Pith. sign in

REVIEW 4 major objections 5 minor 39 references

From Discrete Plans to Real-World Execution: A World-Model-Driven Framework for Execution-Aware Multi-Agent Path Finding

T0 review · 4 major / 5 minor · reviewed 2026-08-03 · deepseek-v4-flash

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

Pith's one-line read This paper claims that using a learned model of real-world execution time to guide multi-agent path planning cuts deadline violations by up to 20% compared to constant-speed baselines.

desk verdict The abstract and full text describe different papers, and the body's headline 20% gain is only tested on training maps; the core idea is still worth referee time. read the letter →

arxiv 2511.21886 v2 pith:AQ3I2BLN submitted 2025-11-26 cs.RO cs.AI

Jingtian Yan , Shuai Zhou , He Jiang , Stephen F. Smith , Jiaoyang Li This is my paper · ORCID
classification cs.ROcs.AI
keywords multi-agentpathfindingexecutiontimepredictionactiondependencygraphdeadline-awareplanningneuralnetworksworldmodelwarehouseroboticsMAPF-RD
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 tries to close the gap between discrete multi-agent path finding (MAPF) and physical execution. It introduces a neural network, ExecTimeNet, that takes the dependency graph of a candidate plan and predicts each agent's wall-clock completion time. These predictions are fed back into search-based planners so they can optimize for real-world deadlines rather than abstract path length. In the paper's simulations with up to 300 agents, this execution-aware search improves solution quality by up to 20% over baselines that assume a constant execution speed. The significance is that a single learned model, trained once, generalizes to unseen maps and agent counts, making deadline-aware warehouse-scale planning feasible.

What carries the argument

The Action Dependency Graph (ADG) encoding, augmented with node and edge features, is the central object. It turns a discrete MAPF plan into a graph whose nodes are actions and whose edges are intra-agent sequencing and inter-agent passing constraints. ExecTimeNet—a Transformer plus graph-attention model—processes this graph to predict per-agent wall-clock execution times, capturing how delays propagate across agents. The ADG-based representation is map-agnostic, which is what allows the model to transfer to unseen maps and agent counts.

What would settle it

Run the same REMAP evaluation on physical robots with the same maps, agent counts, and deadlines, and compare realized penalties against the constant-speed baselines; if the improvement over baselines disappears or reverses, the claim that learned execution-time prediction transfers to reality is falsified.

Watch

Extended reading notes

Core claim

The central claim is that execution time in MAPF is predictable from the structure of the plan itself. Given a plan, the Action Dependency Graph encodes both the order of each agent's actions and the passing order of agents at shared locations. ExecTimeNet reads this graph and outputs either a point estimate or a distribution of each agent's completion time. REMAP uses these estimates to compute deadline penalties and guides the search of two leading search-based planners, producing plans that, when re-simulated, incur up to 20% lower penalty than plans chosen by constant-speed execution estimates.

Load-bearing premise

The results depend on the simulator used for both training and evaluation faithfully reproducing the physical execution of robots; if that simulator's speed, acceleration, and controller models diverge from real hardware, the predicted improvements may not transfer to deployed fleets.

Editorial extensions

If this is right

  • Deadline-aware planning can be added to existing search-based MAPF solvers with modest changes, without redesigning the planner.
  • A single trained model can serve fleets of varying sizes and unseen warehouse layouts, removing the need to re-calibrate per map.
  • Distributional predictions allow planners to hedge against execution uncertainty, not just optimize the expected time.
  • The runtime cost of ADG construction and inference is small enough to support many search refinement iterations within a minute budget.
  • The approach extends beyond deadline satisfaction to any time-sensitive multi-agent scheduling problem where execution times are hard to model analytically.

Reading between the lines

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

  • The paper's evaluation uses the same simulator for both training and testing; a fair transfer test would require a different simulator or physical robots to confirm the 20% improvement persists.
  • The paper's abstract describes a post-planning procedure and physical-hardware results that are not present in the full text; the body's evidence is simulation-only.
  • The deadline protocol is tuned so that roughly half the agents pass; a natural extension is to evaluate under tight-deadline regimes where the planning-execution gap matters most.
  • Because the ADG encoding is map-agnostic but trained on plans from one planner family, one could test whether the model retains accuracy when plans come from a different planner or with different robot dynamics.
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

4 major / 5 minor

Summary. The manuscript (full text) proposes REMAP, a planning framework for MAPF with Real-world Deadlines (MAPF-RD). REMAP wraps existing search-based MAPF planners (MAPF-LNS and CBS) with a learned execution-time predictor, ExecTimeNet, which takes an Action Dependency Graph (ADG) encoding of a candidate plan and predicts per-agent wall-clock completion times (or distributions thereof). These predictions are used to compute deadline penalties and guide the search. ExecTimeNet is trained on SMART simulator traces generated from six MovingAI maps, and REMAP is evaluated by re-simulating its output plans in SMART. The paper reports up to a 20% improvement in solution quality (penalty gap relative to a Virtual Best Solver) over constant-speed baselines on up to 300 agents. The supplied arXiv abstract additionally claims an ESADG post-processing method, up to 40% improvement, and 15.3% total execution-time reduction on physical hardware, but none of these appear in the full text, which is a REMAP-only simulation study.

Significance. If the central claim holds, the paper's contribution is useful: it shows that a learned surrogate of MAPF execution, taking an ADG structure as input, can act as a search signal and improve deadline-related objectives over simple constant-speed estimates. The paper includes several strengths: a fixed-iteration control in Fig. 4, which mitigates the concern that the performance gap is only an artifact of differing iteration counts; a feature ablation in Table II; a distributional variant with a proper likelihood loss; and a careful discussion of CBS non-optimality in Appendix E. However, the significance is currently limited by the absence of any planner-level out-of-distribution evaluation, by the closed SMART-only loop, and by the fact that the abstract promises hardware/ESADG results that are not present in the body.

major comments (4)
  1. [Abstract vs. Sections I–V] The supplied abstract claims 'ESADG ... up to 40% normalized improvement' and 'on physical hardware ... reduces total execution time by up to 15.3%'. The full text contains no ESADG, no hardware experiment, and no such results. The body's conclusion (Section V) claims only 'up to 20%' improvement in 'realistic simulations'. This is a load-bearing mismatch: the advertised contribution is absent from the manuscript. The authors must either include the ESADG and hardware experiments, or revise the abstract to match the REMAP-only simulation content.
  2. [Section IV-B vs. Section V] The planner-level evaluation of REMAP is performed only on the six-map pool (empty, maze, random, room, den312d, WH-S) that was used to train ExecTimeNet in Section IV-A1. The two out-of-distribution maps, WH-L and boston, appear only in Table I for prediction MAPE, not in the REMAP planner comparisons. The claim in Section V that the model 'generalizes across ... unseen maps' is therefore supported only at the prediction level, not at the task level. The headline 20% improvement could be an in-distribution artifact. This is fixable: report REMAP penalty-gap results on WH-L and boston (or explicitly scope the planner claim to in-distribution maps).
  3. [Section IV-B1 and Appendix F] The deadline regime is selected post hoc: Appendx F describes a grid search over K_D (8 to 16, step 2) and chooses the value that gives roughly 50% of agents meeting their deadlines on each map/agent-count. This makes the comparison regime-dependent, and it is unclear how sensitive the reported gaps are to this choice. The authors should justify that 50% pass rate is a principled operating point a priori and, at minimum, report results for neighboring K_D values. Without this, the headline improvement may be concentrated in a deliberately selected easy/hard regime.
  4. [Section IV-A1 and Section IV-B] The empirical pipeline is closed entirely inside the SMART simulator: ExecTimeNet is trained on SMART labels, the planner optimizes against ExecTimeNet's surrogate of SMART, and the final evaluation re-simulates in SMART. This is not circular by equation, but it leaves the central premise—that optimizing against a learned model of SMART transfers to physical robots—unsupported. The abstract promises 'effective transfer from simulation to real-world deployment' with a 15.3% hardware figure, but no real-robot experiment appears. Either add hardware validation or explicitly restrict the paper's claims to the SMART simulator and remove the hardware/transfer language.
minor comments (5)
  1. [Section II-A] Definition 2 contains a typo: 'communication delays).' has an unbalanced parenthesis. Also, the end of the definition should be punctuated consistently.
  2. [Section IV-B1] The normalization with VBS is described as 'results are normalized using a Virtual Best Solver ... divided by the number of agents', but the exact formula for the 'penalty gap' is not stated. Please write the normalization explicitly to allow reproduction.
  3. [Figures 3–5] No error bars or confidence intervals are shown for the penalty-gap curves. Given only 25 instances per map and intrinsic SMART variability, reporting standard errors or confidence bands, at least for the headline comparisons, would strengthen the presentation.
  4. [Title and front matter] The document's internal title is 'Bridging Planning and Execution: Multi-Agent Path Finding Under Real-World Deadlines', which differs from the arXiv title 'From Discrete Plans to Real-World Execution: A World-Model-Driven Framework for Execution-Aware Multi-Agent Path Finding'. The authors should align these and ensure the abstract matches the body.
  5. [Appendix B] The distributional head is described as modeling a Gaussian in log-space, but the main text (Section III-B2) calls it 'log-normal'. Please reconcile the terminology—this is the same model, but the presentation is confusing.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: ExecTimeNet is a fitted surrogate, but the reported planner improvement is measured against a shared SMART execution oracle, not by construction.

full rationale

The paper's derivation chain is a standard surrogate-based optimization loop: ExecTimeNet is trained on ADG features mapped to SMART execution-time labels; REMAP uses ExecTimeNet's predicted times only as a scoring signal inside CBS/MAPF-LNS; final penalties are then computed by re-simulating the chosen plans in SMART. There is no equation in the paper that sets the predicted execution time equal to the SMART-measured time by construction, nor does the search objective force a plan to have a low SMART penalty tautologically. The constant-speed baselines are also evaluated in the same SMART simulator, and their K_u values are calibrated to SMART statistics, which makes the comparison fair rather than circular. The grid-search selection of K_D applies uniformly to all methods and only chooses a balanced deadline regime. The main self-citation is SMART [33], the authors' own simulation testbed; this is an external-validity concern for claims about physical robots, especially since the separate arXiv abstract mentions hardware results not present in the body, but it does not make the simulation-based 20% improvement reduce to its inputs. The central claim is therefore empirically meaningful within the stated simulation setting, and no specific circular step can be exhibited.

Assumptions & free parameters 3 free parameters · 7 assumptions · 1 invented entities

The central claim rests on a chain of premises that the reader must accept from the authors' own prior work: the ADG execution semantics (refs [1],[29]), the SMART simulator's fidelity (ref [33], same authors), and the deadline generation protocol (K_D tuned in Appendix F). ExecTimeNet's learned weights are fitted to SMART labels, so the planner's objective is calibrated to the same simulator used for evaluation. The only externally grounded elements are the discrete MAPF problem, the CBS/LNS algorithm families, and the MovingAI benchmark maps.

free parameters (3)
  • ExecTimeNet weights = 325,602 parameters (Appendix B); trained on 9,433 SMART execution traces
    The planner's execution-time objective is determined entirely by this learned fit to the authors' simulator; the predictor's accuracy (Table I) is the foundation of the approach.
  • K_D (deadline scaling factor) = grid-searched over {8,10,12,14,16} per map and agent count
    Chosen in Appendix F so that ~50% of agents meet deadlines; this post-hoc regime selection affects all reported penalty results, though equally across methods.
  • K_u (baseline constant-speed factors) = 0.1, 0.05, 0.03; 0.05 = mean path-length/execution-time ratio in SMART
    ConstExec baselines are calibrated to SMART statistics (Section IV-B1), so the comparison is against a well-tuned constant rather than a naive one.
assumptions (7)
  • domain assumption ADG construction from a MAPF plan guarantees collision-free asynchronous execution (Type-1 and Type-2 precedence edges)
    ExecTimeNet's input graph and the semantics of 'execution' both come from the ADG formalism of refs [1],[29]; Section III-B1 builds on it without derivation.
  • domain assumption SMART's execution model represents real warehouse robot execution
    SMART (ref [33], same authors) is the sole ground truth for training labels and evaluation (Section IV); no physical-robot validation appears in the full text.
  • domain assumption Discrete grid actions map one-to-one to executable robot actions (move/wait/rotate with kinodynamic cost captured by SMART)
    The 11-dimensional action features (Section III-B1) assume this mapping holds; turns and rotations are treated as actions with real costs only inside SMART.
  • domain assumption Predictor trained on six maps generalizes to unseen maps and agent counts
    Table I supports this empirically (OOD MAPE 4-8%), but planner-level gains on WH-L/boston rest on the same assumption unchecked at the planning level.
  • ad hoc to paper Execution times are log-normally distributed for the distributional variant
    Section III-B2 postulates log-normal 'because execution times are strictly positive and often right-skewed'; no goodness-of-fit test is reported.
  • standard math Standard MAPF theory for CBS optimality and LNS anytime refinement
    CBS/LNS background (refs [8],[11]) is assumed; the paper itself shows in Appendix E that CBS optimality no longer holds under REMAP.
  • domain assumption Small MAPE of ExecTimeNet implies planner decisions are not systematically misled
    No analysis of how prediction errors propagate into CBS/LNS penalty ordering; Appendix E concedes prediction inaccuracy contributes to suboptimality.
invented entities (1)
  • ExecTimeNet
    purpose: Learned world model (Transformer + GATv2 over ADG) predicting each agent's wall-clock execution time or its log-normal distribution from a discrete MAPF plan, used to score planner solutions
    Its only validation is against SMART, the same simulator that generated its training labels (Section IV); the abstract's physical-robot evidence is absent from the body, so there is no external falsifiable handle in this version.

how reviews work

0 comments
Cite this review

Pith. "Pith review of From Discrete Plans to Real-World Execution: A World-Model-Driven Framework for Execution-Aware Multi-Agent Path Finding." pith.science (2026). https://pith.science/paper/AQ3I2BLN

@misc{pith2026251121886,
  author       = {Pith},
  title        = {Pith review of: From Discrete Plans to Real-World Execution: A World-Model-Driven Framework for Execution-Aware Multi-Agent Path Finding},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/AQ3I2BLN}},
  note         = {Machine review of arXiv:2511.21886}
}
read the original abstract

Multi-Agent Path Finding (MAPF) studies how to coordinate multiple agents to reach their goals without collisions and underpins a range of large-scale robotic systems, including automated warehousing and manufacturing. Recent advances enable MAPF solvers to compute high-quality plans for hundreds of agents. However, these plans are generated using simplified robot models with discretized time and action spaces. When they are deployed in physical systems, heterogeneous robot dynamics, asynchronous interactions, communication delays, and other real-world factors can lead to substantial deviations from planned performance. We bridge the gap between discrete planning and real-world execution through ExecTimeNet, a learned world model of MAPF execution that predicts how a discrete MAPF solution will unfold on physical robots, mapping each discrete action to its realized execution state, including its wall-clock completion time and the kinodynamic state in which it ends. Building on this capability, we first propose REMAP, an execution-aware MAPF framework that integrates execution-time estimation into planning, guiding the search toward MAPF solutions with improved execution performance. We also introduce ESADG, a post-planning optimization procedure that optimizes the execution schedule of a given MAPF solution while preserving path feasibility. We evaluate proposed frameworks in high-fidelity simulation with up to 300 agents and on physical robots. In simulation, ExecTimeNet predicts the execution state accurately and transfers to unseen maps and agent counts. Across simulation benchmarks spanning diverse map topologies, REMAP reduces delays by up to 21% over baselines, while ESADG achieves up to 40% normalized improvement. On physical hardware, the full pipeline reduces total execution time by up to 15.3%, demonstrating effective transfer from simulation to real-world deployment.

Figures

Figures reproduced from arXiv: 2511.21886 by the authors.

Figure 1
Figure 1. Overview of REMAP. Updated node features GATv2 Encoder Aggregated inter-agent features MLP MLP Execution time mean Execution time variance (Optional) Final prediction Intra-Agent Temporal Encoder Transformer Encoders Inter-Agent Graph Encoder Transformer Encoders Mean pooling Execution Time Head Embedding [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Model structure of ExecTimeNet. Pink arrows indicate the direction of information propagation. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Comparison of the linear and percentage penalty on MAPF-LNS. The terms “Linear” and “Percentage” in parentheses [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Comparison of the linear penalty on MAPF-LNS with [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: Comparison of the linear penalty on CBS. [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: Detailed model structure of ExecTimeNet (dist). [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]
Figure 8
Figure 8. Figure 8: Results on MAPF-LNS with different neighborhood [PITH_FULL_IMAGE:figures/full_fig_p010_8.png]
Figure 9
Figure 9. Figure 9: Toy example to show that CBS under MAPF-RD is [PITH_FULL_IMAGE:figures/full_fig_p010_9.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

39 extracted references · 4 linked inside Pith

  1. [1]

    Persistent and robust execution of MAPF schedules in warehouses,

    W. H ¨onig, S. Kiesel, A. Tinka, J. W. Durham, and N. Ayanian, “Persistent and robust execution of MAPF schedules in warehouses,” IEEE Robotics and Automation Letters, vol. 4, no. 2, pp. 1125–1131, 2019

    2019

  2. [2]

    Intersection coordination with priority-based search for autonomous vehicles,

    J. Li, T. A. Hoang, E. Lin, H. L. Vu, and S. Koenig, “Intersection coordination with priority-based search for autonomous vehicles,” in Proceedings of the AAAI Conference on Artificial Intelligence, vol. 37, 2023, pp. 11 578–11 585

    2023

  3. [3]

    Planning, scheduling and monitoring for airport surface operations,

    R. Morris, C. S. Pasareanu, K. Luckow, W. Malik, H. Ma, T. S. Kumar, and S. Koenig, “Planning, scheduling and monitoring for airport surface operations,” inAAAI Workshop: Planning for Hybrid Systems, 2016

    2016

  4. [4]

    Deadline-aware multi-agent tour planning,

    T. Huang, V . Shivashankar, M. Caldara, J. W. Durham, J. Li, B. Dilkina, and S. Koenig, “Deadline-aware multi-agent tour planning,” inPro- ceedings of the International Conference on Automated Planning and Scheduling, 2023, pp. 189–197

    2023

  5. [5]

    Multi-agent path finding with deadlines,

    H. Ma, G. Wagner, A. Felner, J. Li, T. K. S. Kumar, and S. Koenig, “Multi-agent path finding with deadlines,” inProceedings of the Inter- national Joint Conference on Artificial Intelligence, 2018, pp. 417–423

    2018

  6. [6]

    Multi-agent path finding with kinematic constraints,

    W. H ¨onig, T. K. S. Kumar, L. Cohen, H. Ma, H. Xu, N. Ayanian, and S. Koenig, “Multi-agent path finding with kinematic constraints,” in Proceedings of the International Conference on Automated Planning and Scheduling, vol. 26, 2016, pp. 477–485

    2016

  7. [7]

    Multi-agent motion planning for differential drive robots through stationary state search,

    J. Yan and J. Li, “Multi-agent motion planning for differential drive robots through stationary state search,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 39, no. 22, 2025, pp. 23 360– 23 368

    2025

  8. [8]

    Conflict-based search for optimal multi-agent pathfinding,

    G. Sharon, R. Stern, A. Felner, and N. R. Sturtevant, “Conflict-based search for optimal multi-agent pathfinding,”Artificial Intelligence, vol. 219, pp. 40–66, 2015

    2015

Show all 39 references
  1. [9]

    EECBS: A bounded-suboptimal search for multi-agent path finding,

    J. Li, W. Ruml, and S. Koenig, “EECBS: A bounded-suboptimal search for multi-agent path finding,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 35, 2021, pp. 12 353–12 362

    2021

  2. [10]

    Improved heuristics for conflict-based search for multi-agent path finding,

    J. Li, A. Felner, E. Boyarski, H. Ma, and S. Koenig, “Improved heuristics for conflict-based search for multi-agent path finding,” inProceedings of the International Joint Conference on Artificial Intelligence, 2019, pp. 442–449

    2019

  3. [11]

    Anytime multi-agent path finding via large neighborhood search,

    J. Li, Z. Chen, D. Harabor, P. J. Stuckey, and S. Koenig, “Anytime multi-agent path finding via large neighborhood search,” inProceedings of the International Joint Conference on Artificial Intelligence, 2021, pp. 4127–4135

    2021

  4. [12]

    Anytime multi-agent path finding via machine learning-guided large neighborhood search,

    T. Huang, J. Li, S. Koenig, and B. Dilkina, “Anytime multi-agent path finding via machine learning-guided large neighborhood search,” in Proceedings of the AAAI Conference on Artificial Intelligence, 2022, pp. 9368–9376

    2022

  5. [13]

    MAPF- LNS2: Fast repairing for multi-agent path finding via large neighborhood search,

    J. Li, Z. Chen, D. Harabor, P. J. Stuckey, and S. Koenig, “MAPF- LNS2: Fast repairing for multi-agent path finding via large neighborhood search,” inProceedings of the AAAI Conference on Artificial Intelli- gence, vol. 36, 2022, pp. 10 256–10 265

    2022

  6. [14]

    Primal: Pathfinding via reinforcement and imitation multi- agent learning,

    G. Sartoretti, J. Kerr, Y . Shi, G. Wagner, T. K. S. Kumar, S. Koenig, and H. Choset, “Primal: Pathfinding via reinforcement and imitation multi- agent learning,”IEEE Robotics and Automation Letters, vol. 4, no. 3, pp. 2378–2385, 2019

    2019

  7. [15]

    Primal 2: Pathfinding via reinforcement and imitation multi-agent learning - lifelong,

    M. Damani, Z. Luo, E. Wenzel, and G. Sartoretti, “Primal 2: Pathfinding via reinforcement and imitation multi-agent learning - lifelong,”IEEE Robotics and Automation Letters, vol. 6, no. 2, pp. 2666–2673, 2021

    2021

  8. [16]

    Priority inheri- tance with backtracking for iterative multi-agent path finding,

    K. Okumura, M. Machida, X. D ´efago, and Y . Tamura, “Priority inheri- tance with backtracking for iterative multi-agent path finding,”Artificial Intelligence, vol. 310, p. 103752, 2022

    2022

  9. [17]

    Multi-robot path planning with due times,

    H. Wang and W. Chen, “Multi-robot path planning with due times,” IEEE Robotics and Automation Letters, vol. 7, no. 2, pp. 4829–4836, 2022

    2022

  10. [18]

    Scalable rail planning and replanning: Winning the 2020 flatland challenge,

    J. Li, Z. Chen, Y . Zheng, S.-H. Chan, D. Harabor, P. J. Stuckey, H. Ma, and S. Koenig, “Scalable rail planning and replanning: Winning the 2020 flatland challenge,” inProceedings of the International Conference on Automated Planning and Scheduling, 2021, pp. 477–485

    2020

  11. [19]

    Anonymous multi-agent path finding with individual deadlines,

    G. Fine, D. Atzmon, and N. Agmon, “Anonymous multi-agent path finding with individual deadlines,” inProceedings of the International Conference on Autonomous Agents and Multi-agent Systems, 2023, pp. 869–877

    2023

  12. [20]

    Conflict-based search for multi-robot motion planning with kinodynamic constraints,

    J. Kottinger, S. Almagor, and M. Lahijanian, “Conflict-based search for multi-robot motion planning with kinodynamic constraints,” in Proceedings of the IEEE International Conference on Intelligent Robots and Systems, 2022, pp. 13 494–13 499

    2022

  13. [21]

    Temporal rea- soning with kinodynamic networks,

    H. Zhang, N. Tiruviluamala, S. Koenig, and T. S. Kumar, “Temporal rea- soning with kinodynamic networks,” inProceedings of the International Conference on Automated Planning and Scheduling, vol. 31, 2021, pp. 415–425

    2021

  14. [22]

    Provably safe and deadlock- free execution of multi-robot plans under delaying disturbances,

    M. ˇC´ap, J. Gregoire, and E. Frazzoli, “Provably safe and deadlock- free execution of multi-robot plans under delaying disturbances,” in Proceedings of the IEEE International Conference on Intelligent Robots and Systems, 2016, pp. 5113–5118

    2016

  15. [23]

    Congestion prediction for large fleets of mobile robots,

    G. Yu and M. T. Wolf, “Congestion prediction for large fleets of mobile robots,” inProceedings of the IEEE International Conference on Robotics and Automation, 2023, pp. 7642–7649

    2023

  16. [24]

    Online re-planning and adaptive parameter update for multi-agent path finding with stochas- tic travel times,

    A. Kita, N. Suenari, M. Okada, and T. Taniguchi, “Online re-planning and adaptive parameter update for multi-agent path finding with stochas- tic travel times,”arXiv preprint, vol. arXiv:2302.01489, 2023

    2023 arXiv

  17. [25]

    Spatial-temporal synchronous graph convolutional networks: A new framework for spatial-temporal network data forecasting,

    C. Song, Y . Lin, S. Guo, and H. Wan, “Spatial-temporal synchronous graph convolutional networks: A new framework for spatial-temporal network data forecasting,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 34, no. 01, 2020, pp. 914–921

    2020

  18. [26]

    Enough waiting for the couriers: Learning to estimate package pick-up arrival time from couriers’ spatial-temporal behaviors,

    H. Wen, Y . Lin, F. Wu, H. Wan, Z. Sun, T. Cai, H. Liu, S. Guo, J. Zheng, C. Songet al., “Enough waiting for the couriers: Learning to estimate package pick-up arrival time from couriers’ spatial-temporal behaviors,” ACM Transactions on Intelligent Systems and Technology, vol....

    2023

  19. [27]

    ETA prediction with graph neural networks in Google Maps,

    A. Derrow-Pinion, J. She, D. Wong, O. Lange, T. Hester, L. Perez, M. Nunkesser, S. Lee, X. Guo, B. Wiltshireet al., “ETA prediction with graph neural networks in Google Maps,” inProceedings of the 30th ACM International Conference on Information & Knowledge Management, 2021, p...

    2021

  20. [28]

    Subdimensional expansion for multirobot path planning,

    G. Wagner and H. Choset, “Subdimensional expansion for multirobot path planning,”Artificial Intelligence, vol. 219, pp. 1–24, 2015

    2015

  21. [29]

    Streamlining the action dependency graph framework: Two key enhancements,

    J. Dunkel, “Streamlining the action dependency graph framework: Two key enhancements,”arXiv preprint, vol. arXiv:2412.01277, 2024

    2024 arXiv

  22. [30]

    Attention is all you need,

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, Ł. Kaiser, and I. Polosukhin, “Attention is all you need,” inAdvances in Neural Information Processing Systems, 2017, pp. 5998–6008

    2017

  23. [31]

    How attentive are graph attention networks?

    S. Brody, U. Alon, and E. Yahav, “How attentive are graph attention networks?” inInternational Conference on Learning Representations, 2022

    2022

  24. [32]

    What uncertainties do we need in Bayesian deep learning for computer vision?

    A. Kendall and Y . Gal, “What uncertainties do we need in Bayesian deep learning for computer vision?” inAdvances in Neural Information Processing Systems, vol. 30, 2017, pp. 5574–5584

    2017

  25. [33]

    Advancing mapf towards the real world: A scalable multi-agent realistic testbed (smart),

    J. Yan, Z. Li, W. Kang, Y . Zhang, S. Smith, and J. Li, “Advancing mapf towards the real world: A scalable multi-agent realistic testbed (smart),” arXiv preprint, vol. arXiv:2503.04798, 2025

    2025 arXiv

  26. [34]

    Multi-agent pathfinding: Definitions, variants, and benchmarks,

    R. Stern, N. R. Sturtevant, A. Felner, S. Koenig, H. Ma, T. T. Walker, J. Li, D. Atzmon, L. Cohen, T. K. S. Kumar, E. Boyarski, and R. Bart ´ak, “Multi-agent pathfinding: Definitions, variants, and benchmarks,” in Proceedings of the International Symposium on Combinatorial Sea...

    2019

  27. [35]

    Reliable and efficient multi-agent coordination via graph neural net- work variational autoencoders,

    Y . Meng, N. Majcherczyk, W. Liu, S. Kiesel, C. Fan, and F. Pecora, “Reliable and efficient multi-agent coordination via graph neural net- work variational autoencoders,”arXiv preprint, vol. arXiv:2503.02954, 2025

    2025 arXiv

  28. [36]

    An adaptive large neighborhood search heuristic for the pickup and delivery problem with time windows,

    S. Ropke and D. Pisinger, “An adaptive large neighborhood search heuristic for the pickup and delivery problem with time windows,” Transportation Science, vol. 40, no. 4, pp. 455–472, 2006. SHELLet al.: A SAMPLE ARTICLE USING IEEETRAN.CLS FOR IEEE JOURNALS 9 VI. APPENDIX A. Da...

    2006

  29. [37]

    Adaptive Neighborhood:The original MAPF-LNS em- ploys a set of three neighborhood and selects one of them in each iteration, namely an agent-based neighborhood, a map- based neighborhood and a random neighborhood. The agent- based neighborhood generation includes a seed agenta...

    2021

  30. [38]

    We define a set of deadline violation countsV C={vc 1, vc2,

    Failure-based Neighborhood:Failure-based neighbor- hood targets agents that frequently violate deadlines. We define a set of deadline violation countsV C={vc 1, vc2, . . . , vcM }, initially set to zero for all agents. In each iteration, after obtaining predicted execution tim...

  31. [39]

    However, path 2 requires fewer turns, leading to fewer acceleration and deceleration phases than path 1

    When CBS resolves a conflict at the red X location in path 1, the low-level planner selects path 2 for the robot, which has the same length as path 1. However, path 2 requires fewer turns, leading to fewer acceleration and deceleration phases than path 1. This decreases execut...

Pith tools

Reviewed August 3, 2026 · model on record in the stance chip above.