Pith. sign in

REVIEW 9 minor 17 references

AI Planning: A Primer and Survey (Preliminary Report)

T0 review · 0 major / 9 minor · reviewed 2026-08-11 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read AI planning's transferable core is the structured symbolic model, and this survey shows how such models enable efficient solving, learning, and generalisation to unseen problems.

desk verdict A useful, accurate planning-RL primer whose new organizing frames (MDP access levels, generalisation in planning) are worth adopting; the one flagged technical issue is not actually an error. read the letter →

arxiv 2412.05528 v1 pith:I236KOBI submitted 2024-12-07 cs.AI

Dillon Z. Chen , Pulkit Verma , Siddharth Srivastava , Michael Katz , Sylvie Thiébaux This is my paper · ORCID
classification cs.AI
keywords AIplanningstructuredrepresentationsMarkovdecisionprocessesheuristicsearchgeneralisedactionmodellearningdomain-independentgeneralisation
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 message is that AI planning's distinctive and transferable contribution is the structured, symbolic model: a decision problem is described with predicates, objects, and action schemata rather than as an opaque simulator. It argues that this compact representation is what makes long-horizon, sparse-reward problems tractable, and it surveys the heuristic search, decomposition, and learning methods that exploit it. The authors aim this primer at researchers in reinforcement learning, foundation models, and operations research who know the decision-making problem but not planning's toolbox. If the framing is accepted, planning becomes a source of representations and algorithms that complement learning-based AI rather than a separate, isolated subfield.

What carries the argument

The load-bearing object is the structured MDP in the STRIPS/PDDL style: predicates and objects generate propositions; a state is any set of propositions under the closed-world assumption; goals are propositional formulas; and transitions come from action schemata with preconditions, add lists, and delete lists. This single representation carries the whole survey's argument, because each major technique is presented as a way of exploiting the compactness and logical form of that representation, and each learning subfield is presented as a way of acquiring or using it.

What would settle it

A decisive check would be a controlled benchmark on a domain family with unbounded object counts: train a structure-agnostic learner on small instances, test on much larger instances, and compare against the same learner given the symbolic action schemata; if the structure-agnostic version matches the structured version on long-horizon sparse-reward tasks, the paper's central claim that structure is what enables efficient generalisation is contradicted.

Watch

Extended reading notes

Core claim

The paper's central claim, stated on its own terms, is that AI planning's core insight is that structure in world models is a resource for decision-making. It formalises a planning task as a factored MDP whose states are sets of ground propositions under the closed-world assumption, whose goals are compact logical formulas, and whose transitions are induced by grounded action schemata with preconditions, add effects, and delete effects. From this representation the paper derives the field's main assets: attention-like relevance, declarative automation, compact encoding of combinatorially large state spaces, domain-independent heuristic search, and problem decomposition through landmarks, mutexes, factoring, and symmetries. It then presents the learning-oriented subfields—learning planning models, learning for planning, and generalised planning—as different ways to acquire, exploit, or upgrade structure, culminating in the claim that structured representations enable generalisation to unseen tasks with arbitrary numbers of objects.

Load-bearing premise

The survey's structure-centred case rests on the accuracy of its account of prior planning work; if a technical characterisation such as the claimed EXPSPACE-completeness of the minimal formalism is wrong, the survey's reliability as a reference is undermined.

Editorial extensions

If this is right

  • Planning's structured representations let an agent upgrade its access to a problem from raw interaction to a compact model, making long-horizon sparse-reward tasks solvable without hand-writing a solver.
  • Domain-independent heuristic search, width-based methods, and decomposition techniques transfer across domains expressed in a planning language, rather than being re-learned from scratch.
  • The same structure supports automatic generation of reward functions and training labels, so learning methods need not rely on hand-designed rewards.
  • Generalised plans and sketches can encode solutions for entire domains with arbitrarily many objects, offering explainable and verifiable alternatives to black-box policies.
  • Because the underlying ideas extend to probabilistic, temporal, numeric, hybrid, and multi-agent settings, the survey's structure-centred view covers decision-making well beyond the classical STRIPS block world.

Reading between the lines

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

  • A consequence the authors leave implicit is that reinforcement-learning agents should receive a symbolic domain model whenever one is available, rather than learning latent dynamics from raw pixel or state-vector observations.
  • The generalisation-in-planning setup gives a concrete evaluation protocol for testing whether large language models can plan: prompt on small instances and test on instances with more objects, comparing against a classical planner's guarantees.
  • One testable extension is to treat learned heuristic functions as ranking functions rather than cost estimators in more settings, since the survey notes that this framing often matches how search actually uses them.
  • A practical recipe suggested by the survey is to use classical planners to generate dense reward functions or training labels, then train fast learned policies; this pipeline could be tested across domains with unbounded object counts.
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

0 major / 9 minor

Summary. The paper is a brief, intentionally nonexhaustive primer and survey of AI planning aimed at readers from RL and adjacent communities. It introduces a hierarchy of MDP access levels, presents classical planning as a structured, first-order MDP representation with closed-world semantics, sketches probabilistic, nondeterministic, temporal, numeric, and multi-agent extensions, reviews structure-exploiting solver techniques (heuristics, heuristic search, decomposition, symmetries), and surveys AP subfields where learning is used to acquire or exploit structure. It concludes with pointers to benchmarks, software, and textbooks. The central framing is that AP's distinctive contribution is a compact symbolic model that enables efficient solving and generalization.

Significance. As a survey, the paper's value is organizational and pedagogical rather than new technical results. The level-of-access taxonomy and the Generalisation-in-Planning formulation are useful lenses for cross-community communication, and the paper is generous and accurate in its pointers to the literature. I checked the potentially contentious complexity statement in §3.2: for the lifted STRIPS formalism defined in §3.1, where action-schema arity is unbounded, EXPSPACE-completeness is the standard result from Erol, Nau, and Subrahmanian (1995); the concern that this should be PSPACE-completeness applies to ground propositional STRIPS or to bounded-arity schemata, not to the formalism as written. The remaining technical characterizations I sampled (MDP definitions, admissible heuristics, A*/GBFS/Weighted A*, LAO*/LRTDP/UCT, landmarks/mutexes/sketches, generalized-planning terminology) are consistent with standard references.

minor comments (9)
  1. [§3.1] The sentence 'can be grounded from task objects to induce a high-order polynomial number of actions and transitions' is imprecise when the arity of action schemata is part of the input: for unbounded arity, the number of ground actions is exponential in the input size, not merely polynomial. Please rephrase so that this statement is consistent with the EXPSPACE-completeness claim in §3.2.
  2. [§3.2] The EXPSPACE-completeness statement is correct for the lifted STRIPS representation defined in §3.1 with unbounded predicate/action arity. For the benefit of readers who know the PSPACE-completeness of ground propositional STRIPS, the paper should add a one-sentence clarification that bounded-arity schemata yield the polynomial-size ground case, which is PSPACE-complete, whereas the compact lifted representation is what gives EXPSPACE-completeness.
  3. [§4.3] The sentence 'relative simple representations for generalised plans and are equivalent to counter-based models of computing, and thus equivalent to Turing equivalent' is ungrammatical. It should read, for example, 'relatively simple representations for generalised plans are equivalent to counter-based models of computing, and are thus Turing equivalent.'
  4. [§1] In the second bullet of the Introduction, 'On top of using learning to make planning feasible the discovery of structured models' is missing a connecting phrase; it should be something like 'On top of using learning to make the discovery of structured models feasible, we can also learn to make planning fast.'
  5. [§3.5] The package name 'Scickit-Decide' should be 'Scikit-Decide'.
  6. [Figure 1] Figure 1 is not referenced in the text, and the caption as rendered ('train = test train size = test size train size ⊂ test size ...') is not self-explanatory. Please add a pointer to the figure in Section 4 and provide a readable caption that explains the nesting of generalization setups.
  7. [§3.4] The description of GBFS and Weighted A* is ambiguous: the phrase 'where g(n)=0 and f(n)=g(n)+w·h(s), respectively' should be split into two explicit definitions, e.g., GBFS uses f(n)=h(s) (with g(n)=0) and Weighted A* uses f(n)=g(n)+w·h(s).
  8. [§3.4] The phrase 'UCT ... are anytime optimal algorithms' is imprecise: UCT is an anytime algorithm whose value estimates converge to optimal values in the limit, but it does not provide an optimality guarantee at every stopping time. Please reword.
  9. [§2] The phrase 'distribution models, also known as model-based RL' is loose; Sutton and Barto's dichotomy is between distribution models and sample models, and the term 'model-based RL' in the modern literature includes learned models and broader settings. Consider a more careful phrasing.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper is a survey and primer that reports external results; its framing and definitions do not reduce to their own inputs.

full rationale

This paper is secondary literature rather than a derivation chain. Its central claims are organizational and expository: that planning's contribution is structured symbolic models, and that such structure can be exploited for efficient solving and generalization. These claims are supported by definitions (Secs. 2 and 3.1) and by citations to external, independently developed results. There are no fitted parameters, no empirical predictions derived from fitted values, and no result that is asserted on the basis of a self-citation chain. The technical statements, including the EXPSPACE-completeness claim in Sec. 3.2, are presented as citations to Erol, Nau, and Subrahmanian (1995) for the compact lifted representation described in Sec. 3.1, and are not derived within the paper. Some cited works include authors of this survey (e.g., Katz, Srivastava, Verma, Thiébaux, and Chen), but those citations identify existing literature and are not load-bearing premises that force the survey's conclusions. The Generalisation in Planning setup in Sec. 4 is introduced as a definitional taxonomy for organizing existing work, not as a mathematical derivation. Similarly, the statement in Sec. 4.3 that generalized plans correspond to counter-based models of computing is attributed to prior work and is independent of the survey's own argument. The paper also includes appropriate caveats, such as the footnote on proper policies and pointers to fuller treatments. Because there is no self-derivation, no renamed fit, and no uniqueness claim imported from the authors' own prior work, the appropriate circularity score is 0.

Assumptions & free parameters 0 free parameters · 3 assumptions · 0 invented entities

The survey introduces no new free parameters or entities. It relies on standard math (MDPs, logic) and on the correctness of cited works. Its conceptual frameworks (access hierarchy, generalisation in planning) are organizational, not formal constructs requiring independent evidence.

assumptions (3)
  • standard math Standard definitions of MDPs and Bellman equations are accepted.
    Used in Sec. 2 to frame planning in RL terms.
  • domain assumption The closed-world assumption and STRIPS semantics are standard in planning.
    Introduced in Sec. 3.1 to define states and transitions.
  • domain assumption The survey assumes the cited works correctly support the described results.
    The survey is a secondary source; its correctness depends on the primary literature.

how reviews work

0 comments
Cite this review

Pith. "Pith review of AI Planning: A Primer and Survey (Preliminary Report)." pith.science (2026). https://pith.science/paper/I236KOBI

@misc{pith2026241205528,
  author       = {Pith},
  title        = {Pith review of: AI Planning: A Primer and Survey (Preliminary Report)},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/I236KOBI}},
  note         = {Machine review of arXiv:2412.05528}
}
read the original abstract

Automated decision-making is a fundamental topic that spans multiple sub-disciplines in AI: reinforcement learning (RL), AI planning (AP), foundation models, and operations research, among others. Despite recent efforts to ``bridge the gaps'' between these communities, there remain many insights that have not yet transcended the boundaries. Our goal in this paper is to provide a brief and non-exhaustive primer on ideas well-known in AP, but less so in other sub-disciplines. We do so by introducing the classical AP problem and representation, and extensions that handle uncertainty and time through the Markov Decision Process formalism. Next, we survey state-of-the-art techniques and ideas for solving AP problems, focusing on their ability to exploit problem structure. Lastly, we cover subfields within AP for learning structure from unstructured inputs and learning to generalise to unseen scenarios and situations.

Figures

Figures reproduced from arXiv: 2412.05528 by the authors.

Figure 1
Figure 1. Generalisation setups for decision-making. AP ap [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

17 extracted references · 14 canonical work pages

  1. [6]

    A Planning Approach to Repair Domains with Incom- plete Action Effects. InICAPS. Gretton, C.; and Thi ´ebaux, S. 2004. Exploiting First-Order Regression in Inductive Policy Selection. InUAI. Grohe, M. 2021. The Logic of Graph Neural Networks. In LICS. Grundke, C.; R ¨oger, G.; and Helmert, M. 2024. Formal Rep- resentations of Classical Planning Domains. I...

    work page 2004

  2. [8]

    A Survey of Zero-shot Generalisation in Deep Rein- forcement Learning.J. Artif. Intell. Res., 76: 201–264. Kocsis, L.; and Szepesv ´ari, C. 2006. Bandit Based Monte- Carlo Planning. InECML. Konidaris, G. 2019. On the necessity of abstraction.Current Opinion in Behavioral Sciences, 29: 1–7. Articial Intelli- gence. Kumar, N.; Silver, T.; McClinton, W.; Zh...

    work page 2006

  3. [11]

    Model-based Reinforcement Learning: A Survey. F ound. Trends Mach. Learn., 16: 1–118. Morris, C.; Ritzert, M.; Fey, M.; Hamilton, W. L.; Lenssen, J. E.; Rattan, G.; and Grohe, M. 2019. Weisfeiler and Leman Go Neural: Higher-Order Graph Neural Networks. InAAAI. Muggleton, S. H.; and Raedt, L. D. 1994. Inductive Logic Programming: Theory and Methods.J. Log....

    work page 2019

  4. [15]

    SymNet 2.0: Effectively handling Non-Fluents and Actions in Generalized Neural Policies for RDDL Relational MDPs. InUAI. Shen, W.; Trevizan, F.; and Thi ´ebaux, S. 2020. Learning Domain-Independent Planning Heuristics with Hypergraph Networks. InICAPS. Shleyfman, A.; Katz, M.; Helmert, M.; Sievers, S.; and Wehrle, M. 2015. Heuristics and Symmetries in Cla...

    work page 2020

  5. [1998]

    Technical report

    PDDL-the planning domain denition language. Technical report. Micheli, A.; and V alentini, A. 2021. Synthesis of Search Heuristics for Temporal Planning via Reinforcement Learn- ing. InAAAI. Moerland, T. M.; Broekens, J.; Plaat, A.; and Jonker, C. M

    work page 2021

  6. [2005]

    Activity Planning for the Mars Exploration Rovers. InICAPS. Brockman, G. 2016. OpenAI Gym.arXiv preprint arXiv:1606.01540. Callanan, E.; V enezia, R. D.; Armstrong, V .; Paredes, A.; Kang, J.; Chakraborti, T.; and Muise, C. 2022. MACQ: A Unied Library for Action Model Acquisition. InInterna- tional Conference on Automated Planning and Scheduling: System ...

    arXiv 2016

  7. [2008]

    Beyond Classical Planning: Procedural Control Knowledge and Preferences in State-of-the-Art Planners. In AAAI. Barcel´o, P .; Kostylev, E. V .; Monet, M.; P ´erez, J.; Reutter, J. L.; and Silva, J. P . 2020. The Logical Expressiveness of Graph Neural Networks. InICLR. Bercher, P .; Alford, R.; and H ¨oller, D. 2019. A Survey on Hierarchical Planning - One...

    work page 2020

  8. [2011]

    InPro- ceedings of Inductive Logic Programming

    Active Learning of Relational Action Models. InPro- ceedings of Inductive Logic Programming. Rodriguez, I. D.; Bonet, B.; Romero, J.; and Geffner, H

    work page

Show all 17 references
  1. [2014]

    In ICAPS

    LP-Based Heuristics for Cost-Optimal Planning. In ICAPS. Porteous, J.; Sebastia, L.; and Hoffmann, J. 2001. On the Extraction, Ordering, and Usage of Landmarks in Planning. InSixth European Conference on Planning. Richter, S.; Helmert, M.; and Westphal, M. 2008. Land- marks Re...

    2001

  2. [2016]

    Silver, T.; and Chitnis, R

    Mastering the game of Go with deep neural networks and tree search.Nat., 529: 484–489. Silver, T.; and Chitnis, R. 2020. PDDLGym: Gym Environ- ments from PDDL Problems. InInternational Conference on Automated Planning and Scheduling (ICAPS) PRL Work- shop. Silver, T.; Chitnis,...

    2020

  3. [2017]

    Adapting Novelty to Classical Planning as Heuristic Search. InICAPS. Keller, T.; and Helmert, M. 2013. Trial-Based Heuristic Tree Search for Finite Horizon MDPs. InICAPS. Khardon, R. 1999. Learning Action Strategies for Planning Domains.Artif. Intell., 113: 125–148. Kirk, R.; ...

    2013

  4. [2018]

    Surv., 50: 84:1–84:32

    Cooperative Multi-Agent Planning: A Survey.ACM Comput. Surv., 50: 84:1–84:32. Toyer, S.; Thi´ebaux, S.; Trevizan, F.; and Xie, L. 2020. AS- Nets: Deep Learning for Generalised Planning.J. Artif. In- tell. Res., 68: 1–68. Toyer, S.; Trevizan, F. W.; Thi ´ebaux, S.; and Xie, L. ...

    2020 arXiv

  5. [2019]

    Neural Logic Machines. InICLR. Dzeroski, S.; Raedt, L. D.; and Driessens, K. 2001. Rela- tional Reinforcement Learning.Mach. Learn., 43: 7–52. Edelkamp, S. 2002. Symbolic Pattern Databases in Heuristic Search Planning. InAIPS. Edelkamp, S. 2003. Taming Numbers and Durations in...

    2001

  6. [2021]

    Learning First-Order Representations for Planning from Black Box States: New Results. InKR. R¨oger, G.; and Helmert, M. 2010. The More, the Merrier: Combining Heuristic Estimators for Satiscing Planning. In ICAPS. Russell, S.; and Norvig, P . 2020.Articial Intelligence: A Mo...

    2010

  7. [2022]

    Learn., 111: 147–172

    Inductive logic programming at 30.Mach. Learn., 111: 147–172. Culberson, J. C.; and Schaeffer, J. 1996. Searching with Pattern Databases. InAI. Culberson, J. C.; and Schaeffer, J. 1998. Pattern Databases. Comput. Intell., 14: 318–334. Cˆot´e, M.-A.; ´Akos K ´ad´ar; Y uan, X.; ...

    1996 arXiv

  8. [2023]

    InNeurIPS

    Optimize Planning Heuristics to Rank, not to Estimate Cost-to-Goal. InNeurIPS. Cimatti, A.; Pistore, M.; Roveri, M.; and Traverso, P . 2003. Weak, strong, and strong cyclic planning via symbolic model checking.Artif. Intell., 147: 35–84. Coquelin, P .; and Munos, R. 2007. Band...

    2003

  9. [2024]

    Intell., 328: 104062

    From statistical relational to neurosymbolic articial intelligence: A survey.Artif. Intell., 328: 104062. Mart´ın, M.; and Geffner, H. 2004. Learning General- ized Policies from Planning Examples Using Concept Lan- guages.Appl. Intell., 20: 9–19. Mausam; and Kolobov, A. 2012....

    2004

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.