSN 0278-3649. doi:10.1177/02783649211032721

  39. [48]

    Manschitz, J

    S. Manschitz, J. Kober, M. Gienger, and J. Peters. Learning to sequence movement prim- itives from demonstrations. 2014 IEEE/RSJ International Conference on Intelligent Robots and Systems, page 4414–4421, Sept. 2014. doi:10.1109/IROS.2014.6943187

  40. [49]

    S. Teng, L. Chen, Y . Ai, Y . Zhou, Z. Xuanyuan, and X. Hu. Hierarchical interpretable imitation learning for end-to-end autonomous driving. IEEE Transactions on Intelligent Vehicles, 8(1): 673–683, Jan. 2023. ISSN 2379-8904. doi:10.1109/TIV .2022.3225340

  41. [51]

    Loula, K

    J. Loula, K. Allen, T. Silver, and J. Tenenbaum. Learning constraint-based planning mod- els from demonstrations. In 2020 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), page 5410–5416. IEEE, 2020. URL https://ieeexplore.ieee.org/ abstract/document/9341535/

  42. [52]

    Y . Zhu, P. Stone, and Y . Zhu. Bottom-up skill discovery from unsegmented demonstrations for long-horizon robot manipulation. IEEE Robotics and Automation Letters, 7(2):4126–4133, Apr. 2022. ISSN 2377-3766, 2377-3774. doi:10.1109/LRA.2022.3146589

  43. [53]

    R. S. Sutton, D. Precup, and S. Singh. Between mdps and semi-mdps: A framework for temporal abstraction in reinforcement learning. Artificial Intelligence, 1999. ISSN 0004-3702. doi:https://doi.org/10.1016/S0004-3702(99)00052-1

  44. [54]

    ignoring delete lists

    J. Hoffmann. The metric-ff planning system: Translating“ignoring delete lists”to numeric state variables. Journal of artificial intelligence research, 20:291–341, 2003. 13

  45. [55]

    C. Chi, S. Feng, Y . Du, Z. Xu, E. Cousineau, B. Burchfiel, and S. Song. Diffusion policy: Visuomotor policy learning via action diffusion. In Proceedings of Robotics: Science and Systems (RSS), 2023. 14 A Appendix Figure 6: Illustration of the Nut Assembly envi- ronment in Ro...

  46. [56]

    Curtis, T

    A. Curtis, T. Silver, J. B. Tenenbaum, T. Lozano-P ´erez, and L. Kaelbling. Discovering state and action abstractions for generalized task and motion planning. Proceedings of the AAAI Conference on Artificial Intelligence, 36(5):5377–5384, 2022. ISSN 2374-3468, 2159-5399. doi:...

  47. [2021]

    URL https://doi.org/10.24963/kr.2021/51

    doi:10.24963/kr.2021/51. URL https://doi.org/10.24963/kr.2021/51

  48. [2023]

    arXiv:2301.06294 [cs]

    URL http://arxiv.org/abs/2301.06294. arXiv:2301.06294 [cs]

  49. [8005]

    doi:10.1017/S0269888918000188

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.