Pith. sign in

REVIEW 3 major objections 5 minor 29 references

Where Paths Collide: A Comprehensive Survey of Classic and Learning-Based Multi-Agent Pathfinding

T0 review · 3 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read This survey claims that MAPF research is split into classical and learning-based camps with incompatible evaluation habits, and that the split hides how much more thoroughly classical methods are stress-tested.

desk verdict A useful MAPF survey that bridges classical and learning-based work, but the headline scale-disparity claim rests on an unreleased 200-paper corpus and should be treated as plausible, not verified. read the letter →

arxiv 2505.19219 v2 pith:Z5TMGNZQ submitted 2025-05-25 cs.AI cs.LGcs.MAmath.CO

Shiyue Wang , Haozheng Xu , Yuhan Zhang , Jingran Lin , Changhong Lu , Xiangfeng Wang , Wenhao Li This is my paper · ORCID
classification cs.AIcs.LGcs.MAmath.CO
keywords multi-agentpathfindingMAPFconflict-basedsearchreinforcementlearningevaluationmethodologybenchmarkinglearning-basedplanningsurvey
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

This survey tries to unify the two largely separate strands of multi-agent pathfinding (MAPF) research, classical search and compilation solvers on one side and learning-based methods on the other, under a single framework. It claims that classical methods are typically evaluated on far larger instances, up to 200 by 200 grids with more than 1000 agents, while learning-based methods are mostly tested on 10 to 100 agents. It argues that evaluation metrics, map types, and baseline choices vary so widely across papers that reported performance numbers are not directly comparable, and it calls for standardized benchmarking protocols. A reader cares because this comparison determines which methods actually scale in practice and whether learning-based results generalize to real deployments.

What carries the argument

The central mechanism is the systematic meta-analysis of experimental practices across more than 200 MAPF papers, organized by a unified taxonomy of search-based, compilation-based, and learning-based methods. The taxonomy classifies methods into Conflict-Based Search variants, Priority-Based Search, Large Neighborhood Search, SAT/SMT/CSP/ASP/MIP compilations, reinforcement learning, supervised learning, and hybrid approaches, and the meta-analysis then tabulates map sizes, agent counts, evaluation metrics, and baseline choices to expose the scale gap.

What would settle it

A reader could compile every MAPF paper from a fixed set of venues and years using explicit inclusion rules and extract map size, agent count, metrics, and baselines; if learning-based papers are found to use instances as large as classical ones, or if classical papers typically use small instances, the claimed scale disparity would not hold.

Watch

Extended reading notes

Core claim

The paper's central claim is that MAPF has a common formal core, an undirected graph with collision-free path constraints, but the field has diverged into classical and learning camps with different priorities and evaluation habits. Through a systematic analysis of experimental practices across more than 200 papers, the survey finds a striking scale disparity: classical search-based and compilation-based methods are routinely tested on grids up to 200 by 200 with over 1000 agents, whereas reinforcement learning and other learning-based approaches are predominantly evaluated with 10 to 100 agents. The survey also documents wide variation in evaluation metrics, environment types, and baseline selections, and uses this meta-analysis to argue that the field needs standardized benchmarking to make cross-paradigm comparisons meaningful.

Load-bearing premise

The survey assumes the more than 200 papers it analyzed are a fair and accurately reported sample of MAPF evaluation practice, yet it provides no selection criteria, paper list, or data-extraction protocol to check that assumption.

Editorial extensions

If this is right

  • Learning-based MAPF papers that report results only on 10 to 100 agents are not yet comparable to classical scalability claims of 1000 or more agents.
  • A standardized benchmarking protocol with fixed map sizes, agent counts, metrics, and baseline algorithms would make reported performance meaningful across paradigms.
  • Hybrid solvers that insert learned modules into classical search or compilation pipelines can inherit classical scalability while gaining adaptability in dynamic environments.
  • Future work on many-agent pathfinding, neural solvers, and language-grounded MAPF should be measured against the hundred-to-thousand-agent classical scale.
  • The evaluation disparity suggests that claims of learning-based superiority in small grids do not transfer automatically to warehouse-scale deployment.

Reading between the lines

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

  • The scale gap may partly reflect training cost and the one-shot versus lifelong problem framing, factors the survey does not separate; a follow-up analysis controlling for those variables would clarify whether the gap is algorithmic or economic.
  • A concrete testable extension would be to take current learning-based policies and run them on the same 200 by 200, 1000-agent benchmarks used by classical solvers, reporting success rate, runtime, and solution cost.
  • If the gap is real, warehouse and logistics deployments of learning-based MAPF should be stress-tested at hundreds of agents before adoption, rather than extrapolating from small-grid results.
  • The survey's meta-analysis could be reproduced and verified if the authors released the full list of the 200-plus papers, their extracted settings, and the coding protocol.
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

3 major / 5 minor

Summary. The paper presents a broad survey of multi-agent pathfinding (MAPF), organizing the literature into search-based, compilation-based, and learning-based approaches, and it proposes a unified framework plus several illustrative pseudocode algorithms. Its most distinctive claim is a meta-analysis of experimental practices across more than 200 papers, asserting that classical methods are typically evaluated on much larger instances (up to 200x200 grids with 1000+ agents) than learning-based methods (predominantly 10-100 agents). The survey also discusses metrics, baselines, and future research directions such as mixed-motive MAPF, language-grounded planning, and neural solvers.

Significance. If the experimental meta-analysis were fully documented and reproducible, the paper would make a valuable contribution by quantifying evaluation disparities and motivating standardized MAPF benchmarking. The survey also usefully consolidates a large and fragmented literature, including recent hybrid and learning-augmented methods, and its taxonomy and pseudocode excerpts could serve as a practical reference. However, the central empirical claim currently rests on an unreleased corpus and unspecified extraction methodology, which substantially limits the paper's contribution as a survey with a data-driven message. The classical-algorithm review itself appears broadly sound and informative, though some algorithm descriptions contain technical errors that need correction.

major comments (3)
  1. [Abstract and §8] The central claim that classical MAPF evaluations reach 200x200 grids and 1000+ agents while learning-based evaluations are predominantly 10-100 agents is presented as the result of a 'systematic analysis' of 200+ papers, but the manuscript provides no list of the surveyed papers, no inclusion/exclusion criteria, no extraction schema, and no aggregation data or code. This claim is load-bearing for the paper's call for standardized benchmarking protocols, yet it cannot be independently verified or checked for robustness to venue selection, year range, problem variant (e.g., lifelong versus one-shot), or the classification of hybrid methods. I ask the authors to release the full paper list, the per-paper extracted experimental settings, and the aggregation code as supplementary material, and to report the raw counts underlying the scale-disparity statement. Without this, the abstract's quantitative claim is an unverifiable observation rather than a demonstrated result.
  2. [§3.1.3, Algorithm 1] The pseudocode for basic CBS contains a control-flow error: the final 'return No feasible solution' is placed inside the 'while' loop, immediately after the branching 'for' loop. As written, whenever the first conflict is found, the algorithm creates two child nodes and then unconditionally returns 'No feasible solution' without ever popping the next node from the open list. This would make the algorithm incomplete even for trivially solvable instances. The return statement should appear after the 'while' loop terminates (i.e., when the open list is exhausted). This is a technical correctness issue in a core algorithmic description of the survey.
  3. [§4.1.2, Algorithm 19] The sum-of-costs bound in the baseline SAT algorithm is inconsistent with the surrounding text. The text defines C_i,t as indicating that agent i has not yet reached its goal at time t, so summing C_i,t over t equals the agent's path cost; the total SoC is then bounded by xi_0 + Delta. Algorithm 19, however, writes the constraint as 'sum_{i,t} C_i,t <= Delta', which would force each agent to finish within its shortest-path length for Delta=0 and contradicts the earlier definition of xi = xi_0 + Delta. Either the definition of C_i,t or the constraint needs to be corrected; as written, the algorithm would not find optimal SoC solutions and the accompanying explanation is misleading.
minor comments (5)
  1. [§1, Figure 1] The text says 'MAPF sees practical deployment in an extensive set of application domains, as shown in Figure 1', but Figure 1 displays Google Scholar publication trends, not application domains. The reference should point to a figure that illustrates the cited application areas, or the sentence should be reworded.
  2. [§3.4, opening paragraph] The sentence 'The illustration of how agent-priority ordering affects the outcome of prioritized planning in PBS is shown in 8' is missing the word 'Figure' before the numeral '8'.
  3. [§4.4, final paragraph] The phrase 'adopting APS- or MaxSAT-specific toolchains' appears to contain a typo: 'APS' should likely be 'ASP', matching the discussion of Answer Set Programming in Section 4.3.2.
  4. [§3.3.1, Table 2] In the table row for ECBS, the notation 'w · Cost(opt)' is used without explicitly defining 'opt' as the optimal cost; this is a minor clarity issue for readers who encounter the table before reading the text.
  5. [§5.2.2] Several representative learning approaches (e.g., Zhang et al., 2022b; Wang et al., 2023; Yang et al., 2024) are described only briefly and without the full citation details in the visible text; please ensure all these entries appear in the bibliography with complete bibliographic information.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: this is a literature survey whose claims rest on cited prior work, not on a derivation that reintroduces its own inputs.

full rationale

This paper is a survey and meta-analysis of existing MAPF literature, not a derivation of new results. Its central empirical claim—that classical methods are typically evaluated on larger instances (up to 200×200 grids with 1000+ agents) than learning-based methods (predominantly 10–100 agents)—is presented as an observation about the surveyed papers, not as a conclusion derived from any fitted parameter or from the paper's own framework. The survey does not define its concepts in terms of the conclusion, does not fit a parameter and then rename it a prediction, and does not invoke a self-citation chain to force its taxonomy or its benchmarking recommendation. The fact that the 200+ paper corpus, inclusion criteria, and extraction code are not provided is a reproducibility limitation, but it is not circular reasoning: the claim is externally checkable in principle even if the underlying data are not currently released. Similarly, the survey's extensive use of citations to prior work is standard scholarly practice and does not constitute circularity unless a load-bearing premise is justified only by the authors' own unverified prior results, which is not the case here. Overall, the paper contains no derivation chain that reduces to its own inputs.

Assumptions & free parameters 0 free parameters · 2 assumptions · 0 invented entities

The survey does not introduce new free parameters or invented entities. It relies on standard domain assumptions about the MAPF formulation and the representativeness of the cited literature.

assumptions (2)
  • domain assumption The graph-based MAPF formulation in Section 2 is the mainstream representation.
    The survey organizes all methods around this formulation, which may not fully cover continuous or other variants. This is a reasonable assumption for a survey but may exclude some non-standard formulations.
  • domain assumption The 200+ papers analyzed are representative of the MAPF literature.
    No systematic selection methodology is given, so there is a risk of selection bias in the meta-analysis.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Where Paths Collide: A Comprehensive Survey of Classic and Learning-Based Multi-Agent Pathfinding." pith.science (2026). https://pith.science/paper/Z5TMGNZQ

@misc{pith2026250519219,
  author       = {Pith},
  title        = {Pith review of: Where Paths Collide: A Comprehensive Survey of Classic and Learning-Based Multi-Agent Pathfinding},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/Z5TMGNZQ}},
  note         = {Machine review of arXiv:2505.19219}
}
read the original abstract

Multi-Agent Path Finding (MAPF) is a fundamental problem in artificial intelligence and robotics, requiring the computation of collision-free paths for multiple agents navigating from their start locations to designated goals. As autonomous systems become increasingly prevalent in warehouses, urban transportation, and other complex environments, MAPF has evolved from a theoretical challenge to a critical enabler of real-world multi-robot coordination. This comprehensive survey bridges the long-standing divide between classical algorithmic approaches and emerging learning-based methods in MAPF research. We present a unified framework that encompasses search-based methods (including Conflict-Based Search, Priority-Based Search, and Large Neighborhood Search), compilation-based approaches (SAT, SMT, CSP, ASP, and MIP formulations), and data-driven techniques (reinforcement learning, supervised learning, and hybrid strategies). Through systematic analysis of experimental practices across 200+ papers, we uncover significant disparities in evaluation methodologies, with classical methods typically tested on larger-scale instances (up to 200 by 200 grids with 1000+ agents) compared to learning-based approaches (predominantly 10-100 agents). We provide a comprehensive taxonomy of evaluation metrics, environment types, and baseline selections, highlighting the need for standardized benchmarking protocols. Finally, we outline promising future directions including mixed-motive MAPF with game-theoretic considerations, language-grounded planning with large language models, and neural solver architectures that combine the rigor of classical methods with the flexibility of deep learning. This survey serves as both a comprehensive reference for researchers and a practical guide for deploying MAPF solutions in increasingly complex real-world applications.

Figures

Figures reproduced from arXiv: 2505.19219 by the authors.

Figure 1
Figure 1. The trends of the cumulative numbers of Google Scholar papers that contain the [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Structure diagram of the paper (Section 2 - Section 9). [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Illustration of (a): a MAPF instance and (b): different types of conficts. (a): the [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (18 more)
Figure 4
Figure 4. Figure 4: A chronological timeline of classical algorithms developed between 2012 and 2024, [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: Illustration of the CBS process on a simple two-agent instance. Left: Two related [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]
Figure 6
Figure 6. Figure 6: A 2 × 3 grid environment for the CBS example. Agent 1 aims to move from (0, 0) to (1, 2); Agent 2 aims to move from (1, 0) to (0, 2). We now provide a step-by-step example of how CBS resolves conflicts and arrives at a collision-free solution, using the small 2 × 3 gri…
Figure 7
Figure 7. Figure 7: An evolutionary graph of the research work conducted on CBS. Starting from the [PITH_FULL_IMAGE:figures/full_fig_p014_7.png]
Figure 8
Figure 8. Figure 8: Illustration of how agent-priority ordering affects the outcome of prioritized [PITH_FULL_IMAGE:figures/full_fig_p021_8.png]
Figure 9
Figure 9. Figure 9: Illustration of the core workflow in LNS. Left: all agents (or path segments) are [PITH_FULL_IMAGE:figures/full_fig_p028_9.png]
Figure 10
Figure 10. Figure 10: Illustration of LaCAM. Top left (Map & Instance): A 3 [PITH_FULL_IMAGE:figures/full_fig_p033_10.png]
Figure 11
Figure 11. Figure 11: A toy 2D grid snippet (6 vertices). Agent 1 must go from A0 to F0, while Agent 2 does exactly the reverse. We illustrate (A0 ↔ B ↔ C) on top row and (D ↔ E ↔ F0) below. Agents can also move vertically between the top and bottom rows. Note that A0 and F0 are effectivel…
Figure 12
Figure 12. Figure 12: An evolution diagram of learning-based MAPF algorithms from 2017 to 2024, [PITH_FULL_IMAGE:figures/full_fig_p050_12.png]
Figure 13
Figure 13. Figure 13: A general algorithmic flowchart for solving the MAPF problem using reinforce [PITH_FULL_IMAGE:figures/full_fig_p059_13.png]
Figure 14
Figure 14. Figure 14: A flowchart of solving the MAPF problem using Monte Carlo Tree Search [PITH_FULL_IMAGE:figures/full_fig_p069_14.png]
Figure 15
Figure 15. Figure 15: A supervised learning framework for solving the MAPF problem. The process [PITH_FULL_IMAGE:figures/full_fig_p072_15.png]
Figure 16
Figure 16. Figure 16: A set of four sub-flowcharts illustrating different hybrid approaches to solving [PITH_FULL_IMAGE:figures/full_fig_p073_16.png]
Figure 17
Figure 17. Figure 17: Representative heatmap (schematic) depicting experimental configurations of [PITH_FULL_IMAGE:figures/full_fig_p087_17.png]
Figure 18
Figure 18. Figure 18: Representative heatmap (schematic) depicting experimental configurations of [PITH_FULL_IMAGE:figures/full_fig_p088_18.png]
Figure 19
Figure 19. Figure 19: Histogram (schematic) illustrating the frequency of classical baseline algorithms [PITH_FULL_IMAGE:figures/full_fig_p089_19.png]
Figure 20
Figure 20. Figure 20: Histogram (schematic) illustrating the frequency of various baseline algorithms [PITH_FULL_IMAGE:figures/full_fig_p089_20.png]
Figure 21
Figure 21. Figure 21: Results of the evaluation of DCC algorithm in static and dynamic envirnments [PITH_FULL_IMAGE:figures/full_fig_p092_21.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

29 extracted references · 17 canonical work pages

  1. [5]

    Odrm* optimal multirobot path plan- ning in low dimensional search spaces

    Cornelia Ferner, Glenn Wagner, and Howie Choset. Odrm* optimal multirobot path plan- ning in low dimensional search spaces. In 2013 IEEE international conference on robotics and automation, pp. 3854–3859. IEEE,

    work page 2013

  2. [7]

    Social Behavior as a Key to Learning-based Multi-Agent Pathfinding Dilemmas

    Chengyang He, Tanishq Duhan, Parth Tulsyan, Patrick Kim, and Guillaume Sartoretti. Social behavior as a key to learning-based multi-agent pathfinding dilemmas. arXiv preprint arXiv:2408.03063, 2024a. Chengyang He, Tianze Yang, Tanishq Duhan, Yutong Wang, and Guillaume Sartoretti. Alpha: Attention-based long-horizon pathfinding in highly-structured areas. ...

    work page Pith review arXiv

  3. [8]

    Formalization of optimality conditions for smooth constrained optimization problems

    Chenyi Li, Shengyang Xu, Chumin Sun, Li Zhou, and Zaiwen Wen. Formalization of optimality conditions for smooth constrained optimization problems. arXiv preprint arXiv:2503.18821,

    arXiv

  4. [11]

    Evolution of heuristics: Towards efficient automatic algorithm design using large language model

    Fei Liu, Tong Xialiang, Mingxuan Yuan, Xi Lin, Fu Luo, Zhenkun Wang, Zhichao Lu, and Qingfu Zhang. Evolution of heuristics: Towards efficient automatic algorithm design using large language model. In ICML, 2024a. Qi Liu, Jianqi Gao, Dongjie Zhu, Zhongjian Qiao, Pengbin Chen, Jingxiang Guo, and Yanjie Li. Multi-agent target assignment and path finding for ...

    arXiv

  5. [12]

    Lifelong multi-agent path finding for online pickup and delivery tasks

    Hang Ma, Jiaoyang Li, TK Kumar, and Sven Koenig. Lifelong multi-agent path finding for online pickup and delivery tasks. arXiv:1705.10868, 2017a. Hang Ma, Jingxing Yang, Liron Cohen, TK Kumar, and Sven Koenig. Feasibility study: Moving non-homogeneous teams in congested video game environments. In AIIDE, 2017b. Hang Ma, Daniel Harabor, Peter J Stuckey, Ji...

    arXiv

  6. [13]

    Flatland-RL: Multi-agent reinforcement learning on trains

    Sharada Mohanty, Erik Nygren, Florian Laurent, Manuel Schneider, Christian Scheller, Nil- abha Bhattacharya, Jeremy Watson, Adrian Egli, Christian Eichenberger, Christian Baum- berger, et al. Flatland-RL: Multi-agent reinforcement learning on trains. arXiv:2012.05893,

    arXiv 2012

  7. [14]

    Improving LaCAM for Scalable Eventually Optimal Multi-Agent Pathfinding

    Keisuke Okumura. Improving lacam for scalable eventually optimal multi-agent pathfind- ing. arXiv:2305.03632, 2023a. Keisuke Okumura. LaCAM: Search-based algorithm for quick multi-agent pathfinding. In AAAI, 2023b. Keisuke Okumura. Engineering LaCAM*: Towards real-time, large-scale, and near-optimal multi-agent pathfinding. In AAMAS,

    work page Pith review arXiv

  8. [15]

    winPIBT: Extended Prioritized Algorithm for Iterative Multi-agent Path Finding

    Keisuke Okumura, Yasumasa Tamura, and Xavier D´efago. winpibt: Extended prioritized algorithm for iterative multi-agent path finding. arXiv preprint arXiv:1905.10149,

    work page Pith review arXiv 1905

Show all 29 references
  1. [17]

    Multi agent path finding using evolutionary game theory

    Sheryl Paul and Jyotirmoy V Deshmukh. Multi agent path finding using evolutionary game theory. arXiv:2212.02010,

    arXiv

  2. [18]

    Confidence-based curricu- lum learning for multi-agent path finding

    Thomy Phan, Joseph Driscoll, Justin Romberg, and Sven Koenig. Confidence-based curricu- lum learning for multi-agent path finding. arXiv preprint arXiv:2401.05860, 2024a. Thomy Phan, Taoan Huang, Bistra Dilkina, and Sven Koenig. Adaptive anytime multi-agent path finding using ...

    2011 arXiv

  3. [19]

    Multi-agent navigation based on deep reinforcement learning and traditional pathfinding algorithm

    Hongda Qiu. Multi-agent navigation based on deep reinforcement learning and traditional pathfinding algorithm. arXiv:2012.09134,

    2012 arXiv

  4. [20]

    Mapfast: A deep algorithm selector for multi agent path finding using shortest path embeddings

    Jingyao Ren, Vikraman Sathiyanarayanan, Eric Ewing, Baskin Senbaslar, and Nora Ayanian. Mapfast: A deep algorithm selector for multi agent path finding using shortest path embeddings. arXiv preprint arXiv:2102.12461,

    arXiv

  5. [21]

    Llmdr: Llm-driven dead- lock detection and resolution in multi-agent pathfinding

    Seungbae Seo, Junghwan Kim, Minjeong Shin, and Bongwon Suh. Llmdr: Llm-driven dead- lock detection and resolution in multi-agent pathfinding. arXiv preprint arXiv:2503.00717,

    arXiv

  6. [22]

    Automatic algorithm selection in multi-agent pathfinding

    Devon Sigurdson, Vadim Bulitko, Sven Koenig, Carlos Hernandez, and William Yeoh. Automatic algorithm selection in multi-agent pathfinding. arXiv preprint arXiv:1906.03992,

    1906 arXiv

  7. [23]

    POGEMA: A benchmark platform for cooperative multi-agent navigation

    Alexey Skrynnik, Anton Andreychuk, Anatolii Borzilov, Alexander Chernyavskiy, Kon- stantin Yakovlev, and Aleksandr Panov. POGEMA: A benchmark platform for cooperative multi-agent navigation. arXiv:2407.14931, 2024a. Alexey Skrynnik, Anton Andreychuk, Maria Nesterova, Konstanti...

    arXiv

  8. [25]

    Benchmarking large neighborhood search for multi-agent path finding

    Jiaqi Tan, Yudong Luo, Jiaoyang Li, and Hang Ma. Benchmarking large neighborhood search for multi-agent path finding. arXiv:2407.09451,

    arXiv

  9. [26]

    Ensembling prioritized hybrid policies for multi-agent pathfinding

    Huijie Tang, Federico Berto, and Jinkyoo Park. Ensembling prioritized hybrid policies for multi-agent pathfinding. arXiv:2403.07559, 2024a. Jingtao Tang, Zining Mao, and Hang Ma. Large-scale multi-robot coverage path planning on grids with path deconfliction. arXiv:2411.01707,...

    arXiv

  10. [27]

    M*: A complete multirobot path planning algorithm with performance bounds

    Glenn Wagner and Howie Choset. M*: A complete multirobot path planning algorithm with performance bounds. In 2011 IEEE/RSJ international conference on intelligent robots and systems, pp. 3260–3267. IEEE,

    2011

  11. [28]

    LNS2+ RL: Combining multi-agent reinforcement learning with large neighborhood search in multi-agent path finding

    Yutong Wang, Tanishq Duhan, Jiaoyang Li, and Guillaume Sartoretti. LNS2+ RL: Combining multi-agent reinforcement learning with large neighborhood search in multi-agent path finding. arXiv:2405.17794, 2024d. Mengdie Wu, Wenyao Yan, Haslin Hasan, and Robiatul A’Dawiah Jamaluddin...

    arXiv

  12. [29]

    Perceive, reflect, and plan: Designing llm agent for goal-directed city navigation without instructions

    Qingbin Zeng, Qinglong Yang, Shunan Dong, Heming Du, Liang Zheng, Fengli Xu, and Yong Li. Perceive, reflect, and plan: Designing llm agent for goal-directed city navigation without instructions. arXiv preprint arXiv:2408.04168,

    arXiv

  13. [2016]

    Formal verification of piece-wise linear feed-forward neural networks

    Ruediger Ehlers. Formal verification of piece-wise linear feed-forward neural networks. In ATV A 2017,

    2017

  14. [2017]

    Scalable mechanism design for multi-agent path finding

    Paul Friedrich, Yulun Zhang, Michael Curry, Ludwig Dierks, Stephen McAleer, Jiaoyang Li, Tuomas Sandholm, and Sven Seuken. Scalable mechanism design for multi-agent path finding. arXiv preprint arXiv:2401.17044,

  15. [2018]

    Conflict handling framework in generalized multi-agent path finding: Advantages and shortcomings of satisfiability modulo approach

    Pavel Surynek. Conflict handling framework in generalized multi-agent path finding: Advantages and shortcomings of satisfiability modulo approach. In ICAART, 2019a. Pavel Surynek. Multi-agent path finding with continuous time viewed through satisfiability modulo theories (smt)...

    1903 arXiv

  16. [2019]

    Offline time- independent multi-agent path planning

    Keisuke Okumura, Franc ¸ois Bonnet, Yasumasa Tamura, and Xavier D´efago. Offline time- independent multi-agent path planning. arXiv:2105.07132, 2021a. Keisuke Okumura, Yasumasa Tamura, and Xavier D´efago. Iterative refinement for real-time multi-robot path planning. In IROS, 2...

    arXiv

  17. [2020]

    Anytime multi- agent path finding via large neighborhood search

    Jiaoyang Li, Zhe Chen, Daniel Harabor, Peter J Stuckey, and Sven Koenig. Anytime multi- agent path finding via large neighborhood search. In IJCAI, 2021a. Jiaoyang Li, Zhe Chen, Yi Zheng, Shao-Hung Chan, Daniel Harabor, Peter J Stuckey, Hang Ma, and Sven Koenig. Scalable rail ...

    2020

  18. [2021]

    Accessed: 2024-12-03

    URL https://www.politesi.polimi.it/bitstream/ 10589/191682/4/Tesi Bignoli.pdf. Accessed: 2024-12-03. Eli Boyarski, Ariel Felner, Guni Sharon, and Roni Stern. Don’t split, try to work it out: Bypassing conflicts in multi-agent pathfinding. In ICAPS, 2015a. Eli Boyarski, Ariel F...

    2024

  19. [2022]

    Towards deployment-efficient and collision-free multi-agent path finding (student abstract)

    Feng Chen, Chenghe Wang, Fuxiang Zhang, Hao Ding, Qiaoyong Zhong, Shiliang Pu, and Zongzhang Zhang. Towards deployment-efficient and collision-free multi-agent path finding (student abstract). In AAAI, 2023a. 101 Preprint. Under review. Lin Chen, Yaonan Wang, Zhiqiang Miao, Ya...

    arXiv

  20. [2023]

    Multi-agent path finding in continuous spaces with projected diffusion models

    Jinhao Liang, Jacob K Christopher, Sven Koenig, and Ferdinando Fioretto. Multi-agent path finding in continuous spaces with projected diffusion models. arXiv preprint arXiv:2412.17993,

    arXiv

  21. [2024]

    Mapfaster: A faster and simpler take on multi-agent path finding algorithm selection

    Jean-Marc Alkazzi, Anthony Rizk, Michel Salomon, and Abdallah Makhoul. Mapfaster: A faster and simpler take on multi-agent path finding algorithm selection. In 2022 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pp. 10088–10093. IEEE,

    2022

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.