Research on Autonomous Vehicle Lane-Keeping and Navigation System Based on Deep Reinforcement Learning: From Simulation to Real-World Application

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This paper presents an autonomous vehicles lane keeping and navigation system.

DRL was used for training and test by four different algorithms. Besides, a simulation for real  is applied to overcome real world limitation.

 

1. Some figures are hard to read, the resolution is too bad ,such as the fig. 8, fig.9, fig.16, please update these figures making them easy to read.
2. The references seems did not update for many years, such as the [15] and [20], accessed on 8 December 2020, the [1] accessed on 9 December 2021. Besides, there was no new recently published references, can we infer that this paper was likely written around 2021? and within the past four years, the authors did not update this paper? In fact, the autonomous area developed very fast with the recent AI technology, strongly suggest authors to update the references and also revise some comments in the full content.
3. For the lane keeping state, the author only gave one duckiebotvehicle case, other vehicles influences on lane keeping is very important, so it is better to consider other vehicles influences on lane keeping, as well as the navigation.

Author Response

Response To Reviewer Comments

 

The authors would like to thank the anonymous reviewers for their careful examination of this paper and the constructive comments they have provided. In this letter, we respond to each of the comments and show how the paper has been modified in response to the reviewers’ recommendations.

 

1.Some figures are hard to read, the resolution is too bad ,such as the fig. 8, fig.9, fig.16, please update these figures making them easy to read.

Response:
    Thank you for pointing out the issue regarding figure quality. We have updated Figures 8, 9, and 16 with higher-resolution versions to ensure that the content is clear and the annotations are easy to read, thereby improving overall readability.

2.The references seems did not update for many years, such as the [15] and [20], accessed on 8 December 2020, the [1] accessed on 9 December 2021. Besides, there was no new recently published references, can we infer that this paper was likely written around 2021? and within the past four years, the authors did not update this paper? In fact, the autonomous area developed very fast with the recent AI technology, strongly suggest authors to update the references and also revise some comments in the full content.

Response:
    Thank you for this important suggestion. We have carefully reviewed and updated the references throughout the manuscript and incorporated more recent research to enhance the timeliness and academic completeness of the paper.

3.For the lane keeping state, the author only gave one duckiebot vehicle case, other vehicles influences on lane keeping is very important, so it is better to consider other vehicles influences on lane keeping, as well as the navigation.

Response:
    Thank you for your suggestion. Due to limitations in funding, equipment, and available testing space, we primarily used the Duckiebot platform for our experiments. This platform offers flexibility and ease of adjustment, which supports iterative testing and development. We have clarified this limitation in the revised manuscript.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

The workload of this paper is relatively substantial, but there are still significant issues in areas such as chart presentation and writing. It would require at least a major revision before it can be considered for publication.

Specific suggestions:

1. In line 9 of the abstract, before the first sentence, it is recommended to add two sentences. One should introduce the background and significance of the study, while the other should highlight the shortcomings of current methods, thereby leading to the problem that the authors aim to solve using the DRL approach.

2. Currently, the title of section 2 is "Background." However, in my view, the first paragraph of section 1, "Introduction," actually serves as the background for this paper. Section 2 of the article is dedicated to reviewing current methods.

3. As the authors mentioned in the introduction, autonomous driving requires multiple sensors, such as cameras, radars, as well as inertial navigation systems, odometers, etc. However, these sensors provide relative positions within local or self-consistent coordinate systems, rather than absolute positions in the real-world coordinate system. For the "real world application" mentioned in the title, there is actually a need for a true reference provided by GNSS or pseudolites, which is likely to be urgently required for the next generation of autonomous driving. Therefore, the authors might consider referencing and analyzing this article in the introduction: Liu, T., Liu, J., Wang, J., Zhang, H., Zhang, B., Ma, Y., ... & Xu, G. (2023). Pseudolites to support location services in smart cities: Review and prospects. Smart Cities, 6(4), 2081-2105.

4. In fact, the most critical aspect of current autonomous driving technology is safety. However, this paper addresses this issue relatively little. Perhaps a discussion on this should be added.

5. The figures in the article currently do not meet the standards for publication. For example, the font size in Figures 8 and 9 is too small, while that in Figure 14 is too large. What is the significance of Figure 29? Perhaps Figure 29 should be removed. It is rare to see articles with a similar approach to figure presentation.

6. The experimental discussion and conclusion sections lack a comparison with current methods. That is, how much improvement does the DRL method proposed in this paper offer compared to existing methods, and what is the significance of this improvement? In short, this paper reads more like an experimental report than an academic paper.

Author Response

Response To Reviewer Comments

 

The authors would like to thank the anonymous reviewers for their careful examination of this paper and the constructive comments they have provided. In this letter, we respond to each of the comments and show how the paper has been modified in response to the reviewers’ recommendations.

 

1. In line 9 of the abstract, before the first sentence, it is recommended to add two sentences. One should introduce the background and significance of the study, while the other should highlight the shortcomings of current methods, thereby leading to the problem that the authors aim to solve using the DRL approach.

Response:

    Thank you for the suggestion. In the revised manuscript, we have added background information and the shortcomings of existing methods in the abstract to emphasize the significance of applying deep reinforcement learning in this study.

 

2. Currently, the title of section 2 is "Background." However, in my view, the first paragraph of section 1, "Introduction," actually serves as the background for this paper. Section 2 of the article is dedicated to reviewing current methods.

Response:

    We have renamed Section 2 from "Background" to "Related Work" and reorganized its content to clearly separate the background information in the Introduction from the review of current methods, improving the manuscript’s structure and readability.

 

3. As the authors mentioned in the introduction, autonomous driving requires multiple sensors, such as cameras, radars, as well as inertial navigation systems, odometers, etc. However, these sensors provide relative positions within local or self-consistent coordinate systems, rather than absolute positions in the real-world coordinate system. For the "real world application" mentioned in the title, there is actually a need for a true reference provided by GNSS or pseudolites, which is likely to be urgently required for the next generation of autonomous driving. Therefore, the authors might consider referencing and analyzing this article in the introduction: Liu, T., Liu, J., Wang, J., Zhang, H., Zhang, B., Ma, Y., ... & Xu, G. (2023). Pseudolites to support location services in smart cities: Review and prospects. Smart Cities, 6(4), 2081-2105.

Response:

    We sincerely thank the reviewer for taking the time to evaluate our submission and for providing valuable reference literature. The recommended paper significantly enriches our understanding and background regarding localization systems, and we appreciate your guidance.

 

4. In fact, the most critical aspect of current autonomous driving technology is safety. However, this paper addresses this issue relatively little. Perhaps a discussion on this should be added.

Response:

    We appreciate the reviewer’s insightful comment. We have specifically emphasized the critical role of safety in autonomous driving in the conclusion section. Furthermore, we explain how our approach employs simulation-based training and Sim2Real techniques to reduce risks associated with real-world testing. We will continue to deepen our discussion on safety aspects in future work to enhance the practical applicability of our system.

 

5. The figures in the article currently do not meet the standards for publication. For example, the font size in Figures 8 and 9 is too small, while that in Figure 14 is too large. What is the significance of Figure 29? Perhaps Figure 29 should be removed. It is rare to see articles with a similar approach to figure presentation.

Response:

    We have revised all figures according to the reviewer’s suggestions, adjusting font sizes and layout to improve clarity and readability. Regarding Figure 29, this figure visualizes the attention of the convolutional layers in the neural network by fusing feature maps from multiple layers and overlaying the resulting heatmap on the original input image. This visualization helps demonstrate the consistency of the model’s focus areas between the simulation and real-world environments, thereby providing valuable insights into the model’s interpretability and the effectiveness of the Sim2Real approach. Considering its significance in illustrating these points, we believe it is important to retain this figure in the manuscript.

 

6. The experimental discussion and conclusion sections lack a comparison with current methods. That is, how much improvement does the DRL method proposed in this paper offer compared to existing methods, and what is the significance of this improvement? In short, this paper reads more like an experimental report than an academic paper.

Response:

    Thanks very much for the reviewer’s helpful suggestion about adding more comparisons with existing methods. Because of differences in datasets and setups, it’s a bit challenging to do direct quantitative comparisons in this current study. That said, we have added some discussion to better explain how our approach compares to other work based on the available literature. We agree that this is an important aspect, and we plan to strengthen and improve this part in our future work. We believe that further efforts in this direction will help demonstrate the strengths and practical value of our approach more clearly.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

Overall, the paper exhibits excellent quality, and the creation of a deep reinforcement learning-based autonomous vehicle system enables the Duckiebot to maintain its lane in both simulated and real-world environments.ironments. Moreover, this technique is designed to determine the shortest route between the starting point and the destination in order to accomplish the second phase of the navigation function. This model can be utilized in the navigation of lane-keeping; the Dueling DQN algorithm is the most effective, based on the experimental results mentioned in the paper. When it involves navigating or lane-keeping, the competing DQN algorithm works well as compared to your other mathematical algorithms. However, the Dueling DQN algorithm not only outperforms the other algorithms in terms of efficiency but also demonstrates greater adaptability to varying traffic conditions. This adaptability is crucial for real-time applications, ensuring safe and reliable navigation in complex environments. You can add few more references/citations in the recent development on these areas of innovations and future studies that could be possible more exciting to readers and researchers, otherwise it is a great paper.

Author Response

Response To Reviewer Comments

 

The authors would like to thank the anonymous reviewers for their careful examination of this paper and the constructive comments they have provided. In this letter, we respond to each of the comments and show how the paper has been modified in response to the reviewers’ recommendations.

 

1. Overall, the paper exhibits excellent quality, and the creation of a deep reinforcement learning-based autonomous vehicle system enables the Duckiebot to maintain its lane in both simulated and real-world environments. Moreover, this technique is designed to determine the shortest route between the starting point and the destination in order to accomplish the second phase of the navigation function. This model can be utilized in the navigation of lane-keeping; the Dueling DQN algorithm is the most effective, based on the experimental results mentioned in the paper. When it involves navigating or lane-keeping, the competing DQN algorithm works well as compared to your other mathematical algorithms. However, the Dueling DQN algorithm not only outperforms the other algorithms in terms of efficiency but also demonstrates greater adaptability to varying traffic conditions. This adaptability is crucial for real-time applications, ensuring safe and reliable navigation in complex environments. You can add few more references/citations in the recent development on these areas of innovations and future studies that could be possible more exciting to readers and researchers, otherwise it is a great paper.

Response:

        We sincerely appreciate the reviewer taking the time to review our manuscript. In response to the suggestion to incorporate more recent relevant studies and references, we have added several important recent publications in the revised manuscript and appropriately cited them throughout the text. This has helped enhance the completeness and forward-looking nature of the research background and technical context.

Author Response File: Author Response.pdf

Reviewer 4 Report

Comments and Suggestions for Authors

The paper presents a limited real-word validation, with experiments being conducted in a small custom-build Dukiebot platform, not on a full sized vehicle or in complex real world traffic scenarios. This implies that it is a small-scale testing in a limited environment complexity, lacking diversity, unpredictability and the complexity of actual road conditions.

The paper acknowledges discrepancies between simulated and real environment and respective insufficient domain adaptation. Moreover, the paper only focus on lane keeping and basic navigation, which leaves outside of the study parameters such as obstacle avoidance, traffic signals, interaction with other vehicles or pedestrians and these are absolutely critical for autonomous driving.

Furthermore, 4 DRL algorithms are compared, but the study does not compare it with traditional methods (non DRL) and therefore, it is not possible to evaluate the added value of DRL.

Lastly, there is no in-depth analysis about failure cases, error types or conditions under which the models fail after sim-to-real transfer. The fact that the deployment is on a Jetson Nano and a small robot, may not be generalized to the computational and sensor requirements of full autonomous vehicles, which puts into question the scalability of the approach. No mention of releasing code, trained models or datasets, which limits the ability of others to replicate or build upon the work.

Author Response

Response To Reviewer Comments

 

The authors would like to thank the anonymous reviewers for their careful examination of this paper and the constructive comments they have provided. In this letter, we respond to each of the comments and show how the paper has been modified in response to the reviewers’ recommendations.

 

1. The paper presents a limited real-word validation, with experiments being conducted in a small custom-build Dukiebot platform, not on a full sized vehicle or in complex real world traffic scenarios. This implies that it is a small-scale testing in a limited environment complexity, lacking diversity, unpredictability and the complexity of actual road conditions.

Response:

        Thank you very much for your valuable comments. Due to cost and space limitations, we were unable to test on full-sized real vehicles, and instead used a small autonomous vehicle platform to allow for greater flexibility and convenience. In the future, if conditions permit, we plan to expand the scale and complexity of the simulation environment and even attempt to transfer the model to real vehicles for testing, in order to further validate the practicality and feasibility of our approach.

2. The paper acknowledges discrepancies between simulated and real environment and respective insufficient domain adaptation. Moreover, the paper only focus on lane keeping and basic navigation, which leaves outside of the study parameters such as obstacle avoidance, traffic signals, interaction with other vehicles or pedestrians and these are absolutely critical for autonomous driving.

Response:

        Thank you for your valuable suggestion. This study currently focuses primarily on lane keeping and basic navigation functionalities, representing an initial experimental stage. In the future, we plan to gradually integrate more realistic driving scenarios and complex conditions to advance more comprehensive and in-depth research in autonomous driving.

3. Furthermore, 4 DRL algorithms are compared, but the study does not compare it with traditional methods (non DRL) and therefore, it is not possible to evaluate the added value of DRL.

Response:

        Thank you for your valuable feedback. Regarding comparisons with traditional methods, since this study primarily aims to evaluate the performance of deep reinforcement learning algorithms on specific tasks, comprehensive direct comparisons with traditional non-DRL methods remain challenging. In the future, we will consider incorporating additional methods for comparison to further enhance the completeness and persuasiveness of our research.

4. Lastly, there is no in-depth analysis about failure cases, error types or conditions under which the models fail after sim-to-real transfer. The fact that the deployment is on a Jetson Nano and a small robot, may not be generalized to the computational and sensor requirements of full autonomous vehicles, which puts into question the scalability of the approach. No mention of releasing code, trained models or datasets, which limits the ability of others to replicate or build upon the work.

Response:

        Thank you for your detailed comments. We sincerely apologize for the lack of in-depth analysis regarding failure cases, error types, and conditions under which the models fail after Sim2Real transfer. We commit to strengthening the related investigations and analyses in our future research. Regarding the public release of code, trained models, and datasets, due to certain considerations, we are currently unable to provide them. We appreciate your understanding.

Author Response File: Author Response.pdf

Reviewer 5 Report

Comments and Suggestions for Authors

This paper  proposed a  Lane-Keeping and Navigation method using  deep reinforcement learning. However, the contents and contributions of the current version do not meet the standard of scientific publication.

Author Response

Response To Reviewer Comments

 

The authors would like to thank the anonymous reviewers for their careful examination of this paper and the constructive comments they have provided. In this letter, we respond to each of the comments and show how the paper has been modified in response to the reviewers’ recommendations.

 

1. This paper  proposed a  Lane-Keeping and Navigation method using  deep reinforcement learning. However, the contents and contributions of the current version do not meet the standard of scientific publication.

 

Response: Thank you for taking the time and effort to review our manuscript. We sincerely thank the reviewer for taking the time to review our manuscript. In response to the comments, we have revised the paper by strengthening the description of technical contributions, updating recent relevant references, and improving the quality and explanations of the figures. We hope these changes enhance the scientific value of the manuscript and adequately address the reviewer’s valuable suggestions.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

After carefully read the revised version, the reviewer think there are still some parts need to be improved.

1. In the abstract section, the author tried to explain why  they using the deep learning in the proposed method. However, it is not necessary give so much descriptions to introduce the importance of the reinforcement and deep reinforcement learning. Thus the second paragraph in the abstract should be  more concise.
2. There are still some figures need to be improved with high resolution. Such as the fig.8,fig.9 , fig.16. the words or font size in these figures were too small, and some parameters formatting were too bad,  such the “st”in the middle of fig.8.  Besides, variables should be italicized, please revise and check all the variables in the whole paper.   
3. The font size in different figures are quite different, such as in fig.7 is relative good, but in  fig.29 the font size of “0, 1”is too big.
4. The related work review need improve. Most of the review references were used to explain some special algorithm, but no give any deep review about the reference itself, some key parts should be contained when review related works, such as, what method was proposed, and how the method work and what kinds of results were obtained in these references. Though the authors did many improvements on this section, but still need to do more deep review.   
5. The novelty or contribution should be clearly mentioned in section 1 introduction. The current section 1 is too simple .
6. In page 28, table 5, the success rate is dueling DQN 62% and  PPO 52%, to be honesty, the success rate is indeed relatively low. The author should discuss potential reasons for these.  Besides, in table 3, the success rate is 96.4% and 66%. the gap between the table 3 and table 5 also should be explained.

Author Response

Response To Reviewer Comments

 

The authors would like to thank the anonymous reviewers for their careful examination of this paper and the constructive comments they have provided. In this letter, we respond to each of the comments and show how the paper has been modified in response to the reviewers’ recommendations.

 

1.In the abstract section, the author tried to explain why  they using the deep learning in the proposed method. However, it is not necessary give so much descriptions to introduce the importance of the reinforcement and deep reinforcement learning. Thus the second paragraph in the abstract should be  more concise.

Response:

    Thank you for your valuable suggestion on the abstract. In response to your comment, we have revised the second paragraph to remove redundant background explanations, making it more concise and focused. The revised content has been highlighted in yellow in the manuscript.

 

2.There are still some figures need to be improved with high resolution. Such as the fig.8,fig.9 , fig.16. the words or font size in these figures were too small, and some parameters formatting were too bad,  such the “st”in the middle of fig.8.  Besides, variables should be italicized, please revise and check all the variables in the whole paper.

Response:

    We appreciate your feedback on the figure quality. We have replaced Figures 8, 9, and 16 with higher-resolution versions, adjusted the font sizes for better readability, and corrected all variable formatting to be italicized as appropriate. These changes have been highlighted in yellow in the manuscript.

 

3.The font size in different figures are quite different, such as in fig.7 is relative good, but in  fig.29 the font size of “0, 1”is too big.

Response:

    We have carefully reviewed and standardized the font sizes in all figures. Specifically, the font size in Figure 29 (“0, 1”) has been adjusted to match the overall style and to ensure visual consistency with other figures such as Figure 7. This revision has also been highlighted in yellow in the manuscript.

 

4.The related work review need improve. Most of the review references were used to explain some special algorithm, but no give any deep review about the reference itself, some key parts should be contained when review related works, such as, what method was proposed, and how the method work and what kinds of results were obtained in these references. Though the authors did many improvements on this section, but still need to do more deep review.

Response:

    Thank you for your valuable comments on the related work section. Based on your suggestions, we have made appropriate revisions and additions to improve the completeness and clarity of this part. The revised content has been highlighted in yellow in the manuscript.

 

5.The novelty or contribution should be clearly mentioned in section 1 introduction. The current section 1 is too simple .

Response:

Thank you for pointing out the insufficient description of the research contributions in the introduction. As per your suggestion, we have revised Section 1 to include a clearer statement of our study's contributions and research positioning. We hope this provides readers with a better understanding of the value and objectives of our work. The revised content has been highlighted in yellow in the manuscript.

 

6.In page 28, table 5, the success rate is dueling DQN 62% and  PPO 52%, to be honesty, the success rate is indeed relatively low. The author should discuss potential reasons for these.  Besides, in table 3, the success rate is 96.4% and 66%. the gap between the table 3 and table 5 also should be explained.

Response:

    Thank you for pointing out the relatively low success rates in Table 5 and the significant gap compared to Table 3. In response, we have added a detailed explanation in Section 4.4 discussing the causes of this gap, such as differences in environment complexity and limitations of domain adaptation techniques. We also revised the conclusion to highlight this issue and outline directions for future improvements. The revised content has been highlighted in yellow in the manuscript.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

This version shows significant improvement. I believe it’s ready for publication after minor textual edits. Here are a few suggestions to further enhance the manuscript:

1.  In the conclusion and results analysis, consider including navigation accuracy metrics (e.g., in meters or decimeters), as these are critical for the autonomous driving industry. Providing such data would strengthen the paper’s relevance.

2. While the title of this paper focus is on the Navigation System, the current descriptions and discussions lack depth. The connection between localization/navigation concepts and the deep learning methods employed remains underdeveloped. Strengthening this alignment in the introduction, results analysis, and other sections would improve clarity.

3. Where possible, enhance the aesthetics of figures and tables. For instance, inconsistencies in spacing (e.g., between Tables 4, 5 and surrounding text) should be addressed for a polished appearance.

Author Response

Response To Reviewer Comments

 

The authors would like to thank the anonymous reviewers for their careful examination of this paper and the constructive comments they have provided. In this letter, we respond to each of the comments and show how the paper has been modified in response to the reviewers’ recommendations.

1. In the conclusion and results analysis, consider including navigation accuracy metrics (e.g., in meters or decimeters), as these are critical for the autonomous driving industry. Providing such data would strengthen the paper’s relevance.

Response:

    Thank you for your suggestion regarding the inclusion of navigation accuracy metrics. In our current study, the evaluation criteria are based on whether the agent can stably follow the designated path and complete the entire navigation route without deviation, rather than on numerical distance errors (e.g., in meters or decimeters). This is due to the nature of the Duckietown environment, where localization and navigation decisions rely primarily on image inputs and planned trajectory information. Both training and testing are designed to assess task completion and path adherence. Nevertheless, we acknowledge the importance of precise distance-based accuracy in practical autonomous driving applications and will consider incorporating such metrics in future work to improve the system's applicability and comparability.

2. While the title of this paper focus is on the Navigation System, the current descriptions and discussions lack depth. The connection between localization/navigation concepts and the deep learning methods employed remains underdeveloped. Strengthening this alignment in the introduction, results analysis, and other sections would improve clarity.

Response:

    Thank you for pointing out the areas where the discussion of the navigation system could be strengthened. We understand your concern regarding the limited explanation of how navigation concepts are connected to the deep learning methods used in this study. In response, we have expanded relevant sections in the introduction and results analysis to clarify these relationships.

    In this work, the navigation task serves as the application context for evaluating deep reinforcement learning models. The study explores how models trained in a simulated environment can be applied to real-world tasks using a Duckiebot platform. Given this focus, much of the discussion has emphasized the process of transferring learned behavior from simulation to physical implementation. Following your suggestion, we have adjusted the manuscript to better convey the connection between navigation and the learning methods employed, and to improve overall coherence and clarity.

3. Where possible, enhance the aesthetics of figures and tables. For instance, inconsistencies in spacing (e.g., between Tables 4, 5 and surrounding text) should be addressed for a polished appearance.

Response:

        Thank you for your valuable suggestion. We have revised the formatting of figures and tables to improve visual consistency, including adjustments to the spacing around Tables 4 and 5. All relevant changes have been highlighted in yellow in the revised manuscript.

Author Response File: Author Response.pdf

Reviewer 4 Report

Comments and Suggestions for Authors

Dear Authors,

I acknowledge your efforts to reply to my points; however, they are mostly structural errors and not a matter of writing, therefore, they cannot be solved with the edition of multiple versions. Therefore, my opinion is that the paper is not publishable at this journal.

Kind regards

Author Response

Response To Reviewer Comments

The authors would like to thank the anonymous reviewers for their careful examination of this paper and the constructive comments they have provided. In this letter, we respond to each of the comments and show how the paper has been modified in response to the reviewers’ recommendations.

1. I acknowledge your efforts to reply to my points; however, they are mostly structural errors and not a matter of writing, therefore, they cannot be solved with the edition of multiple versions. Therefore, my opinion is that the paper is not publishable at this journal.

Response:

    Thank you for taking the time to review our manuscript and for your valuable feedback. We understand your concerns regarding the structural issues and will thoroughly review and reorganize the relevant sections, including the presentation of the methodology and results, to improve the clarity and logical flow of the paper.

We sincerely appreciate your time and guidance.

 

Author Response File: Author Response.pdf

Round 3

Reviewer 1 Report

Comments and Suggestions for Authors

In section 1, the contributions still need to be clearly stated, for example:
"The main contributions of this paper are as follows:
(1) XXX;
(2) XXX."

 

Author Response

Response To Reviewer Comments

 

The authors would like to thank the anonymous reviewers for their careful examination of this paper and the constructive comments they have provided. In this letter, we respond to each of the comments and show how the paper has been modified in response to the reviewers’ recommendations.

 

1.In section 1, the contributions still need to be clearly stated, for example:
"The main contributions of this paper are as follows:
(1) XXX;
(2) XXX."

Response:

        Thank you for your valuable suggestion. In accordance with your comments, we have clearly added a description of the research contributions at the end of Section 1 (Introduction). We have adopted the format you recommended: “The main contributions of this paper are as follows: (1)... (2)...” to clearly present the key contributions of this study. All modifications made in the manuscript are highlighted with a light blue background.

Author Response File: Author Response.docx

Reviewer 4 Report

Comments and Suggestions for Authors

Dear Authors,

The main reason I rejected this paper twice is not possible to solve with rewriting as it is a matter of structural faults along the study. Therefore, I keep the same opinion: reject.

Kind regards

Author Response

Response To Reviewer Comments

The authors would like to thank the anonymous reviewers for their careful examination of this paper and the constructive comments they have provided. In this letter, we respond to each of the comments and show how the paper has been modified in response to the reviewers’ recommendations.

1.The main reason I rejected this paper twice is not possible to solve with rewriting as it is a matter of structural faults along the study. Therefore, I keep the same opinion: reject.

Response:

        Thank you for taking the time to review our manuscript and provide your valuable comments. We understand your concerns regarding the structural aspects of our study and fully respect your decision based on your professional judgment. We sincerely appreciate your candid feedback and the time you devoted to the review.

Author Response File: Author Response.docx