A Scalable Post-Processing Pipeline for Large-Scale Free-Space Multi-Agent Path Planning with PiBT

Arjo Chakravarty; M. A. Viraj J. Muthugala; Michael X. Grey; Mohan Rajesh Elara

REVIEW 4 major objections 5 minor 26 references

A Scalable Post-Processing Pipeline for Large-Scale Free-Space Multi-Agent Path Planning with PiBT

T0 review · 4 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read This paper shows that PiBT, extended to 8-connected grids and combined with safety-aware string pulling and a SIPP fallback, scales to over 500 agents in free-space multi-agent path planning while staying near-optimal in sparse domains.

desk verdict Useful engineering paper with a genuinely new safety-aware smoother, but the safety guarantee is overstated and the 500-agent claim needs runtime data. read the letter →

arxiv 2506.16748 v1 pith:GGZAMKHD submitted 2025-06-20 cs.RO cs.MA

Arjo Chakravarty , Michael X. Grey , M. A. Viraj J. Muthugala , Mohan Rajesh Elara This is my paper · ORCID

classification cs.ROcs.MA

keywords Multi-AgentPathFindingFree-spaceplanningsmoothingStringpullingPriorityInheritancewithBacktrackingSafeIntervalAny-angleScalability

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 paper asks whether a rule-based multi-agent planner can be made practical for free-space environments with hundreds of robots. The authors answer by extending Priority Inheritance with Backtracking (PiBT) to 8-connected grids, then applying a safety-aware string-pulling smoother that shortens paths while avoiding areas where agents are about to interact; a Safe Interval Path Planning (SIPP) fallback cleans up any new intersections the smoothing introduces. If correct, the framework shows that a bounded-runtime, suboptimal planner can handle scales of 500 or more agents at which optimal any-angle planners time out, while staying within about 0.8 percent of optimal path length on the small cases where optimality is computable.

What carries the argument

The load-bearing mechanism is safety-aware string pulling built on PiBT's locality: agents within two hops are directly interacting, and groups of interacting agents are defined transitively. The smoother never pulls strings while agents are directly interacting; it treats the region within one hop of any agent as a critical interaction area and forbids smoothing into that region by treating it as a time-bound obstacle in the line-of-sight check. Residual intersections are then handed to SIPP with a priority scheme. The 8-connected grid extension supplies the shorter initial paths and the diagonal-lane capacity that make the subsequent string pulling safe.

What would settle it

Run the pipeline on an adversarial or larger MAPF instance and check whether any pair of smoothed trajectories intersects at the same time and place while SIPP returns no collision-free resolution; a single such instance would refute the safety claim as stated.

Watch

Extended reading notes

Core claim

The paper's central discovery is that the local interaction structure of PiBT can be exploited for safe post-hoc smoothing: PiBT negotiations are local, so the smoother only operates on trajectory segments where agents are not directly interacting, and treats the one-hop critical interaction regions around other agents as time-bound obstacles during line-of-sight checks. This selective smoothing introduces far fewer new trajectory intersections than naive string pulling, an order of magnitude fewer in their benchmarks, and the residual intersections are resolved by a priority-based SIPP fallback. The result is a pipeline that extends PiBT to 8-connected grids and delivers collision-free free-space plans for over 500 agents in large open maps, with path lengths near those of optimal continuous-space planning in sparse domains and runtimes that scale far beyond the comparison planner.

Load-bearing premise

The load-bearing premise is that the SIPP fallback always succeeds in resolving every newly created trajectory intersection; if some instance produces an unresolvable intersection, the pipeline's claimed safety guarantee collapses.

Editorial extensions

If this is right

In open maps the pipeline produces collision-free paths for hundreds to over 500 agents in seconds, whereas the optimal any-angle baseline times out at a handful of agents.
Path smoothing cuts route length by up to 27.8 percent in free-space maps, with an average reduction of 7.3 percent across all tested scenarios and a modest runtime overhead of about 19 percent at 100 or more agents, dropping to as low as 2 percent at small scales.
Critical-region-aware smoothing introduces far fewer post-smoothing intersections than naive string pulling, keeping the SIPP fallback cheap to run.
Makespan is not reduced, and in dense or grid-like maps smoothing can yield near-zero improvement; the benefit concentrates in large open areas.

Reading between the lines

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

Our inference: the same local-planner plus safety-aware-smoother plus local-fallback recipe could be applied to other rule-based or priority-based MAPF planners, not just PiBT, whenever their interaction structure is local.
Our inference: a theoretical bound on the number of intersections introduced by the smoother, or on when SIPP is guaranteed to resolve them, would turn the empirical safety result into a certificate; the paper leaves this open.
Our inference: the effectiveness pattern suggests a practical deployment heuristic of applying this smoother only in low-density regions of a map and falling back to grid-based dense planning in corridors.

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.

Desk Editor's Note

read the letter

Short version: this is a useful engineering paper, not a breakthrough. The safety-aware string pulling with critical interaction regions is a genuinely new post-processing step, and the ablation showing it cuts new trajectory intersections by an order of magnitude over vanilla string pulling is the strongest part. The 8-connected PiBT extension is minor, but the combination with the smoother and SIPP fallback is a working pipeline, with code and benchmark data public.

The main flaw is the safety claim. Section 3 starts by saying the smoother 'guarantees collision-free paths', but Section 3.2 explicitly says it doesn't, and the last line of defense is a SIPP fallback that is only benchmarked, not proven. More specifically, Algorithm 1 resolves intersections pairwise and never rechecks the whole trajectory set after a fixpoint, so a SIPP modification for one pair could, in principle, put that agent in conflict with a third one. The stress-test note is right about this. It's not a fatal flaw for an empirical robotics paper, but the abstract and Section 3 title overstate the guarantee.

Second issue: the 500-agent claim in the abstract. The runtime tables stop at 100 agents; Table 4 shows only intersection counts for 500 agents, not solve times. That claim needs either a full runtime table or a softened abstract.

Minor: no error bars, one baseline (CCBS+SIPP from Open-RMF), and the path-length comparison only covers small instances because the baseline doesn't scale. The speed assumption (robots can move fast enough to cover a diagonal in one timestep) is stated but not justified in detail, and kinodynamics are admittedly out of scope.

Given the code and data, the paper deserves a serious referee. The central idea still holds up as a practical advance for free-space MAPF at scale, and the authors are honest in the conclusion about not claiming optimality or completeness. I'd send it to review with comments asking for a fixpoint check (or at least a clear limitation statement), a 500-agent runtime table, and a corrected safety statement in Section 3 and the abstract.

Referee Report

4 major / 5 minor

Summary. The paper proposes a post-processing pipeline for large-scale free-space multi-agent path planning. It extends the rule-based PiBT planner from 4-connected to 8-connected grids, applies a safety-aware string-pulling smoother that avoids locally interacting agents and critical interaction regions, and resolves any newly introduced trajectory intersections with a SIPP-based fallback. The authors report that the pipeline scales to hundreds of agents, reduces path lengths relative to vanilla PiBT, and produces paths within about 0.8% of optimal arclength on the small instances where an optimal baseline (AA-CCBS+SIPP) is available. Evaluation uses standard MAPF benchmarks (Stern et al.) and the authors provide code and raw results.

Significance. If the safety and scaling claims hold, this would be a practically valuable real-time MAPF approach for free-space domains, complementing optimal but poorly scaling methods. The paper includes commendable practices: open-source code, external benchmark evaluation, an ablation study on the critical-region mechanism, and honest discussion of limitations. However, the central safety guarantee is not established, and the headline 500-agent scaling claim is not backed by runtime data. The significance is therefore conditional on revision of these load-bearing points.

major comments (4)

[Section 3 (opening) and Section 3.2, Algorithm 1] The paper's central safety claim is internally inconsistent. Section 3 opens with 'We propose a path-smoother that guarantees collision-free paths,' but Section 3.2 states 'we do not have a guaranteed collision free path' and instead relies on a SIPP fallback that is only observed empirically. Moreover, Algorithm 1's pairwise resolution loop (lines 37-43) calls ApplySIPP on each detected pair intersection but never rechecks whether the modified trajectory introduces new conflicts with agents outside the pair, nor does it iterate to a fixpoint. Consequently, the output labelled 'Collision free smoothed path' is not supported by any invariant. Please either provide a proof (or formal conditions) that the pairwise SIPP resolution terminates with a globally collision-free set, or substantially weaken the claims in the abstract and Section 3 to describe an empirically safe pipeline rather than a guaranteed one.
[Abstract and Table 4] The abstract claims the method 'can scale to over 500 agents in large free-space environments,' but the only 500-agent data appear in Table 4, which reports intersection counts only—not solve times, success rates, or path lengths. No runtime or scaling evidence for 500 agents is presented anywhere in the paper. Please add 500-agent runtime benchmarks (ideally on multiple maps) or explicitly restrict the scaling claim to the 100-agent experiments in Tables 1 and 3.
[Tables 1-3] All runtime and arclength tables report single measurements with no error bars, confidence intervals, or multiple trials. Given that the paper's central claim is that PiBT+PS outperforms CCBS+SIPP in runtime, the lack of variance information makes it impossible to assess whether the reported advantages are significant or stable across instances. Please provide multiple runs (e.g., different random seeds or scenario instances) with standard deviations, or clearly state that the tables show single representative runs.
[Section 4 and Table 2] The claim that PiBT+PS produces paths '0.8% worse than the CCBS+SIPP optimal approach across the entire mapf benchmark' is not supported by the data shown. Table 2 covers only two maps, and for the larger agent counts (10, 15, 100) the CCBS+SIPP column is 'Unsolved' for Berlin or absent for Warehouse. The 0.8% figure also appears to be an aggregate across all scenarios, but the comparison is only possible where CCBS+SIPP terminates, which is heavily biased toward small, sparse instances. Please either restrict the near-optimality claim to the regime where the baseline is solvable, or provide a per-map breakdown with the number of comparable instances.

minor comments (5)

[Throughout] The planner name is spelled inconsistently as both 'PiBT' and 'PIBT'; please standardize (the original paper uses PIBT).
[Algorithm 1 header] The caption 'Parameters: PiBT Path Smoothing' is confusing; the procedure name is String-Pull and the algorithm implements PIBT-Smoother. Please clarify the caption.
[Section 5 vs Table 4] The text says 'the number of new intersections stays well below 20 in most cases,' but Table 4 reports 21 intersections for Berlin 1 256 with 500 agents. Please reconcile this statement or qualify 'most cases.'
[Section 3.2] The definition of the critical interaction region as 'within 1 hop' is a heuristic design choice; please state explicitly that the 1-hop radius is a fixed parameter of the algorithm and discuss its sensitivity to agent speed or grid resolution.
[Bibliography] Reference [2] is cited for AA-CCBS, but the text in Section 1 uses the name 'Any-Angle Continuous-time Conflict Based Search (AA-CCBS)'; the related-work section spells the author as 'Adreychuk' while the reference list uses 'Andreychuk'. Please ensure names match the reference list.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the pipeline's empirical claims are tested against independent external benchmarks, and the smoother's design parameters are fixed choices rather than fitted outputs.

full rationale

The paper's central derivation chain is self-contained with respect to its own predictions. PiBT's scalability is an external result [1]; the paper's contribution is an 8-connected extension plus a smoothing post-processor. The smoother's parameters — two-hop direct interaction, one-hop critical interaction, line-of-sight string pulling, and SIPP fallback — are stated design decisions in Section 3.2, and they are not fitted to the targets the paper claims (runtime, path-length reduction, number of intersections). Evaluation uses the independent Stern et al. MAPF benchmark and an external baseline (CCBS+SIPP from Open-RMF), so the claimed runtime and cost comparisons are not forced by construction. The manuscript contains an internal tension between the Section 3 heading ('guarantees collision-free paths') and the Section 3.2 concession ('Even after these restrictions, we do not have a guaranteed collision free path'), and the SIPP fallback's success is only empirically observed rather than proven. However, this is a correctness and completeness concern, not circularity, because the fallback is an independent solver invoked on externally generated intersections rather than a quantity derived from the claim being tested. No self-citations are load-bearing: the cited PiBT work is by different authors and is used as a building block, not as evidence for this paper's predictions. Therefore no step reduces by construction to its inputs, and the circularity score is 0.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

All parameters are algorithmic thresholds (interaction radius, critical interaction radius) chosen from the PiBT literature or the smoother design; none are fitted to the benchmark data. The paper introduces no physical entities or fitted constants. The main reliance is on PiBT's completeness, the variable-speed assumption, and a heuristic critical-interaction radius, all explicit in the text.

free parameters (2)

interaction_radius_hops = 2 hops
Agents within two hops are considered directly interacting (borrowed from PiBT [1]); this governs when smoothing is disabled.
critical_interaction_radius_hops = 1 hop
Regions within one hop of another agent are treated as time-bound obstacles during string pulling (Section 3.2, Figure 3). Chosen by hand, not fitted to data, but affects collision rate and smoothing aggressiveness.

assumptions (4)

domain assumption PiBT is complete for MAPD on bi-connected graphs (from [1])
The paper uses PiBT as its base configuration generator and inherits its behavior without re-proving completeness; Section 2.1 and Section 3.1 rely on PiBT properties.
domain assumption Robots can move at any speed up to the diagonal traversal time of a cell
Section 3.1 assumes this to allow diagonal 8-connected moves and to justify later string-pulled segments. If robot max speed is lower, the smoothed paths may be kinematically infeasible.
ad hoc to paper A 1-hop critical interaction region captures all imminent collision risks during smoothing
Section 3.2 and Figure 3 define critical interaction areas as within 1 hop; this is a heuristic design choice, not derived from a safety theorem.
standard math Any line segment inside a square is no longer than the square's diagonal
Used in Section 3.1 to justify the max-speed assumption; a standard geometric fact.

how reviews work

0 comments

Cite this review

Pith. "Pith review of A Scalable Post-Processing Pipeline for Large-Scale Free-Space Multi-Agent Path Planning with PiBT." pith.science (2026). https://pith.science/paper/GGZAMKHD

@misc{pith2026250616748,
  author       = {Pith},
  title        = {Pith review of: A Scalable Post-Processing Pipeline for Large-Scale Free-Space Multi-Agent Path Planning with PiBT},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GGZAMKHD}},
  note         = {Machine review of arXiv:2506.16748}
}

read the original abstract

Free-space multi-agent path planning remains challenging at large scales. Most existing methods either offer optimality guarantees but do not scale beyond a few dozen agents, or rely on grid-world assumptions that do not generalize well to continuous space. In this work, we propose a hybrid, rule-based planning framework that combines Priority Inheritance with Backtracking (PiBT) with a novel safety-aware path smoothing method. Our approach extends PiBT to 8-connected grids and selectively applies string-pulling based smoothing while preserving collision safety through local interaction awareness and a fallback collision resolution step based on Safe Interval Path Planning (SIPP). This design allows us to reduce overall path lengths while maintaining real-time performance. We demonstrate that our method can scale to over 500 agents in large free-space environments, outperforming existing any-angle and optimal methods in terms of runtime, while producing near-optimal trajectories in sparse domains. Our results suggest this framework is a promising building block for scalable, real-time multi-agent navigation in robotics systems operating beyond grid constraints.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

26 extracted references · 21 canonical work pages

[1]

Artificial Intelligence, 103752 (2022) https://doi.org/10.1016/j.artint.2022.103752

Okumura, K., Machida, M., D´ efago, X., Tamura, Y.: Priority inheritance with backtracking for iterative multi-agent path finding. Artificial Intelligence, 103752 (2022) https://doi.org/10.1016/j.artint.2022.103752

arXiv 2022
[2]

In: 2024 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pp

Yakovlev, K., Andreychuk, A., Stern, R.: Optimal and bounded suboptimal any- angle multi-agent pathfinding. In: 2024 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pp. 7996–8001 (2024). https://doi.org/ 10.1109/IROS58592.2024.10801691

arXiv 2024
[3]

In: 2020 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) (2020)

Macenski, S., Martin, F., White, R., Gin´ es Clavero, J.: The marathon 2: A navi- gation system. In: 2020 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) (2020)

work page 2020
[4]

Artificial Intelligence219, 40–66 (2015) https: //doi.org/10.1016/j.artint.2014.11.006 14

Sharon, G., Stern, R., Felner, A., Sturtevant, N.R.: Conflict-based search for optimal multi-agent pathfinding. Artificial Intelligence219, 40–66 (2015) https: //doi.org/10.1016/j.artint.2014.11.006 14

doi:10.1016/j.artint.2014.11.006 2015
[5]

In: Proceedings of the AAAI Conference on Artificial Intelligence, vol

Li, J., Ruml, W., Koenig, S.: Eecbs: A bounded-suboptimal search for multi-agent path finding. In: Proceedings of the AAAI Conference on Artificial Intelligence, vol. 35, pp. 12353–12362 (2021)

work page 2021
[6]

arXiv preprint arXiv:2410.03072 (2024)

Shaoul, Y., Mishani, I., Vats, S., Li, J., Likhachev, M.: Multi-robot motion planning with diffusion models. arXiv preprint arXiv:2410.03072 (2024)

arXiv 2024
[7]

In: Proceedings of the Fourteenth International Symposium on Combinatorial Search, SOCS 2021, Online, July 16-18, 2021, pp

Gordon, O., Filmus, Y., Salzman, O.: Revisiting the complexity analy- sis of conflict-based search: New computational techniques and improved bounds. In: Proceedings of the Fourteenth International Symposium on Combinatorial Search, SOCS 2021, Online, July 16-18, 2021, pp. 64–

work page 2021
[8]

In: Proceedings of AAAI Conference on Artificial Intelligence (AAAI) (2023)

Okumura, K.: Lacam: Search-based algorithm for quick multi-agent pathfinding. In: Proceedings of AAAI Conference on Artificial Intelligence (AAAI) (2023)

work page 2023

Show all 26 references

[9]

In: IJCAI, pp

Surynek, P.: Problem compilation for multi-agent path finding: a survey. In: IJCAI, pp. 5615–5622 (2022)

2022
[10]

In: Proceedings of the International Sym- posium on Combinatorial Search, vol

Surynek, P., Felner, A., Stern, R., Boyarski, E.: Sub-optimal sat-based approach to multi-agent path-finding problem. In: Proceedings of the International Sym- posium on Combinatorial Search, vol. 9, pp. 99–105 (2018)

2018
[11]

IEEE Robotics and Automation Letters4(3), 2378–2385 (2019)

Sartoretti, G., Kerr, J., Shi, Y., Wagner, G., Kumar, T.S., Koenig, S., Choset, H.: Primal: Pathfinding via reinforcement and imitation multi-agent learning. IEEE Robotics and Automation Letters4(3), 2378–2385 (2019)

2019
[12]

In: 2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pp

Wang, Y., Xiang, B., Huang, S., Sartoretti, G.: Scrimp: Scalable communication for reinforcement-and imitation-learning-based multi-agent pathfinding. In: 2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pp. 9301–9308 (2023). IEEE

2023
[13]

In: Proceedings of the AAAI Conference on Artificial Intelligence, vol

Andreychuk, A., Yakovlev, K., Panov, A., Skrynnik, A.: Mapf-gpt: Imitation learning for multi-agent pathfinding at scale. In: Proceedings of the AAAI Conference on Artificial Intelligence, vol. 39, pp. 23126–23134 (2025)

2025
[14]

In: ICAPS 2024 System’s Demonstration Track (2024)

Chan, S.-H., Chen, Z., Guo, T., Zhang, H., Zhang, Y., Harabor, D., Koenig, S., Wu, C., Yu, J.: The league of robot runners competition: Goals, designs, and implementation. In: ICAPS 2024 System’s Demonstration Track (2024)

2024
[15]

Symposium on Combinatorial Search (SoCS), 151–158 (2019)

Stern, R., Sturtevant, N.R., Felner, A., Koenig, S., Ma, H., Walker, T.T., Li, J., Atzmon, D., Cohen, L., Kumar, T.K.S., Boyarski, E., Bartak, R.: Multi-agent pathfinding: Definitions, variants, and benchmarks. Symposium on Combinatorial Search (SoCS), 151–158 (2019)

2019
[16]

https://github.com/open-rmf/mapf

GitHub - open-rmf/mapf: Multi-agent (path finding) planning framework — 15 github.com. https://github.com/open-rmf/mapf. [Accessed 23-04-2025]

2025
[17]

Proceedings of the AAAI Conference on Artificial Intelligence36(9), 10256–10265 (2022) https:// doi.org/10.1609/aaai.v36i9.21266

Li, J., Chen, Z., Harabor, D., Stuckey, P.J., Koenig, S.: Mapf-lns2: Fast repairing for multi-agent path finding via large neighborhood search. Proceedings of the AAAI Conference on Artificial Intelligence36(9), 10256–10265 (2022) https:// doi.org/10.1609/aaai.v36i9.21266

2022 doi
[18]

In: AAAI, vol

Nash, A., Daniel, K., Koenig, S., Felner, A.: Thetaˆ*: Any-angle path planning on grids. In: AAAI, vol. 7, pp. 1177–1183 (2007)

2007
[19]

In: Proceedings of the AAAI Conference on Artificial Intelligence, vol

Nash, A., Koenig, S., Tovey, C.: Lazy theta*: Any-angle path planning and path length analysis in 3d. In: Proceedings of the AAAI Conference on Artificial Intelligence, vol. 24, pp. 147–154 (2010)

2010
[20]

PhD thesis, University of Southern California (2012)

Nash, A.: Any-angle path planning. PhD thesis, University of Southern California (2012)

2012
[21]

International Journal of Artificial Intelligence & Applications2(4), 1 (2011)

Cui, X., Shi, H.: Direction oriented pathfinding in video games. International Journal of Artificial Intelligence & Applications2(4), 1 (2011)

2011
[22]

In: AAAI, pp

Surynek, P.: An optimization variant of multi-robot path planning is intractable. In: AAAI, pp. 1261–1263 (2010). https://doi.org/10.1609/aaai.v24i1.7767 . https://doi.org/10.1609/aaai.v24i1.7767

2010 doi
[23]

In: 2011 IEEE International Conference on Robotics and Automation, pp

Phillips, M., Likhachev, M.: Sipp: Safe interval path planning for dynamic envi- ronments. In: 2011 IEEE International Conference on Robotics and Automation, pp. 5628–5635 (2011). IEEE

2011
[24]

In: Distributed Autonomous Robotic Systems: The 10th International Symposium, pp

Alonso-Mora, J., Breitenmoser, A., Rufli, M., Beardsley, P., Siegwart, R.: Optimal reciprocal collision avoidance for multiple non-holonomic robots. In: Distributed Autonomous Robotic Systems: The 10th International Symposium, pp. 203–216 (2013). Springer

2013
[25]

In: 2023 International Symposium on Multi-Robot and Multi-Agent Systems (MRS), pp

Gao, Z., Yang, G., Prorok, A.: Online control barrier functions for decentralized multi-agent navigation. In: 2023 International Symposium on Multi-Robot and Multi-Agent Systems (MRS), pp. 107–113 (2023). IEEE 16

2023
[72]

https://doi.org/10.1609/socs.v12i1.18552

AAAI Press, ??? (2021). https://doi.org/10.1609/socs.v12i1.18552 . https://ojs.aaai.org/index.php/SOCS/article/view/18552

2021 doi

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.

[1] [1]

Artificial Intelligence, 103752 (2022) https://doi.org/10.1016/j.artint.2022.103752

Okumura, K., Machida, M., D´ efago, X., Tamura, Y.: Priority inheritance with backtracking for iterative multi-agent path finding. Artificial Intelligence, 103752 (2022) https://doi.org/10.1016/j.artint.2022.103752

arXiv 2022

[2] [2]

In: 2024 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pp

Yakovlev, K., Andreychuk, A., Stern, R.: Optimal and bounded suboptimal any- angle multi-agent pathfinding. In: 2024 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pp. 7996–8001 (2024). https://doi.org/ 10.1109/IROS58592.2024.10801691

arXiv 2024

[3] [3]

In: 2020 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) (2020)

Macenski, S., Martin, F., White, R., Gin´ es Clavero, J.: The marathon 2: A navi- gation system. In: 2020 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) (2020)

work page 2020

[4] [4]

Artificial Intelligence219, 40–66 (2015) https: //doi.org/10.1016/j.artint.2014.11.006 14

Sharon, G., Stern, R., Felner, A., Sturtevant, N.R.: Conflict-based search for optimal multi-agent pathfinding. Artificial Intelligence219, 40–66 (2015) https: //doi.org/10.1016/j.artint.2014.11.006 14

doi:10.1016/j.artint.2014.11.006 2015

[5] [5]

In: Proceedings of the AAAI Conference on Artificial Intelligence, vol

Li, J., Ruml, W., Koenig, S.: Eecbs: A bounded-suboptimal search for multi-agent path finding. In: Proceedings of the AAAI Conference on Artificial Intelligence, vol. 35, pp. 12353–12362 (2021)

work page 2021

[6] [6]

arXiv preprint arXiv:2410.03072 (2024)

Shaoul, Y., Mishani, I., Vats, S., Li, J., Likhachev, M.: Multi-robot motion planning with diffusion models. arXiv preprint arXiv:2410.03072 (2024)

arXiv 2024

[7] [7]

In: Proceedings of the Fourteenth International Symposium on Combinatorial Search, SOCS 2021, Online, July 16-18, 2021, pp

Gordon, O., Filmus, Y., Salzman, O.: Revisiting the complexity analy- sis of conflict-based search: New computational techniques and improved bounds. In: Proceedings of the Fourteenth International Symposium on Combinatorial Search, SOCS 2021, Online, July 16-18, 2021, pp. 64–

work page 2021

[8] [8]

In: Proceedings of AAAI Conference on Artificial Intelligence (AAAI) (2023)

Okumura, K.: Lacam: Search-based algorithm for quick multi-agent pathfinding. In: Proceedings of AAAI Conference on Artificial Intelligence (AAAI) (2023)

work page 2023