Exploring Decentralized Warehouse Management Using Large Language Models: A Proof of Concept

Round 1

Reviewer 1 Report (Previous Reviewer 5)

Comments and Suggestions for Authors

A proof of concept was conducted to explore decentralized warehouse management using LLMs. The topic is interesting and worth investigating. Please consider enhancing the quality of the article by addressing the comments below.
Comment 1. Strengthen the research results in the abstract.
Comment 2. The Fourth Industrial Revolution was mentioned in the abstract; however, both Industry 4.0 and 5.0 were mentioned in the first sentence of the first section.
Comment 3. Section 1 Introduction:
(a) Enhance the literature review by providing a concise summary of the methodologies, results, and limitations of the existing works (mainly recent 5-year journal articles).
(b) Add a paragraph to discuss the research contributions of the paper.
Comment 4. Among different techniques, why were LLMs considered, provided that LLMs are not mature and training an LLM for a specific application is challenging?
Comment 5. Section 2 Theoretical background and literature review:
(a) As a theoretical background, LLMs are very difficult to explain in only short descriptions.
(b) The section is titled “literature review”, which is not appropriate based on the content being presented.
Comment 6. Section 3 Material and methods:
(a) Figure 1: Some wordings were cut. In addition, the content did not fully cover all elements of each layer and linkages between layers.
(b) The authors explained the scenarios of two package agents and warehouse agents in Figure 2. How about the generic case?
(c) Zoom in your file to confirm no content is blurred (Figures).
(d) Table 1: What is “Nr.”? In addition, only limited scenarios were tested. How can you confirm that the proposal can be applied to general cases in real-world environments?
Comment 7. Figures 7-15: Clarify agents 1 to 4 and AI. Perhaps adding a table would enhance the clarity.
Comment 8. A comparison with the existing works is expected.
Comment 9. As a proof of concept, what are the major limitations of your work?

Author Response

We would like to express our sincere gratitude for your insightful comments and valuable suggestions, which significantly contributed to improving the quality of our manuscript titled "Exploring Decentralized Warehouse Management Using Large Language Models: A Proof of Concept". We carefully addressed all points raised during the review process and made the necessary revisions to the manuscript. The suggestions were instrumental in refining our work and we are grateful for their time and expertise.

All answers are in the attached file.

Author Response File: Author Response.pdf

Reviewer 2 Report (Previous Reviewer 3)

Comments and Suggestions for Authors

I have appreciated the deep revision of the contents and the present form of this manuscript. All my previous concerns have been accurately addressed. I think that this paper can be accepted.

Author Response

Dear Reviewer,

We would like to sincerely thank you for your thoughtful and encouraging feedback. We greatly appreciate that the extensive revisions we have made have been recognized, and we are grateful that our efforts to address all concerns have been well received. We are pleased to hear that the current version of the manuscript meets the reviewers’ expectations, and we thank them for their support in recommending acceptance of the paper.

Reviewer 3 Report (New Reviewer)

Comments and Suggestions for Authors

This manuscript presents an original and timely exploration of using Large Language Models (LLMs) as autonomous agents in decentralized warehouse management systems. The proposed multi-layered architecture and LLM-driven negotiation mechanisms represent a novel approach to shared manufacturing and logistics. The simulation-based validation demonstrates promising results, especially in capacity pooling scenarios.

However, despite its strengths, the paper requires significant revisions to improve scientific structure, clarity, methodology presentation, and interpretative depth. The following points detail the areas where the manuscript should be improved.

• The manuscript lacks explicitly stated research questions or hypotheses, which are essential for scientific clarity and alignment between the research goals and results.

• The introduction section does not conclude with an overview of the paper’s structure, making it harder for the reader to navigate the manuscript.

• There is no dedicated review of the state of the art (SoA) in decentralized warehouse management, LLM-based agent systems, or shared logistics—only a theoretical foundation is presented.

• The manuscript’s overall length is disproportionate for a proof-of-concept study, and Section 4 (Real-world application) is extremely short compared to other sections.

• The Methods section (Section 3) mixes system architecture with experimentation setup. These should be separated:

  • A dedicated section for “Proposed Solution” or “System Architecture”.

  • A separate “Validation” or “Experimental Setup” section for simulation design and metrics.

• The manuscript jumps from the Results directly to the Conclusion, omitting a Discussion section. A proper Discussion section should:

  • Analyze the simulation results.

  • Revisit the research objectives.

  • Compare findings to the SoA.

  • Acknowledge limitations and future work directions.

• The English language needs editing to improve clarity, conciseness, and readability.

• There is redundant use of full terms and acronyms throughout the paper (e.g., repeating “Large Language Models (LLMs)” instead of just using “LLMs” after the first mention).

• Some main sections jump directly into subsections (e.g., Section 3 → 3.1) without any introductory paragraph to contextualize the content.

• The manuscript includes figures in a row without narrative breaks, reducing readability and overwhelming the reader.

• Figure 3 is placed mid-sentence or mid-paragraph, breaking the flow of the text.

• Figure 6 spans an entire page, disrupting the main narrative. It would be better placed in an appendix or resized.

• Some equations are embedded within sentences instead of being formatted as separate display elements between paragraphs.

• The listing of prompts is not cited or contextualized in the text. It should be referenced and discussed.

• Numbered equations in Section 2.1 are not cited in the body text, making it difficult to follow the logical flow.

• The Results section includes little narrative analysis, and the findings are not tied back to goals, limiting their impact.

• The LLM prompting approach is not well detailed. Important aspects such as the format, parsing logic, settings, and handling variability are missing.

• The baseline for comparison is too simplistic—random agents. Including rule-based or optimization-based agents would offer a more rigorous benchmark.

• The simulation was only repeated five times per scenario, which is statistically weak. More repetitions or confidence intervals would improve reliability.

• No statistical tests or error measures are provided, making it hard to evaluate the significance of the observed performance improvements.

• Results are not contextualized with related work, so it is unclear whether the performance achieved is competitive with state-of-the-art techniques.

• The “Real-World Application” section is promising but underdeveloped. It lacks discussion of:

  • Cost of implementation.

  • Real-time performance concerns.

  • Deployment scalability.

  • Explainability and trust in LLM decisions.

• The paper does not acknowledge limitations of the proof-of-concept approach (e.g., reliance on stateless prompts, simulation scope, or lack of real-world prototype).

Author Response

Dear Reviewer,

We would like to express our sincere gratitude for your insightful comments and valuable suggestions, which significantly contributed to improving the quality of our manuscript titled "Exploring Decentralized Warehouse Management Using Large Language Models: A Proof of Concept". We carefully addressed all points raised during the review process and made the necessary revisions to the manuscript. The suggestions were instrumental in refining our work and we are grateful for their time and expertise.

All answers are in the attached file.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report (Previous Reviewer 5)

Comments and Suggestions for Authors

The authors have made significant enhancements to the quality of the article. I have some follow-up comments on the revised article.
Follow-up Comment 1. The results of the existing works were not clearly summarized and discussed in the first section.
Follow-up Comment 2. Clearly summarize the research contributions in the first section.
Follow-up Comment 3. Presenting artificial intelligence and LLMs in the same subsection (2.4) is not clear.
Follow-up Comment 4. Clarify the inputs and outputs in Figures 3 to 5.

Author Response

Dear Reviewer,

Thank you very much for the quality review and good comments that improved our manuscript. Responses to the comments are in the attached document.

Author Response File: Author Response.pdf

Reviewer 3 Report (New Reviewer)

Comments and Suggestions for Authors

• The way you put RQs and hypotheses doesn't work very well, since you have a 1:1 relationship, which results in repetitive ideas. Usually, in a typical scientific research work, we formulate one hypothesis, following by some (related) RQs that result from the hypothesis. The research methodology focus on finding answers to the RQs, which will led to confirm or decline the hypothesis. If you like, you can also drop the hypotheses and keep only the RQs.
• Still regarding the RQs, you need to have a clear discussion and conclusions about these at the end of the paper. Did you find anwsers to the RQs? What are your findings? What did you conclude?
• The SoA should follow the Theoretical background. You can't expect the reader to fully understand the SoA before reading the background and aquire the baseline knowledge.. 
• Also, the SoA included is poor and doesn't reflect a good literature revision. Are you saying that there isn't any previous work that used LLM for ODW?
• Still, regarding the SoA, this should be in a dedicated section or embedded in a large section together with the theoretical background. Keep the introduction to introduce the work, with a proper context, motivation, problem definition and research goals.
• Finally, at the end where you compare the proposed solution with SoA approaches, this can be considered a comparison.. You state the proposed solution is better, but where are the evidence? What metrics did you use to compare? Where did the proposed approach resulted when the existing SoA failed? Also, I didn't even see the Wang & Yue and Jamili et al. work mentioned in the SoA, why did you try to compare your approach with these traditional approaches, if you don't have means to do a proper comparison of results?
• Did you actually have a real-world application of the proposed solution? By the content in section 5, I guess not. What you have is a point of view about the potential of applying the proposed solution in a real-world use case. This is not suitable for a dedicated section, you are falsely leading the reader thinking you actually applyed and validated the proposed solution to a real use case, besides the simulation approach. So, this section could easily be reduced to a small part of the conclusion about potential applications of the proposed approach..
• I understand what you are saying regarding the limitation of (only) random agents comparison and statistic validation. But for a paper this size, where the SoA analysis and result comparison is weak, you need to explore other formats for a stronger validation of the proposed solution. Otherwhise, it is really difficult to support the significance of the work.
• Personal note: For the work you are trying to report, I bet you could fit everyting in a strong 8-page paper for a conference, with lots of room for follow up papers with extended content/results.

Author Response

Dear Reviewer,

Thank you very much for your review. Responses to the comments are in the attached document.

Author Response File: Author Response.pdf

This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The paper offered for review deals with proof of concept in implementing an artificial intelligence model. It makes for a very interesting read, with good overall structure.

There are however some aspects that might, in my opinion, benefit from some further work. These are, as follows:

1. The abstract of the paper should be solely concentrated on what has been done in the paper, with an emphasis on the novelty of the work.

2. Chapters 2 (Theoretical background) and 3 (Literature review) both contain references to work by other authors, it is hard to distinguish between the purpose of the two. I recommend blending the two and only keeping one title that might be  "State of the art" or another name that the authors deem appropriate.

3. Chapter 4.1 offers a very good description of the theoretical aspects of the paper. Chapter 4.2 also however does not do such a good job in setting up the general parameters of the simulation.
    I have a few questions here:
    - No information if offered here on how the simulation will actually be performed, other that is uses ChatGPT API calls. How is this performed? How was it implemented?
    - Since the "heavy lifting" (i assume, there is now information on how the simulation is implemented) is done by ChatGPT's online model, why is the computer's configuration relevant here?

4. Chapter 5, which, in my opinion should be the focal point of the paper, does very little to give a reasonable explanation as to how the results for the simulation were obtained. As such, it's impossible for me to evaluate them.

5. Furthermore, there is absolutely no mention here as to how this could be implemented in the real world. A real-world usage scenario would do a lot to further the paper.

6. I would blend chapters 6 and 7 and call it "Conclusions"

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

Dear Authors,

Thank you for the opportunity to review your work, “Exploring Decentralised Warehouse Management Using Large Language Models: A Proof of Concept.” Your article provides a thorough exploration of decentralized warehouse management using Large Language Models and offers a compelling proof of concept. The study is well-structured, addressing a highly relevant and timely topic in the fields of logistics and AI.

The title accurately reflects the study’s scope and focus.

While the abstract outlines the work you have done, it spends too much time on general topics, especially in the first section. It would be more effective if it elaborated further on your specific research and findings.

The article builds on a solid theoretical foundation, with well-referenced discussions on Multi-Agent Systems, Game Theory, and Decision Theory. However, some parts of this section are overly technical and not all are directly relevant to the main study. For example, the mathematical representations in the introduction to Multi-Agent Systems could be simplified or summarized to improve readability and focus.

The review section, aside from the technical background, is sufficiently detailed and provides valuable insights.

The structure of the article is logical and easy to follow, with well-defined sections that progress systematically. The figures are clear and informative, though Figure 7 has a typo: “emty” should be corrected to “empty.”

The model is well-described and sound. However, the case studies in Chapters 4.2 and 5 are too brief, making it difficult to fully understand the methodology and simulation without rereading. A significant issue is that the AI agent is only compared to randomly acting agents, which are not very hard to beat. To strengthen the study, it would be beneficial to include agents with distinct decision-making strategies—for example, one that accepts all offers, one that rejects most offers, and another using a simple calculation-based strategy. While I understand that implementing this might require additional effort, it would greatly enhance the depth and rigor of the research.

The grammar throughout the article is generally good, though there are occasional awkward phrasings and minor errors. For instance:

• The phrase “optimally of outcomes” should be corrected to “optimality of outcomes.”
• “Environmenal factors” should be revised to “environmental factors.”

The conclusions and discussion effectively summarize the findings but are too brief. They lack strong recommendations or implications for future research. Expanding these sections to include practical applications of your results would significantly increase the impact of your work.

After addressing the points regarding agent decision-making, improving the abstract and conclusions, and correcting the minor grammatical issues, I recommend the article for publication.

 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

Authors investigate whether the decision-making process of autonomous production units can be transferred from agents to artificial intelligence by using simulations of shared warehouse games as a simplified model of shared production in a virtual environment. The results show that artificial intelligence is successful in solving complex industrial problems, which are usually solved by game theory approaches. 

1.  In the abstract, the background description is too long, I advised to rewrite in order to reflect the purpose, method, results, and conclusion. 

2. The authors must clearly explain the difference(s) between the proposed method and similar works  to further highlight the manuscript's innovations and contributions in the introduction。

3. In the Section of Material and Methods, the describing of the proposed method is very simple. Please enhance the describing of the proposed method.  

4. The experimental details and parameters are not well explained in the manuscript in Section 4.

5. The authors need to interpret the meanings of the variables, such as Eq.(7), Eq.(8), ........

6. The literature review is poor in this paper. You must review all significant similar works that have been done. The article can be further enhanced by connecting with some existing literatures. For example,  https://doi.org/10.1109/TIM.2024.3485438; https://doi.org/10.1109/TIM.2024.3415778 and so on.

7. There are some grammatical mistakes and typo errors.

 

the describing of the proposed method is very simple. Please enhance the describing of the proposed method. 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 4 Report

Comments and Suggestions for Authors

 

This article explores Decentralized Warehouse Management Using Large Language Models: A Proof of Concept. The authors work needs several improvements as follows:

 

1.      The opening part does not clearly articulate the specific study topic or hypothesis being examined. Although it addresses collaborative manufacturing and AI-driven decision-making, it does not effectively link to the study's primary aims. This should be articulated to elucidate the objective of the investigation.

2.      The research lacks adequate background on the utilization of large language models (LLMs) in warehouse management. To enhance the theoretical framework, a comprehensive analysis of current studies about the role of LLMs in dispersed manufacturing is essential.

3.      Moreover, some more recent work should be considered in the literature. The authors can also refer to “Enhancing coherence and diversity in multi-class slogan generation systems,” “Mobile robot localization: Current challenges and future prospective,” and so on.

4.      The findings are predominantly qualitative and lack statistical validation or quantitative metrics to substantiate the assertions of AI's efficacy. Incorporating statistical comparisons and performance measures would substantially augment the credibility of the conclusions.

5.      The authors inadequately elucidate the incorporation of game theory into the AI's decision-making framework. They must detail the implementation of game theory principles and evaluate their influence on the AI's behavior under different circumstances.

6.      Figures 6-11 seem very general and low quality.

7.      Moreover, all the figure's text style should be identical to the paper body text.

8.      The research predominantly depends on simulations, neglecting the practicality of real-world applications. It is crucial to examine the practical obstacles and possible remedies for implementing AI-driven systems in decentralized warehouse settings.

9.      The technique section is inadequately developed, including scant information regarding the simulation setup, scenario design, and evaluation parameters. In this section, it is essential to elaborate on the experimental design and decision-making algorithms.

10.   The discourse around discrepancies in AI decision-making is superficial. To yield significant insights, a more thorough analysis is required to identify the potential reasons for these inconsistencies and recommend solutions.

11.   The existing references are very limited. The authors must cite several more of the latest works, especially in the literature review section.

 

 

Comments on the Quality of English Language

Need detailed revision 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 5 Report

Comments and Suggestions for Authors

This research work provided proof of concept for a cutting-edge technology, LLM, on decentralized warehouse management. However, there are various key issues to be considered and addressed by the authors:
Comment 1. Abstract:
(a) Referring to the content, “In our research, we investigate whether the decision-making process of autonomous production units can be transferred from agents to artificial intelligence by using simulations of shared warehouse games as a simplified model of shared production in a virtual environment. ” This does not align with the paper title.
(b) What are the research results and implications?
Comment 2. The introduction should clearly explain the necessity and importance of the research topic. In addition, please add a paragraph to highlight the research contributions.
Comment 3. Section 2 Theoretical background:
(a) As the theoretical background, the authors should provide necessary details that help readers understand the rest of the sections. The current discussion does not benefit the understanding of the following technical sections.
(b) Why did the authors consider AI and LLMs in one subsection?
Comment 4. Section 3 Literature review:
(a) The discussion is considered irrelevant. The scope should be related to the decentralized warehouse management and large language models.
(b) A precise and concise discussion of the methodology, results, and limitations of the latest works is desired.
Comment 5. Section 4 Materials and Methods:
(a) Please provide the full details of the package digital twin.
(b) The presented details do not help readers to understand how to build the system.
(c) Justify the setting in Table 1. By the way, the format of Table 1 (callout in the main text and heading) is not correct.
Comment 6. Enhance the resolution of all figures. Zoom your figures to ensure no content is blurred.

Comments on the Quality of English Language

The organization and written English needs to be improved.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

Great work in revising the paper. Thank you for taking my comments into consideration, the paper is, in my opinion, much better for it.

Congratulations for the work done here.
  Best regards

Author Response

Dear Reviewer,

I would like to sincerely thank you for the time and effort you devoted to reviewing our article. Your insightful comments were invaluable in improving the quality and clarity of the manuscript. Thank you to your constructive feedback, the article is now much stronger and more refined.

Thank you once again for your support and expertise.

Reviewer 2 Report

Comments and Suggestions for Authors

Dear Authors,

Thank you for taking my suggestions into consideration and revising your work accordingly. The poorly defined sections have been removed, the model is much clearer and more professional, and both the introduction and conclusion have been improved. For me, these revisions are sufficient. I accept the work without any further changes.

Author Response

Dear Reviewer,

I would like to sincerely thank you for the time and effort you devoted to reviewing our article. Your insightful comments were invaluable in improving the quality and clarity of the manuscript. Thank you to your constructive feedback, the article is now much stronger and more refined.

Thank you once again for your support and expertise.

Reviewer 3 Report

Comments and Suggestions for Authors

 

According to the revised paper, I have appreciated the deep revision of the contents and the present form of this manuscript. But there is still a little content, which need be revised according to the comment of reviewer in order to meet the requirements of publish. A number of concerns listed as follows:

1. The motivation of this paper should still be carefully presented in Introduction. The authors shall clarify why the problem of decentralised warehouse management should be investigated. 

2.  All the mathematical notation used throughout the paper should be defined. Some are not defined in this paper. 

3. The results should be compared with other recent methods.  The reviewer would say that there should have other efficient approaches. 

4. In order to further highlight the introduction, some advised references should be added to the paper for improving the review part and the connection with the literature.

 

Comments on the Quality of English Language

The English could be improved to more clearly express the research.

Author Response

Dear Reviewer,

I would like to sincerely thank you for the time and effort you devoted to reviewing our article. Your insightful comments were invaluable in improving the quality and clarity of the manuscript. Thank you to your constructive feedback, the article is now much stronger and more refined.

Thank you once again for your support and expertise.

Author Response File: Author Response.pdf

Reviewer 4 Report

Comments and Suggestions for Authors

The authors fail to address the comments carefully, and most importantly, the results presented in Figures 7-12 are very simple and do not give much information.

Comments on the Quality of English Language

 The English could be improved to more clearly express the research.

Author Response

Dear Reviewer,

I would like to sincerely thank you for the time and effort you devoted to reviewing our article. Your insightful comments were invaluable in improving the quality and clarity of the manuscript. Thank you to your constructive feedback, the article is now much stronger and more refined.

Thank you once again for your support and expertise.

Author Response File: Author Response.pdf

Reviewer 5 Report

Comments and Suggestions for Authors

The authors have made extensive revisions to the article. However, there are various key comments that remain unaddressed, and a major revision is recommended.
Comment 1. Enhance clarity. Avoid presenting with lengthy paragraphs.
Comment 2. Clearly state the research contributions.
Comment 3. Present a literature review before theoretical background to enhance clarity and organization.
Comment 4. As a literature review, it is expected the authors will cite the latest references, particularly since we are approaching 2025.
Comment 5. The theoretical background did not provide the necessary information to help readers understand the rest of the content. Limited equations were shared in each part.
Comment 6. Since experiments are based on simulation, more scenarios (Table 1) are expected.
Comment 7. In many figures (7-12), x-labels and y-labels are missing.
Comment 8. Overall, as a proof of concept, please clarify which proof was confirmed.

Author Response

Dear Reviewer,

I would like to sincerely thank you for the time and effort you devoted to reviewing our article. Your insightful comments were invaluable in improving the quality and clarity of the manuscript. Thank you to your constructive feedback, the article is now much stronger and more refined.

Thank you once again for your support and expertise.

Author Response File: Author Response.pdf