Intelligent Agents for Sustainable Maritime Logistics: Architectures, Applications, and the Path to Robust Autonomy

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This is a review paper focusing on the analysis of the application of intelligent agents in the field of sustainable maritime logistics. The analysis in the paper is relatively comprehensive and the research conclusions are relatively credible. The following points need to be further modified and improved:

1. The research objectives should be further clarified in the introduction section.
2. For Section 3, it is recommended to provide a summary discussion for each application area rather than listing literature content, especially in 3.2.1 and 3.2.2.
3. The references in Table 2 have already been cited and explained in their respective application fields, is it necessary to list separately?
4. It is suggested to supplement the limitations of the current study.
5. It is suggested to analyze more future research applications, such as the impact of generative AI and so on.
6. The conclusion section should include some suggestions for development.

 

Author Response

We thank the reviewer for the positive assessment of our work and the encouraging feedback. We have addressed all your suggestions below.

Comment 1: The research objectives should be further clarified in the introduction section.
Response: We agree. We have expanded the final paragraph of the Introduction (Section 1) to clearly state three explicit research questions and the unique value proposition of this review.

Comment 2: For Section 3, it is recommended to provide a summary discussion for each application area rather than listing literature content, especially in 3.2.1 and 3.2.2.
Response: Thank you for this suggestion. To provide a stronger critical synthesis rather than a simple listing, we have added new synthesis paragraphs. Specifically, at the end of Section 3.2.1, we added a discussion on measurable KPIs and alternative power pathways. Furthermore, we added a dedicated synthesis paragraph in Section 3.3 regarding the deployment maturity of the reviewed literature.

Comment 3: The references in Table 2 have already been cited and explained in their respective application fields, is it necessary to list separately?
Response: We appreciate this observation. While the references are discussed in the text, Table 2 serves as a consolidated decision-grade framework for readers. To clarify its purpose, we added a paragraph immediately preceding the table in Section 3.3, explaining that the table explicitly highlights architectural choices and frames the deployment maturity gap.

Comment 4: It is suggested to supplement the limitations of the current study.
Response: We have completely restructured Section 5 (Conclusions). The revised conclusion now features a dedicated second paragraph that explicitly discusses the limitations observed both within the reviewed literature and our study (e.g., the deployment maturity gap and the lack of standardized KPIs).

Comment 5: It is suggested to analyze more future research applications, such as the impact of generative AI and so on.
Response:  We have added a discussion on Generative AI and Large Language Models (LLMs) in Section 4.2 (The "Black Box" Problem), highlighting their potential role in translating numerical AI weights into human-readable XAI reports. We also emphasized this in the final recommendations of the Conclusion.

Comment 6: The conclusion section should include some suggestions for development.
Response: Following your advice, the third paragraph of the newly restructured Section 5 (Conclusions) is now fully dedicated to actionable recommendations for future research and development (e.g., sim-to-real transfer, Generative AI, and zero-trust cybersecurity).

Reviewer 2 Report

Comments and Suggestions for Authors

This manuscript provides a comprehensive review of intelligent agent architectures and their potential applications in sustainable maritime logistics. The topic is timely and relevant given the increasing digitalization of maritime operations and the growing pressure on the sector to improve environmental performance and operational efficiency. The manuscript offers a structured overview of different types of intelligent agents, including reactive, deliberative, hybrid, and multi-agent systems, and discusses how these architectures can support applications such as autonomous navigation, energy efficiency optimization, predictive maintenance, and port coordination. The classification presented in the paper is clear and the literature coverage is generally broad, including several recent studies. The manuscript therefore provides a useful overview of technological developments at the intersection of artificial intelligence and maritime logistics. At the same time, several aspects of the manuscript could be strengthened to improve its scholarly contribution and analytical clarity. First, although the paper presents a large number of studies, the literature synthesis is mostly descriptive. Many studies are summarized sequentially, but there is limited critical comparison between them. The paper would benefit from a more analytical synthesis that highlights differences in methodological approaches, maturity levels of the technologies discussed, and the extent to which the proposed solutions have been validated in real operational environments versus simulation settings. A clearer comparison of these dimensions would strengthen the value of the review for both researchers and practitioners. Second, as the manuscript is presented as a review article, it would be helpful to briefly clarify the approach used to identify and select the literature. The manuscript does not indicate whether the review follows a systematic, semi-systematic, or narrative review approach. A short explanation of the search strategy, time frame, and selection criteria for the included studies would improve transparency and methodological rigor. Third, although sustainability is introduced as a key motivation for the study, the connection between intelligent agent technologies and concrete sustainability outcomes could be articulated more explicitly. In several sections, sustainability is mentioned primarily at a conceptual level, while the discussion focuses more strongly on technological capabilities. The manuscript would benefit from more clearly explaining how the reviewed technologies contribute to measurable sustainability improvements, such as reductions in fuel consumption, emissions, or operational inefficiencies within maritime logistics systems. Fourth, the section discussing current challenges and future directions is informative and highlights several important issues such as the simulation-to-reality gap, explainability of AI systems, interoperability, and cybersecurity. This section is one of the strongest parts of the manuscript and could be further strengthened by linking the identified challenges more directly to specific limitations observed in the studies reviewed earlier in the paper. Finally, the concluding section could be improved by providing a clearer synthesis of the main insights derived from the review rather than primarily reiterating earlier discussions. A more structured conclusion that highlights the key contributions of the review, its implications for future research, and the main practical considerations for maritime stakeholders would enhance the overall clarity of the manuscript. Overall, the manuscript addresses an important and rapidly evolving topic and provides a useful overview of intelligent agent technologies in maritime logistics. With revisions aimed at strengthening the analytical synthesis of the literature, clarifying the review methodology, and more explicitly connecting the discussion to sustainability outcomes, the paper would offer a more robust contribution to the field.

Author Response

We thank the reviewer for the positive assessment of our work and the encouraging feedback. We have addressed all your suggestions below.

General Comment: This manuscript provides a comprehensive review of intelligent agent architectures... With revisions aimed at strengthening the analytical synthesis of the literature, clarifying the review methodology, and more explicitly connecting the discussion to sustainability outcomes, the paper would offer a more robust contribution to the field.
Response: We sincerely thank the reviewer for highlighting the strengths of our paper, particularly the section on challenges. We have carefully implemented your suggestions to strengthen the methodological transparency and analytical synthesis.

Comment 1: The literature synthesis is mostly descriptive... The paper would benefit from a more analytical synthesis that highlights differences in methodological approaches, maturity levels of the technologies discussed, and the extent to which the proposed solutions have been validated in real operational environments versus simulation settings.
Response: We completely agree. To address this, we added a critical synthesis paragraph in Section 3.3 (Synthesis and Comparative Overview) right before Table 2. This new text explicitly addresses the deployment maturity gap, highlighting that while algorithmic capabilities are advanced, physical deployment remains a rare exception compared to simulation environments.

Comment 2: It would be helpful to briefly clarify the approach used to identify and select the literature. The manuscript does not indicate whether the review follows a systematic, semi-systematic, or narrative review approach.
Response: Thank you for pointing this out. We have updated the final paragraph of Section 1 (Introduction) to explicitly state that this is a semi-systematic review. We also detailed the search protocol, including the major academic databases used, the timeframe (2015–2025), and the primary keywords.

Comment 3: The connection between intelligent agent technologies and concrete sustainability outcomes could be articulated more explicitly... explaining how the reviewed technologies contribute to measurable sustainability improvements.
Response: We agree that sustainability needs to be grounded in metrics. We added a new paragraph at the end of Section 3.2.1 (Intelligent Port Operations) explicitly discussing the need for measurable Key Performance Indicators (KPIs) such as reductions in port waiting times, localized emissions, and the integration of onshore power supply (OPS) and microgrids.

Comment 4: This section [Section 4] could be further strengthened by linking the identified challenges more directly to specific limitations observed in the studies reviewed earlier in the paper.*
Response: We have strengthened this link in the new synthesis paragraph in Section 3.3. By explicitly identifying the "maturity gap" (over-reliance on simulation) in the reviewed literature, we establish a direct bridge to the systemic challenges discussed in Section 4 (specifically, the sim-to-real gap).

Comment 5: The concluding section could be improved by providing a clearer synthesis of the main insights... A more structured conclusion that highlights the key contributions of the review, its implications for future research, and the main practical considerations.
Response: We have completely rewritten Section 5 (Conclusions). It is now strictly structured into three distinct parts: (1) Main outcomes and contributions, (2) Limitations (maturity gaps, lack of KPIs), and (3) Actionable recommendations for future research and practical development.

Reviewer 3 Report

Comments and Suggestions for Authors

Dear Authors,

Your manuscript, “Intelligent Agents for Sustainable Maritime Logistics: Architectures, Applications, and the Path to Robust Autonomy,” presents valuable ideas. That said, I would like to share several non-mandatory recommendations that may help strengthen the overall quality, clarity, and academic positioning of the paper:

1. Clarify the manuscript’s unique value proposition versus prior reviews on maritime autonomy/AI by stating 2–3 explicit research questions and a crisp novelty claim (e.g., new taxonomy, maturity model, or actionable research agenda).
2. Add a transparent review protocol (databases, timeframe, keywords, screening steps, inclusion/exclusion criteria, and final paper count). This is essential to de-risk selection bias.
3. The introduction needs more explanation about the main object, the novelty of your issues, and the last paragraph must number and elaborate on the remaining sections of the manuscript.t.
4. Your agent/architecture taxonomy is promising, but classification rules are not operational. Define the decision criteria for reactive, deliberative, hybrid, and multi-agent systems, and include 1–2 borderline examples.
5. Tables summarising approaches would be more decision-grade if you add: typical algorithms, sensor/data prerequisites, safety assurance method, and deployment maturity (simulation vs HIL vs sea trial vs operations).
6. Figures describing the autonomy stack are clear but generic. Tie each block to cited implementations (specific sensors, fusion approach, planner type, COLREG encoding method, and compute placement/latency).
7. COLREG compliance is repeatedly claimed but not formalised. Specify how compliance is encoded and validated (rule-based constraints, game-theoretic models, imitation learning, scenario-based test suites, or formal verification).
8. The sustainability discussion needs measurable KPIs. Map each application area to clear indicators (CO₂/fuel, port waiting time, berth productivity, emissions at berth, congestion) and report what the reviewed studies actually measured.
9. Consider grading evidence quality across studies (simulation-only, HIL, field pilots, operational deployments). This will substantiate maturity-gap statements and prevent over-generalisation.
10. Coverage is ship-navigation heavy; port/landside logistics is thinner and more heterogeneous. Either deepen the port logistics thread or re-scope the paper explicitly as MASS-dominant plus selected port cases.
11. Re-check consistency between your definitions and how methods are labelled in the tables (some DRL policies could be argued as reactive at runtime even if trained with deliberative objectives). Align terminology end-to-end.
12. Strengthen the interoperability discussion by anchoring “intention sharing” and data exchange to established maritime digitalization initiatives/standards, then show exactly where agent-based layers add incremental value.
13. The safety roadmap would be stronger if you integrate system-safety practice: hazard analysis (e.g., STPA), assurance cases, scenario catalogs, regression testing pipelines, and traceability from hazards to controls and evidence.
14. Human–agent teaming needs more operational detail: handover triggers, mode awareness, alarm philosophy, training requirements, and measurable human-factors outcomes (workload, situational awareness, error rates).
15. Cybersecurity is currently high-level. Add a threat model (trust boundaries, attack surfaces) and minimum viable controls for edge autonomy (authentication, key management, anomaly detection, secure comms), linked to safety impact.
16. To further strengthen the sustainability narrative, it would be valuable to expand the discussion on alternative power supply pathways in maritime logistics—such as shore power (OPS), electrification options, renewable integration, and hybrid energy architectures- and position these as high-impact levers for decarbonization. This would also strengthen your intelligent-agent argument by highlighting energy-aware optimisation for routing, port calls, berth allocation, and operational planning. In this regard, you can gain good information from a research of “J. Mar. Sci. Eng.
17. 2024, 12(8), 1290”.
18. The section conclusion must have 3 different pieces of information: the main outcome of the research, the limitations that you faced when you were engaged in the research and lastly, some recommendations for future research in the same field.

Thank you for your attention.

Author Response

We thank the reviewer for the positive assessment of our work and the encouraging feedback. We have addressed all your suggestions below.

General Comment: Your manuscript... presents valuable ideas. That said, I would like to share several non-mandatory recommendations that may help strengthen the overall quality, clarity, and academic positioning of the paper.

Response: We are highly appreciative of your thorough review and the deep domain expertise reflected in your non-mandatory recommendations. We have integrated your core suggestions, which have significantly sharpened the focus of our manuscript.

Comments 1, 2 & 3: Clarify the manuscript’s unique value proposition... Add a transparent review protocol... The introduction needs more explanation.
Response: We have expanded the final paragraph of  Section 1 (Introduction) to explicitly outline our semi-systematic review protocol (databases, timeframe, keywords). Furthermore, we formulated three explicit research questions to clearly state our unique value proposition and outline the paper's structure.

Comments 4 & 11: Your agent/architecture taxonomy is promising, but classification rules are not operational... Re-check consistency between your definitions and how methods are labelled (some DRL policies could be argued as reactive at runtime).
Response: To address the operational nuance, we added a paragraph in Section 2, right after the deliberative agents' subsection. We explicitly clarified that algorithms like DRL blur these boundaries: while training is deliberative, the deployed policy often functions reactively at runtime.

Comments 5 & 9:Tables summarising approaches would be more decision-grade if you add deployment maturity... Consider grading evidence quality across studies.
Response: We completely agree with the need to grade evidence quality. Instead of adding a repetitive column to the table (since the vast majority of current literature is exclusively simulation-based), we addressed this critically in the text. We added a dedicated synthesis paragraph in Section 3.3 that evaluates the deployment maturity gap across the studies, noting the lack of HIL or sea-trial validation.

Comments 8 & 16:The sustainability discussion needs measurable KPIs... it would be valuable to expand the discussion on alternative power supply pathways (OPS) and position these as high-impact levers. (Reference: J. Mar. Sci. Eng. 2024, 12(8), 1290).
Response: We are grateful for this suggestion and the excellent reference. We added a paragraph at the end of Section 3.2.1 addressing the need for measurable KPIs (e.g., emissions, waiting times). We also expanded the discussion on Onshore Power Supply (OPS) and renewable microgrids, explicitly citing the suggested paper [36] to strengthen the sustainability narrative.

Comments 13 & 15: The safety roadmap would be stronger if you integrate system-safety practice... Cybersecurity is currently high-level. Add a threat model.
Response: To address the need for a more operational security perspective, we expanded Section 4.5 (Cybersecurity). We explicitly incorporated concepts of threat modeling, defining trust boundaries, and acknowledging attack surfaces (such as sensor spoofing) that can cascade into safety-critical navigational failures.

Comment 17: The section conclusion must have 3 different pieces of information: the main outcome, limitations, and recommendations.
Response: Following your precise guidance, we have entirely restructured Section 5 (Conclusions). It now consists of exactly three paragraphs: the first synthesizes the main outcomes, the second explicitly states the limitations (maturity gap, lack of KPIs, black-box trust), and the third provides targeted recommendations for future research.

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

The paper has been revised according to the review comments, and the research purpose has been achieved.

Author Response

Comment:
The paper has been revised according to the review comments, and the research purpose has been achieved.

Response:
We sincerely thank the reviewer for the positive assessment of the revised manuscript and for recognizing that the research objectives have been successfully achieved. We appreciate the constructive feedback provided during the first review round, which helped us improve the clarity and positioning of the paper. We are pleased that the revised version meets the reviewer’s expectations.

Reviewer 3 Report

Comments and Suggestions for Authors

Dear authors, thanks for resolving the issues of your manuscript. I just want to let you know I expect more explanations about the outcome of your research in the conclusion section. 

Author Response

Comment:
I expect more explanations about the outcome of your research in the conclusion section.

Response:

We thank the reviewer for this valuable suggestion. The outcomes of the research are articulated in the Conclusion section through the synthesis of the reviewed literature and the identification of key structural patterns in the application of intelligent agents in maritime logistics.

As this paper adopts a semi-systematic review methodology, the primary research outcomes are conceptual and analytical rather than empirical. Specifically, the paper contributes:
    1.    A domain-adapted taxonomy of intelligent agent architectures (reactive, deliberative, hybrid, and multi-agent systems) contextualized for maritime logistics.
    2.    A structured synthesis of application domains, distinguishing between ship-centric autonomy, port operations, and supply chain coordination.
    3.    Identification of a consistent maturity gap between simulation-based validation and real-world deployment of intelligent agent systems.
    4.    A research agenda highlighting key directions necessary for robust autonomy, including explainable AI, sim-to-real transfer methodologies, ontology-based interoperability, and zero-trust cybersecurity architectures.

These outcomes are summarized in the final paragraphs of the Conclusion section, where the paper emphasizes that hybrid architectures dominate ship-centric applications, while multi-agent systems prevail in port and supply-chain coordination contexts. The conclusion further synthesizes the reviewed literature by identifying the deployment maturity gap as the primary structural barrier to industrial adoption.