pith. machine review for the scientific record. sign in

arxiv: 2512.09629 · v2 · submitted 2025-12-10 · 💻 cs.AI · cs.LG

Recognition: 2 theorem links

· Lean Theorem

End-to-end PDDL Planning with Hardcoded and Dynamic Agents

Emanuele La Malfa , Ping Zhu , Samuele Marro , Sara Bernardini , Michael Wooldridge
Authors on Pith no claims yet

Pith reviewed 2026-05-16 23:31 UTC · model grok-4.3

classification 💻 cs.AI cs.LG
keywords end-to-end planningPDDLLLM agentsnatural language to PDDLhardcoded agentsdynamic agentsautomated planningAI planning
0
0 comments X

The pith

An LLM orchestrator converts natural language specs into verified PDDL plans using hardcoded and dynamic agents with no human input.

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

The paper presents a system in which an orchestrator receives a natural language specification and produces a PDDL domain and problem file. Hardcoded agents address fixed issues such as syntax errors, temporal constraints, and optimality requirements using error traces, while dynamic agents adapt without preset goals to revise the underlying planning abstraction for the given domain. The refined model passes to an external planner to generate a solution, after which a final module renders the plan back into readable natural language. The approach is evaluated on more than ten domains and tasks, including classic problems where LLMs alone perform poorly.

Core claim

The central claim is that an LLM-powered orchestrator combined with two categories of agents can produce a complete, validated PDDL model from an ambiguous natural language input: hardcoded agents apply predefined fixes for syntax, time constraints and similar issues, while dynamic agents revise the latent planning abstraction to fit the specific domain, after which any external PDDL engine computes a plan and the system translates it back to natural language while preserving correctness.

What carries the argument

The orchestrator together with its hardcoded agents (pre-defined goals informed by logs and traces) and dynamic agents (no fixed goal, domain-adaptive revision of planning abstraction), all implemented via LLMs, that iteratively refine the PDDL model before external plan generation.

If this is right

  • The framework works with any PDDL planning engine and validator, including Fast Downward, LPG, POPF, VAL and uVAL.
  • It produces usable plans on domains such as Sokoban, Blocksworld and Tower of Hanoi where standalone LLMs fail even on small instances.
  • The final plan is rendered back into natural language while each step remains correct.
  • Performance holds across GPT-4o, GPT-5-mini, GPT-5.4, Gemini-2.5-flash and Gemini-3-flash on the Google NaturalPlan benchmark and Planbench.
  • The system requires zero human intervention from specification receipt through plan output.

Where Pith is reading between the lines

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

  • The hybrid hardcoded-plus-dynamic design may transfer to other tasks that convert informal requirements into formal specifications.
  • Dynamic agents could allow the system to handle entirely new domains without additional hardcoded rules.
  • Success still depends on the coverage of the chosen external validators; gaps in validator expressiveness would limit reliability.
  • Further gains are likely if stronger LLMs reduce the number of refinement cycles needed.

Load-bearing premise

LLM agents can detect and repair every ambiguity, contradiction, syntax problem and constraint violation in the original specification without missing errors or creating new ones that external validators fail to catch.

What would settle it

A trial input containing a clear contradiction or unstated constraint where the generated PDDL passes all validators yet the external planner returns a plan that violates the original natural language requirement.

Figures

Figures reproduced from arXiv: 2512.09629 by Emanuele La Malfa, Michael Wooldridge, Ping Zhu, Samuele Marro, Sara Bernardini.

Figure 1
Figure 1. Figure 1: Overview of our end-to-end planning framework graph that generates a plan, backed up by a planner, from a human [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overview of our planning framework illustrating how the agentic components interact with each other. The “PDDL [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Results for GPT-5-mini on the Google Natural [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Results for GPT-5-mini on increasingly diffi [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Frequency of each agent for the Google Natural Plan Benchmark and Planbench. [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: The overall frequency of each agent in the framework, across the Google Natural Language Benchmark and Plan [PITH_FULL_IMAGE:figures/full_fig_p010_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: The interface of the Planning Copilot [PITH_FULL_IMAGE:figures/full_fig_p012_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: The framework produces the first JSON representation and PDDL domain and problem. [PITH_FULL_IMAGE:figures/full_fig_p013_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: The framework displays the current SAS plan. [PITH_FULL_IMAGE:figures/full_fig_p014_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: The final SAS plan is back-translated into natural language. [PITH_FULL_IMAGE:figures/full_fig_p015_10.png] view at source ↗
read the original abstract

We present an end-to-end framework for planning supported by verifiers. An orchestrator receives a human specification written in natural language and converts it into a PDDL (Planning Domain Definition Language) model, where the domain and problem are iteratively refined by sub-modules (agents) to address common planning requirements, such as time constraints and optimality, as well as ambiguities and contradictions that may exist in the human specification. We support two categories of agents: hardcoded, which are informed by logs and error traces and have a pre-defined goal (e.g., fix issues with PDDL syntax, check temporal constraints), and dynamic, which have no predefined goal but adapt to the specific domain and revise the latent planning abstraction. The validated domain and problem are then passed to an external planning engine to generate a plan. The orchestrator and agents are powered by Large Language Models (LLMs) and require no human intervention at any stage of the process. Finally, a module translates the final plan back into natural language to improve human readability while maintaining the correctness of each step. We demonstrate the flexibility and effectiveness of our framework on GPT-\{4o, 5-mini, 5.4\}, and Gemini-\{2.5, 3\}-flash across more than ten domains and tasks, including the Google NaturalPlan benchmark, Planbench, and classic planning problems like Sokoban, Blocksworld and the Tower of Hanoi, where LLMs are known to struggle even with small instances. Our framework can be integrated with any PDDL planning engine and validator (we successfully tested Fast Downward, LPG, POPF, VAL, and uVAL) and represents a significant step toward end-to-end planning aided by LLMs.

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 / 2 minor

Summary. The manuscript presents an end-to-end LLM-orchestrated framework for PDDL planning: an orchestrator converts natural-language specifications into domain and problem files, which are iteratively refined by hardcoded agents (pre-defined goals for syntax, temporal, and log-based fixes) and dynamic agents (goal-free adaptation of latent abstractions), validated via external tools (VAL, uVAL), solved by standard planners (Fast Downward, LPG, POPF), and finally translated back to readable natural language. The system requires no human intervention and is evaluated on GPT-4o/5-mini/5.4 and Gemini-2.5/3-flash across >10 domains including NaturalPlan, PlanBench, Sokoban, Blocksworld, and Tower of Hanoi.

Significance. If the reliability claims hold, the architecture offers a concrete, modular way to handle specification ambiguities and common PDDL pitfalls with LLMs while preserving compatibility with existing validators and solvers; the separation of hardcoded and dynamic agents is a useful design distinction. However, the absence of quantitative success rates, failure-mode analysis, or baseline comparisons in the reported evaluation weakens the ability to assess whether the framework actually delivers reliable end-to-end correctness.

major comments (2)
  1. [Abstract] Abstract: the central claim that the framework demonstrates 'flexibility and effectiveness' on benchmarks and classic problems where LLMs struggle is unsupported by any quantitative success rates, failure modes, or comparisons to baselines; this data is load-bearing for the assertion of reliable end-to-end operation without human intervention.
  2. [Framework description] Framework description (orchestrator and dynamic-agent sections): dynamic agents operate on error traces without a fixed goal and can therefore converge to a PDDL model whose semantics differ from the original human specification (e.g., altered predicate meanings or unstated constraints) while still passing VAL/uVAL and producing a valid plan; no concrete safeguards, semantic-equivalence checks, or ablation experiments are described to bound this risk.
minor comments (2)
  1. [Abstract] Abstract: the model list 'GPT-{4o, 5-mini, 5.4}' and 'Gemini-{2.5, 3}-flash' should use precise version strings (e.g., gpt-4o-2024-08-06) for reproducibility.
  2. [Evaluation] Evaluation section: integration claims with Fast Downward, LPG, POPF, VAL, and uVAL would benefit from a brief table listing which validator/planner combination was used per domain.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback highlighting the need for quantitative support and safeguards against semantic drift. We address each major comment below and will incorporate revisions to strengthen the manuscript.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central claim that the framework demonstrates 'flexibility and effectiveness' on benchmarks and classic problems where LLMs struggle is unsupported by any quantitative success rates, failure modes, or comparisons to baselines; this data is load-bearing for the assertion of reliable end-to-end operation without human intervention.

    Authors: We agree that the abstract's claims require quantitative backing. The full manuscript reports results across the listed domains and planners, but explicit aggregate success rates, per-domain breakdowns, failure-mode categorization, and baseline comparisons (e.g., direct LLM prompting) are not presented in sufficient detail. We will add a dedicated evaluation subsection with these metrics and comparisons in the revised version. revision: yes

  2. Referee: [Framework description] Framework description (orchestrator and dynamic-agent sections): dynamic agents operate on error traces without a fixed goal and can therefore converge to a PDDL model whose semantics differ from the original human specification (e.g., altered predicate meanings or unstated constraints) while still passing VAL/uVAL and producing a valid plan; no concrete safeguards, semantic-equivalence checks, or ablation experiments are described to bound this risk.

    Authors: We acknowledge the risk of semantic drift when dynamic agents lack a fixed goal. The current design uses VAL/uVAL for syntactic and plan-validity checks and pairs dynamic agents with hardcoded agents that enforce temporal and log-based constraints, but no explicit semantic-equivalence verification to the original natural-language specification or ablation studies isolating the dynamic component are described. In revision we will add a limitations paragraph discussing this risk, outline any implicit safeguards, and include ablation results comparing configurations with and without dynamic agents. revision: yes

Circularity Check

0 steps flagged

No circularity detected in derivation chain

full rationale

The paper describes an LLM-orchestrated framework for converting natural-language specifications into validated PDDL models using hardcoded and dynamic agents, followed by external planning and plan translation. No equations, derivations, fitted parameters, or self-referential logic appear in the architecture or evaluation. The central claims rest on system design choices and empirical results across benchmarks rather than any reduction of outputs to inputs by construction. Self-citations, if present, are not load-bearing for any uniqueness theorem or ansatz. The contribution is therefore self-contained as an engineering description with no circular steps.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

The framework introduces no new mathematical axioms, free parameters, or invented physical entities; it composes existing LLM capabilities, PDDL syntax, and off-the-shelf planners.

pith-pipeline@v0.9.0 · 5623 in / 1144 out tokens · 47588 ms · 2026-05-16T23:31:21.541492+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

42 extracted references · 42 canonical work pages · 6 internal anchors

  1. [1]

    , " * write output.state after.block = add.period write newline

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint howpublished institution isbn journal key month note number organization pages publisher school series title type volume year label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.state.consts #0 'before.a...

    work page

  2. [2]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...

    work page

  3. [3]

    Ahn, M.; Brohan, A.; Brown, N.; Chebotar, Y.; Cortes, O.; David, B.; Finn, C.; Fu, C.; Gopalakrishnan, K.; Hausman, K.; et al. 2022. Do as i can, not as i say: Grounding language in robotic affordances. arXiv preprint arXiv:2204.01691

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  4. [4]

    L.; and Furst, M

    Blum, A. L.; and Furst, M. L. 1997. Fast planning through planning graph analysis. Artificial intelligence, 90(1-2): 281--300

    work page 1997

  5. [5]

    MLE-bench: Evaluating Machine Learning Agents on Machine Learning Engineering

    Chan, J. S.; Chowdhury, N.; Jaffe, O.; Aung, J.; Sherburn, D.; Mays, E.; Starace, G.; Liu, K.; Maksin, L.; Patwardhan, T.; et al. 2024. Mle-bench: Evaluating machine learning agents on machine learning engineering. arXiv preprint arXiv:2410.07095

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  6. [6]

    Chase, H. 2022. LangChain

    work page 2022

  7. [7]

    H.; and Bengio, Y

    Chevalier-Boisvert, M.; Bahdanau, D.; Lahlou, S.; Willems, L.; Saharia, C.; Nguyen, T. H.; and Bengio, Y. 2019. Baby AI : First Steps Towards Grounded Language Learning With a Human In the Loop. In International Conference on Learning Representations

    work page 2019

  8. [8]

    E.; Lee, N.; Kim, S.; Moon, S.; Furuta, H.; Anumanchipalli, G.; Keutzer, K.; and Gholami, A

    Erdogan, L. E.; Lee, N.; Kim, S.; Moon, S.; Furuta, H.; Anumanchipalli, G.; Keutzer, K.; and Gholami, A. 2025. Plan-and-act: Improving planning of agents for long-horizon tasks. arXiv preprint arXiv:2503.09572

    work page arXiv 2025

  9. [9]

    Farquhar, S.; Kossen, J.; Kuhn, L.; and Gal, Y. 2024. Detecting hallucinations in large language models using semantic entropy. Nature, 630(8017): 625--630

    work page 2024

  10. [10]

    E.; and Nilsson, N

    Fikes, R. E.; and Nilsson, N. J. 1971. STRIPS: A new approach to the application of theorem proving to problem solving. Artificial intelligence, 2(3-4): 189--208

    work page 1971

  11. [11]

    A.; De Freitas, N.; and Whiteson, S

    Foerster, J.; Assael, I. A.; De Freitas, N.; and Whiteson, S. 2016. Learning to communicate with deep multi-agent reinforcement learning. Advances in neural information processing systems, 29

    work page 2016

  12. [12]

    Ghallab, M.; Howe, A.; Knoblock, C.; McDermott, D.; Ram, A.; Veloso, M.; Weld, D.; and Wilkins. 1998. Pddl—the planning domain definition language. Technical Report, Tech. Rep

    work page 1998

  13. [13]

    Grosz, B. J. 1996. Collaborative systems (AAAI-94 presidential address). AI magazine, 17(2): 67--67

    work page 1996

  14. [14]

    Gu, J.; Jiang, X.; Shi, Z.; Tan, H.; Zhai, X.; Xu, C.; Li, W.; Shen, Y.; Ma, S.; Liu, H.; Wang, S.; Zhang, K.; Wang, Y.; Gao, W.; Ni, L.; and Guo, J. 2025. A Survey on LLM-as-a-Judge. arXiv:2411.15594

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  15. [15]

    Gundawar, A.; Valmeekam, K.; Verma, M.; and Kambhampati, S. 2024. Robust Planning with Compound LLM Architectures: An LLM-Modulo Approach. arXiv preprint arXiv:2411.14484

    work page arXiv 2024

  16. [16]

    Z.; and De Raedt, L

    Hazra, R.; Dos Martires, P. Z.; and De Raedt, L. 2024. Saycanpay: Heuristic planning with large language models using learnable domain knowledge. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 38, 20123--20133

    work page 2024

  17. [17]

    Helmert, M. 2006. The fast downward planning system. Journal of Artificial Intelligence Research, 26: 191--246

    work page 2006

  18. [18]

    Hoffmann, J. 2001. FF: The fast-forward planning system. AI magazine, 22(3): 57--57

    work page 2001

  19. [19]

    Kambhampati, S.; Valmeekam, K.; Guan, L.; Stechly, K.; Verma, M.; Bhambri, S.; Saldyt, L.; and Murthy, A. 2024. LLMs Can't Plan, But Can Help Planning in LLM-Modulo Frameworks. ArXiv, abs/2402.01817

    work page arXiv 2024

  20. [20]

    Z.; Due \ n ez-Guzman, E

    Leibo, J. Z.; Due \ n ez-Guzman, E. A.; Vezhnevets, A.; Agapiou, J. P.; Sunehag, P.; Koster, R.; Matyas, J.; Beattie, C.; Mordatch, I.; and Graepel, T. 2021. Scalable evaluation of multi-agent reinforcement learning with melting pot. In International conference on machine learning, 6187--6199. PMLR

    work page 2021

  21. [21]

    Liu, B.; Jiang, Y.; Zhang, X.; Liu, Q.; Zhang, S.; Biswas, J.; and Stone, P. 2023. Llm+ p: Empowering large language models with optimal planning proficiency. arXiv preprint arXiv:2304.11477

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  22. [22]

    I.; Tamar, A.; Harb, J.; Pieter Abbeel, O.; and Mordatch, I

    Lowe, R.; Wu, Y. I.; Tamar, A.; Harb, J.; Pieter Abbeel, O.; and Mordatch, I. 2017. Multi-agent actor-critic for mixed cooperative-competitive environments. Advances in neural information processing systems, 30

    work page 2017

  23. [23]

    Mahdavi, S.; Aoki, R.; Tang, K.; and Cao, Y. 2024. Leveraging Environment Interaction for Automated PDDL Translation and Planning with Large Language Models. arXiv:2407.12979

    work page arXiv 2024

  24. [24]

    L.; Petrov, A.; Frieder, S.; Weinhuber, C.; Burnell, R.; Nazar, R.; Cohn, A

    Malfa, E. L.; Petrov, A.; Frieder, S.; Weinhuber, C.; Burnell, R.; Nazar, R.; Cohn, A. G.; Shadbolt, N.; and Wooldridge, M. 2023. Language Models as a Service: Overview of a New Paradigm and its Challenges. arXiv:2309.16573

    work page arXiv 2023

  25. [25]

    McDermott, D.; Ghallab, M.; Howe, A.; Knoblock, C.; Ram, A.; Veloso, M.; Weld, D.; and Wilkins, D. 1998. PDDL --- T he Planning Domain Definition Language. Technical Report CVC TR-98-003 / DCS TR-1165, Yale Center for Computational Vision and Control, New Haven, CT. Technical report

    work page 1998

  26. [26]

    A.; Amato, C.; et al

    Oliehoek, F. A.; Amato, C.; et al. 2016. A concise introduction to decentralized POMDPs, volume 1. Springer

    work page 2016

  27. [27]

    Oswald, J.; Srinivas, K.; Kokel, H.; Lee, J.; Katz, M.; and Sohrabi, S. 2024. Large language models as planning domain generators. In Proceedings of the International Conference on Automated Planning and Scheduling, volume 34, 423--431

    work page 2024

  28. [28]

    Shojaee, P.; Mirzadeh, I.; Alizadeh, K.; Horton, M.; Bengio, S.; and Farajtabar, M. 2025. The illusion of thinking: Understanding the strengths and limitations of reasoning models via the lens of problem complexity. arXiv preprint arXiv:2506.06941

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  29. [29]

    B.; Cardenas, E.; Sharma, A.; Trengrove, J.; and van Luijt, B

    Shorten, C.; Pierse, C.; Smith, T. B.; Cardenas, E.; Sharma, A.; Trengrove, J.; and van Luijt, B. 2024. Structuredrag: Json response formatting with large language models. arXiv preprint arXiv:2408.11061

    work page arXiv 2024

  30. [30]

    Shridhar, M.; Thomason, J.; Gordon, D.; Bisk, Y.; Han, W.; Mottaghi, R.; Zettlemoyer, L.; and Fox, D. 2020 a . Alfred: A benchmark for interpreting grounded instructions for everyday tasks. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 10740--10749

    work page 2020

  31. [31]

    Shridhar, M.; Yuan, X.; C \^o t \'e , M.-A.; Bisk, Y.; Trischler, A.; and Hausknecht, M. 2020 b . Alfworld: Aligning text and embodied environments for interactive learning. arXiv preprint arXiv:2010.03768

    work page internal anchor Pith review Pith/arXiv arXiv 2020

  32. [32]

    B.; Kaelbling, L.; and Katz, M

    Silver, T.; Dan, S.; Srinivas, K.; Tenenbaum, J. B.; Kaelbling, L.; and Katz, M. 2024. Generalized planning in pddl domains with pretrained large language models. In Proceedings of the AAAI conference on artificial intelligence, volume 38, 20256--20264

    work page 2024

  33. [33]

    Sukhbaatar, S.; Fergus, R.; et al. 2016. Learning multiagent communication with backpropagation. Advances in neural information processing systems, 29

    work page 2016

  34. [34]

    Tambe, M. 1997. Towards flexible teamwork. Journal of artificial intelligence research, 7: 83--124

    work page 1997

  35. [35]

    M.; et al

    Toledo, E.; Hambardzumyan, K.; Josifoski, M.; Hazra, R.; Baldwin, N.; Audran-Reiss, A.; Kuchnik, M.; Magka, D.; Jiang, M.; Lupidi, A. M.; et al. 2025. AI Research Agents for Machine Learning: Search, Exploration, and Generalization in MLE-bench. arXiv preprint arXiv:2507.02554

    work page arXiv 2025

  36. [36]

    Valmeekam, K.; Marquez, M.; and Kambhampati, S. 2023. Can Large Language Models Really Improve by Self-critiquing Their Own Plans? ArXiv, abs/2310.08118

    work page arXiv 2023

  37. [37]

    Valmeekam, K.; Marquez, M.; Olmo, A.; Sreedharan, S.; and Kambhampati, S. 2023. Planbench: An extensible benchmark for evaluating large language models on planning and reasoning about change. Advances in Neural Information Processing Systems, 36: 38975--38987

    work page 2023

  38. [38]

    Valmeekam, K.; Olmo, A.; Sreedharan, S.; and Kambhampati, S. 2022. Large Language Models Still Can't Plan (A Benchmark for LLM s on Planning and Reasoning about Change). In NeurIPS 2022 Foundation Models for Decision Making Workshop

    work page 2022

  39. [39]

    Wu, G.; Zhao, C.; Silva, C.; and He, H. 2024. Your co-workers matter: Evaluating collaborative capabilities of language models in blocks world. arXiv preprint arXiv:2404.00246

    work page arXiv 2024

  40. [40]

    R.; and Cao, Y

    Yao, S.; Zhao, J.; Yu, D.; Du, N.; Shafran, I.; Narasimhan, K. R.; and Cao, Y. 2022. React: Synergizing reasoning and acting in language models. In The eleventh international conference on learning representations

    work page 2022

  41. [41]

    arXiv preprint arXiv:2406.04520 , year=

    Zheng, H. S.; Mishra, S.; Zhang, H.; Chen, X.; Chen, M.; Nova, A.; Hou, L.; Cheng, H.-T.; Le, Q. V.; Chi, E. H.; et al. 2024. Natural plan: Benchmarking llms on natural language planning. arXiv preprint arXiv:2406.04520

    work page arXiv 2024

  42. [42]

    P.; Li, X.; Littman, M

    Zuo, M.; Velez, F. P.; Li, X.; Littman, M. L.; and Bach, S. H. 2024. Planetarium: A rigorous benchmark for translating text to structured planning languages. arXiv preprint arXiv:2407.03321

    work page arXiv 2024