Pith. sign in

REVIEW 3 major objections 6 minor 1 cited by

Learning to Solve the Min-Max Mixed-Shelves Picker-Routing Problem via Hierarchical and Parallel Decoding

T0 review · 3 major / 6 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 paper claims that a multi-agent hierarchical attention model with parallel decoding and sequential action selection is the first neural solver to handle the min-max Mixed-Shelves Picker Routing Problem effectively, matching exact…

desk verdict First neural solver for the min-max MSPRP, but the headline comparison is suspect because MAHAM can split a picker's route into multiple tours while the MIP and neural baselines are restricted to one tour per picker. read the letter →

arxiv 2502.10233 v1 pith:UB3ALYH4 submitted 2025-02-14 cs.MA cs.LGstat.ML

Laurin Luttmann , Lin Xie This is my paper
classification cs.MAcs.LGstat.ML
keywords mixed-shelvespickerroutingproblemmin-maxmulti-agentreinforcementlearningneuralcombinatorialoptimizationparalleldecodinghierarchicaldecoderwarehouselogisticssequentialactionselection
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 proposes a learned solver, MAHAM, for the min-max Mixed-Shelves Picker Routing Problem, where several pickers in a mixed-shelves warehouse retrieve items and the goal is to minimize the longest tour. The authors claim that MAHAM coordinates pickers by generating a joint distribution over all agents' next actions in parallel, then resolving conflicts with a sequential action selection mechanism. They report that MAHAM matches Gurobi's optimal solutions on small instances and beats Gurobi on larger instances where the exact solver cannot find an optimum within an hour. If true, this would be the first neural method for the min-max variant and a practical option for real-time, large-scale warehouse routing.

What carries the argument

The load-bearing components are a heterogeneous-graph encoder with shared cross-attention parameters, an agent context encoder with ranking-based positional encoding and multi-head self-attention among pickers, and a hierarchical decoder that first selects shelf actions and then SKU actions. The sequential action selection samples one agent-action pair at a time from a single normalized distribution over the joint logit space, so the policy learns which picker should act first and avoids two agents selecting the same shelf-SKU combination. Training uses a self-improvement scheme that samples many candidate solutions from the current best policy, keeps the best one as a pseudo-optimal demonstration, and trains with cross-entropy loss.

What would settle it

Run MAHAM on instances where the number of pickers is fixed below ceil(sum(dp)/kappa); if the min-max objective degrades sharply or feasibility breaks, the claimed practical relevance to real warehouses with fixed picker counts would fail. A second check would be finding a large instance where Gurobi provably reaches an optimal solution within its time budget and beats MAHAM.

Watch

Extended reading notes

Core claim

The central claim is that MAHAM solves the min-max MSPRP by combining a hierarchical decoder with parallel decoding over all pickers, so that a single shared policy can balance workload between agents without the high latency of purely autoregressive construction. The joint logits over shelf and SKU choices are decoded through a learned sequential action selection, which masks infeasible actions and implicitly ranks agents by priority. The paper reports state-of-the-art objective values among neural baselines across all tested warehouse sizes, with gaps to the best known solutions below 1.7% on in-distribution instances, and shows large-scale generalization to 50-shelf instances with up to 500 SKUs that Gurobi cannot solve within an hour.

Load-bearing premise

The number of pickers is assumed to equal the minimum number of tours required by the demand and capacity, so the method only applies when staffing can be flexibly matched to the workload rather than being fixed in advance.

Editorial extensions

If this is right

  • MAHAM matches Gurobi's optimal solutions on MSPRP10 and beats Gurobi's one-hour results on MSPRP25 and MSPRP40, with gaps below 1% on most in-distribution settings.
  • Large-scale generalization tests show MAHAM finding solutions in seconds for 50-shelf, 500-SKU instances where Gurobi finds nothing in an hour, with objective gaps of 4% or less against the best known solutions.
  • Parameter sharing in the cross-attention layers cuts model size by about 20% and improves both inference speed and out-of-distribution generalization.
  • The sequential action selection avoids the conflicts that PARCO incurs, so MAHAM needs fewer decoding steps and trains faster than the alternative parallel method.
  • The first neural solver for the min-max variant of the MSPRP provides a concrete alternative to hand-crafted heuristics for mixed-shelves warehouse operations.

Reading between the lines

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

  • Editorial inference: the joint-logit sequential selection strategy could be ported to other multi-agent problems with composite action spaces and hard collision constraints, such as multi-robot task allocation, where parallel decoding currently relies on hand-built conflict handlers.
  • Editorial inference: because the number of pickers is set equal to the minimal number of tours implied by demand and capacity, the method implicitly assumes flexible staffing; adapting it to a fixed workforce would require treating the number of pickers as an input and likely a different reward design.
  • Editorial inference: the reported out-of-distribution robustness suggests that learning an agent ordering from logits is a transferable skill, and a testable prediction is that the learned ranking would generalize across different warehouse layouts even if the rest of the decoder is retrained.
  • Editorial inference: the self-improvement training loop selects the best of 100 sampled trajectories, so the reported margins over baselines may depend on the quality of the initial reference policy and the sampling budget.
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 / 6 minor

Summary. The paper proposes MAHAM, a multi-agent hierarchical attention model for the min-max Mixed-Shelves Picker Routing Problem (MSPRP). MAHAM extends the Hierarchical Attention Model (HAM) with a joint logit-space decoder that generates actions for all pickers in parallel, followed by a sequential action-selection scheme to guarantee feasibility. The model is trained with a self-improvement cross-entropy method. Experiments compare MAHAM against Gurobi, a greedy heuristic, HAM, Equity Transformer, 2d-Ptr, and PARCO on MSPRP instances with 10, 25, 40, and 50 shelves, reporting better solution quality and speed, especially on large out-of-distribution instances.

Significance. If the results were established, the paper would make a useful empirical contribution to neural combinatorial optimization for min-max multi-agent routing: the proposed joint logits plus sequential masking is a plausible way to obtain coordination without fully autoregressive decoding, and the ablation of ranking strategies is informative. The public code and the detailed description of the architecture are strengths. However, the central empirical claim is currently compromised by a mismatch between the problem solved by MAHAM and the problem solved by the MIP and the baseline methods, and by the lack of statistical reporting. With the problem definition aligned and repeated experimental runs, the contribution could be significant.

major comments (3)
  1. [Section 3.1, Appendix A, Appendix B] The MDP in Section 3.1 permits a picker to visit the packing station, unload, restore capacity, and continue, so the per-picker distance used in the reward can span multiple station-to-station tours. In contrast, the MIP in Appendix A models each element of B as exactly one independent tour (constraints (15)-(18), (20)) and minimizes the maximum tour length, with no aggregation of tours per picker. Appendix B states that HAM is adapted "thanks to our assumption, that there are exactly as many pickers as there are tours." Consequently, Tables 1 and 2 compare MAHAM on a less constrained problem than the MIP and the one-tour-per-picker baselines, and the reported margins may reflect this extra degree of freedom rather than the proposed hierarchical and parallel decoding. Please align the problem definitions: either make the packing station a terminal action that ends an agent's episode (one tour per picker), or reformulate the MIP and the baselines to allow multiple tours per picker and compare against the same min-max total-distance objective.
  2. [Section 5.1, Table 1] No repeated runs or error bars are reported. The Gurobi rows are described as solving a single test instance per configuration, and the neural baseline entries report only the best objective among 1280 sampled solutions, without standard deviations over instances or training seeds. Given that several margins in Table 1 are below 0.5% (e.g., MAHAM 0.01% vs PARCO 0.13% for MSPRP10 with |P|=3), the paper should report means and standard deviations over multiple test instances and multiple seeds, or otherwise the claim of consistent superiority is not statistically supported.
  3. [Table 2, Section 5.2] The large-scale generalization claims rest on a single comparison protocol in which each neural baseline is evaluated with 1280 sampled solutions. It is not stated how many test instances are used, whether the same instances are used for all methods, or whether the reported times are for the full sampling procedure or for a single greedy rollout. Please specify the evaluation protocol, including the number of test instances and the precise timing convention, so that the speed comparisons in Table 2 are reproducible.
minor comments (6)
  1. [Table 1] Decimal commas appear inconsistently: entries such as "0,30s" and "0,25s" use a comma, while other entries use a period; please use a single decimal separator throughout.
  2. [Table 1] The header for the second block reads "SKUs" while the first block uses "|P|"; please use a consistent label for the number of SKUs.
  3. [Algorithm 1, Section 4.3] The normalization expression for P_ma is typeset incorrectly (the fraction bar is missing); it should read P_ma = exp(L'_ma / beta) / suma over all agent-action pairs of exp(L'_ij / beta).
  4. [Section 4.1, Equation (10)] The scale parameter C is introduced but its value or training scheme is never specified; please state how C is set in the experiments.
  5. [References] References [15] and [16] are duplicate entries for the same Matrix Encoding Networks paper; please merge them.
  6. [Section 5.3, Figure 3] There are minor formatting issues: "Figure3a" and "Figure3b" lack spaces, and "table 2" is lowercase; please fix these for journal style.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; the central SOTA claim is an external empirical comparison, though a tour-count mismatch between MAHAM and baselines is a correctness risk rather than a circularity.

full rationale

The paper's central claim is an empirical performance comparison of a learned policy against Gurobi and independently published neural baselines, evaluated on a fixed instance-generation scheme. The method builds on the authors' own prior work (HAM and PARCO), but no theorem or prediction is obtained by defining the answer into the inputs. The self-improvement training scheme uses the model's own sampled best solutions as training targets; this is a bootstrapped training procedure, not a derivation that presupposes the conclusion. The most serious concern is a problem-formulation mismatch: the MDP allows a picker to visit the packing station, restore capacity, and continue, so an MAHAM trajectory may contain more than M = ceil(sum_p d_p / kappa) tours, whereas the Appendix A MIP models exactly one independent station-to-station tour per element of B and the HAM baseline is adapted under the assumption that there are exactly as many pickers as tours. This is a potential confounding of the experimental comparison and a correctness/validity risk, but it is not circular: MAHAM's improvements are not equivalent to its inputs by construction, and the result still requires training and external evaluation. Self-citations to [18] and [2] are building-block references, not load-bearing justifications of the empirical claim. Therefore no significant circularity is present.

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

The model itself introduces no new physical or mathematical entities; the dummy SKU used as a default action in Algorithm 1 is a bookkeeping device, not an independent entity. The core assumptions are the M = number-of-tours modeling choice and the split-order framework inherited from prior work.

free parameters (4)
  • Scale parameter C (Eq. 10) = not reported
    Scales logits in the decoder cross-attention; typical value in similar AM-based models is 10 but not specified here.
  • Sampling temperature beta (Algorithm 1) = not reported
    Temperature for categorical sampling during sequential action selection; affects exploration and solution diversity.
  • Number of candidate solutions alpha in self-improvement = 100
    Number of samples drawn per instance to select the pseudo-optimal training target; larger alpha improves target quality at higher compute cost.
  • Encoder/decoder hyperparameters (emb dim 256, heads 8, layers 4) = 256/8/4
    Chosen architecture sizes (Appendix C.2); performance depends on these choices.
assumptions (4)
  • domain assumption Number of agents equals number of required tours: M = ceil(sum_p d_p / kappa)
    Section 3 states this to compare against baselines; ties picker count to demand and capacity, not a fixed workforce.
  • domain assumption Split orders and split deliveries are allowed
    Section 3 and prior work [18,27]; relaxes the requirement that an SKU's demand be satisfied in one tour or by one picker.
  • domain assumption Pick quantities are determined by the transition's deterministic heuristic (Eq. 1) with a permutation Omega
    Section 3.1; this heuristic defines the reward and the learning signal, and is not learned.
  • domain assumption Self-improvement with best-of-alpha pseudo-optimal targets is a valid learning signal
    Section 4.4 adopts the method of [19]; the convergence and quality of the learned policy depend on this assumption.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Learning to Solve the Min-Max Mixed-Shelves Picker-Routing Problem via Hierarchical and Parallel Decoding." pith.science (2026). https://pith.science/paper/UB3ALYH4

@misc{pith2026250210233,
  author       = {Pith},
  title        = {Pith review of: Learning to Solve the Min-Max Mixed-Shelves Picker-Routing Problem via Hierarchical and Parallel Decoding},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/UB3ALYH4}},
  note         = {Machine review of arXiv:2502.10233}
}
read the original abstract

The Mixed-Shelves Picker Routing Problem (MSPRP) is a fundamental challenge in warehouse logistics, where pickers must navigate a mixed-shelves environment to retrieve SKUs efficiently. Traditional heuristics and optimization-based approaches struggle with scalability, while recent machine learning methods often rely on sequential decision-making, leading to high solution latency and suboptimal agent coordination. In this work, we propose a novel hierarchical and parallel decoding approach for solving the min-max variant of the MSPRP via multi-agent reinforcement learning. While our approach generates a joint distribution over agent actions, allowing for fast decoding and effective picker coordination, our method introduces a sequential action selection to avoid conflicts in the multi-dimensional action space. Experiments show state-of-the-art performance in both solution quality and inference speed, particularly for large-scale and out-of-distribution instances. Our code is publicly available at http://github.com/LTluttmann/marl4msprp.

Figures

Figures reproduced from arXiv: 2502.10233 by the authors.

Figure 1
Figure 1. Overview of the MAHAM Architecture Q and SKU embeddings as keys K. This contrasts with the MatNet [16] and HAM [18] architectures, which compute separate attention scores for each node type—once as queries and once as keys. Formally we perform: A = QK⊤ √ dk , Q = WQH l−1 V , K = W KH l−1 P (3) where WQ and W K ∈ R dk×D are weight matrices learned per attention head4 and dk is the per-head embedding dimension. The re… view at source ↗
Figure 2
Figure 2. Agent Context Encoder Agent Encoder. To ac￾count for multiple agents, we introduce an Agent Con￾text Encoder, as illustrated in [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. a summarizes the results of an ablation study testing the effectiveness of the proposed components in our MAHAM architecture. The full model with learned rankings and rank-dependent positional encodings (PE) performs signif￾icantly better than the models relying on an index-based or random order, and also achieves better solutions than MAHAM without the positional encoding. Encoder Parameter Sharing: MAHAM introduce… view at source ↗

Discussion (0). Sign in to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Solving the Pod Repositioning Problem with Deep Reinforced Adaptive Large Neighborhood Search

    cs.RO 2025-06 conditional novelty 5.0 of 10

    A reinforcement-learning-controlled adaptive search method finds cheaper pod storage plans than standard heuristics in simulated robotic warehouse tests.

Reference graph

Works this paper leans on

29 extracted references · 21 canonical work pages · cited by 1 Pith paper

  1. [1]

    Ba, J.L., Kiros, J.R., Hinton, G.E.: Layer normalization (2016), https://arxiv.org/ abs/1607.06450

    arXiv 2016

  2. [2]

    arXiv preprint arXiv:2409.03811 (2024)

    Berto, F., Hua, C., Luttmann, L., Son, J., Park, J., Ahn, K., Kwon, C., Xie, L., Park, J.: Parco: Learning parallel autoregressive policies for efficient multi-agent combinatorial optimization. arXiv preprint arXiv:2409.03811 (2024)

    arXiv 2024

  3. [3]

    arXiv preprint arXiv:2306.17100 (2023)

    Berto, F., Hua, C., Park, J., Luttmann, L., Ma, Y., Bu, F., Wang, J., Ye, H., Kim, M., Choi, S., et al.: Rl4co: an extensive reinforcement learning for combinatorial optimization benchmark. arXiv preprint arXiv:2306.17100 (2023)

    arXiv 2023

  4. [4]

    European Journal of Operational Re- search 262(2), 550–562 (2017)

    Boysen, N., Briskorn, D., Emde, S.: Parts-to-picker based order processing in a rack-moving mobile robots environment. European Journal of Operational Re- search 262(2), 550–562 (2017). https://doi.org/10.1016/j.ejor.2017.03.053

    work page doi:10.1016/j.ejor.2017.03.053 2017

  5. [5]

    Self-Labeling the Job Shop Scheduling Problem

    Corsini, A., Porrello, A., Calderara, S., Dell’Amico, M.: Self-labeling the job shop scheduling problem. arXiv preprint arXiv:2401.11849 (2024)

    work page Pith review arXiv 2024

  6. [6]

    European Journal of Operational Research105(1), 1–17 (Feb 1998)

    Daniels, R.L., Rummel, J.L., Schantz, R.: A model for warehouse order picking. European Journal of Operational Research105(1), 1–17 (Feb 1998). https://doi. org/10.1016/S0377-2217(97)00043-X

    work page doi:10.1016/s0377-2217(97)00043-x 1998

  7. [7]

    https://doi.org/10.1016/j.ejor.2006.07.009

    De Koster, R., Le-Duc, T., Roodbergen, K.J.: Design and control of warehouse or- derpicking:Aliteraturereview.EuropeanJournalof OperationalResearch 182(2), 481–501 (2007). https://doi.org/10.1016/j.ejor.2006.07.009

    doi:10.1016/j.ejor.2006.07.009 2007

  8. [8]

    Gurobi Optimization, LLC: Gurobi Optimizer Reference Manual (2024), https: //www.gurobi.com

    work page 2024

Show all 29 references
  1. [9]

    arXiv preprint arXiv:1606.08415 (2016)

    Hendrycks, D., Gimpel, K.: Gaussian error linear units (gelus). arXiv preprint arXiv:1606.08415 (2016)

    2016 arXiv

  2. [10]

    In: Advances in Neural Information Processing Systems (2022)

    Kim, M., Park, J., Park, J.: Sym-nco: Leveraging symmetricity for neural com- binatorial optimization. In: Advances in Neural Information Processing Systems (2022)

    2022

  3. [11]

    net/forum?id=ByxBFsRqYm Hierarchical and Parallel Decoding for Picker-Routing 15

    Kool, W., van Hoof, H., Welling, M.: Attention, learn to solve routing problems! In: International Conference on Learning Representations (2019), https://openreview. net/forum?id=ByxBFsRqYm Hierarchical and Parallel Decoding for Picker-Routing 15

    2019

  4. [12]

    Kool, W., van Hoof, H., Welling, M.: Buy 4 REINFORCE Samples, Get a Baseline for Free! ICLR Workshops (Apr 2019)

    2019

  5. [13]

    Advances in Neural Information Processing Systems33, 21188–21198 (2020)

    Kwon, Y.D., Choo, J., Kim, B., Yoon, I., Gwon, Y., Min, S.: Pomo: Policy op- timization with multiple optima for reinforcement learning. Advances in Neural Information Processing Systems33, 21188–21198 (2020)

    2020

  6. [14]

    In: Larochelle, H., Ran- zato, M., Hadsell, R., Balcan, M., Lin, H

    Kwon, Y.D., Choo, J., Kim, B., Yoon, I., Gwon, Y., Min, S.: POMO: Policy opti- mization with multiple optima for reinforcement learning. In: Larochelle, H., Ran- zato, M., Hadsell, R., Balcan, M., Lin, H. (eds.) Advances in Neural Information Processing Systems. vol. 33, pp. 2...

    2020

  7. [15]

    In: Advances in Neural Informa- tion Processing Systems

    Kwon, Y.D., Choo, J., Yoon, I., Park, M., Park, D., Gwon, Y.: Matrix encoding networks for neural combinatorial optimization. In: Advances in Neural Informa- tion Processing Systems. vol. 34, pp. 5138–5149 (2021)

    2021

  8. [16]

    In: Advances in Neural Informa- tion Processing Systems

    Kwon, Y.D., Choo, J., Yoon, I., Park, M., Park, D., Gwon, Y.: Matrix encoding networks for neural combinatorial optimization. In: Advances in Neural Informa- tion Processing Systems. vol. 34, pp. 5138–5149. Curran Associates, Inc. (2021)

    2021

  9. [17]

    In: Proceedings of the 23rd International Conference on Autonomous Agents and Mul- tiagent Systems

    Liu, Q., Liu, C., Niu, S., Long, C., Zhang, J., Xu, M.: 2d-ptr: 2d array pointer network for solving the heterogeneous capacitated vehicle routing problem. In: Proceedings of the 23rd International Conference on Autonomous Agents and Mul- tiagent Systems. pp. 1238–1246 (2024)

    2024

  10. [18]

    In: Pro- ceedings of the International Conference on Automated Planning and Scheduling

    Luttmann, L., Xie, L.: Neural combinatorial optimization on heterogeneous graphs: An application to the picker routing problem in mixed-shelves warehouses. In: Pro- ceedings of the International Conference on Automated Planning and Scheduling. vol. 34, pp. 351–359 (2024)

    2024

  11. [19]

    Transactions on Machine Learning Research (2024)

    Pirnay, J., Grimm, D.G.: Self-improvement for neural combinatorial optimization: Sample without replacement, but improvement. Transactions on Machine Learning Research (2024)

    2024

  12. [20]

    In: Proceedings of the AAAI Conference on Artificial Intelligence

    Son, J., Kim, M., Choi, S., Kim, H., Park, J.: Equity-transformer: Solving np- hard min-max routing problems as sequential generation with equity context. In: Proceedings of the AAAI Conference on Artificial Intelligence. vol. 38, pp. 20265– 20273 (2024)

    2024

  13. [21]

    John Wiley & Sons,4th edn

    Tompkins, J.A.: Facilities planning. John Wiley & Sons,4th edn. (2010)

    2010

  14. [22]

    In: Advances in Neural Information Processing Systems

    Vaswani,A.,Shazeer,N.,Parmar,N.,Uszkoreit,J.,Jones,L.,Gomez,A.N.,Kaiser, Ł., Polosukhin, I.: Attention is All you Need. In: Advances in Neural Information Processing Systems. vol. 30. Curran Associates, Inc. (2017)

    2017

  15. [23]

    Advances in neural infor- mation processing systems28 (2015)

    Vinyals, O., Fortunato, M., Jaitly, N.: Pointer networks. Advances in neural infor- mation processing systems28 (2015)

    2015

  16. [24]

    Computers &OperationsResearch 95,139–150(Jul2018).https://doi.org/10.1016/j.cor.2018

    Weidinger, F.: Picker routing in rectangular mixed shelves warehouses. Computers &OperationsResearch 95,139–150(Jul2018).https://doi.org/10.1016/j.cor.2018. 03.012

    2018 doi

  17. [25]

    European Journal of Operational Research274(2), 501–515 (Apr 2019)

    Weidinger, F., Boysen, N., Schneider, M.: Picker routing in the mixed-shelves ware- houses of e-commerce retailers. European Journal of Operational Research274(2), 501–515 (Apr 2019). https://doi.org/10.1016/j.ejor.2018.10.021

    2019 doi

  18. [26]

    European Journal of Operational Research307(2), 713–730 (2023)

    Xie, L., Li, H., Luttmann, L.: Formulating and solving integrated order batch- ing and routing in multi-depot agv-assisted mixed-shelves warehouses. European Journal of Operational Research307(2), 713–730 (2023)

    2023

  19. [27]

    European Journal of Operational Research 288(1), 80–97 (2021)

    Xie, L., Thieme, N., Krenzler, R., Li, H.: Introducing split orders and optimizing operational policies in robotic mobile fulfillment systems. European Journal of Operational Research 288(1), 80–97 (2021)

    2021

  20. [28]

    In: Proceedings of the AAAI conference on artificial intelligence

    Zong, Z., Zheng, M., Li, Y., Jin, D.: Mapdp: Cooperative multi-agent reinforce- ment learning to solve pickup and delivery problems. In: Proceedings of the AAAI conference on artificial intelligence. vol. 36, pp. 9980–9988 (2022) 16 Luttmann, Xie A Formal Definition of the MSP...

    2022

  21. [29]

    addresses min-max routing problems by employing a sequential planning approach with sequence generators like the Transformer. It focuses on equitable workload distribution among multiple agents, applying this strategy to chal- lenges like the min-max multi-agent traveling sale...

Pith tools

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