Robotic Simulation Systems and Intelligent Offline Teaching for Urban Rail Transit Maintenance
Round 1
Reviewer 1 Report
Comments and Suggestions for Authors1.
It is clear the integration of Gazebo and Isaac Sim has potential, but you need to explain more clearly how your system is different from or better than other existing simulation or offline teaching methods. Hence, ask to have a table comparing key aspects like fidelity, automation, and performance.
2.
You acknowledge the difference between simulation and real-world performance, but the explanation is a bit light. Since this is a big issue in robotics, especially in critical systems like rail maintenance, it would necessary to go deeper into what causes these discrepancies and how your system handles them.
3.
The 84% success rate is interesting but based on only 44 targets teaching. That is not enough to support strong claims about generalization or scalability. Consider running more tests with a wider range of tasks and show some analysis on why the failures happened and how repeatable your success is.
4.
There are several places where you describe how the system works but you didn't show how well it performs. For instance, how fast does it run? How much memory does it use? What is the frame rate under load? Including some basic performance benchmarks is necessary to validate your claims about efficiency and real-world viability.
5.
It seems as in many setup, like correcting mesh normals, configuring robot joints, and annotating data, is still manual. You should clarify how much of your workflow is automated and what steps require expert input. This is important for understanding how scalable and practical the system is in larger deployments.
6.
Your future work section points toward advanced goals, such as digital twins and real-time rail management, but your current results don’t yet support those ambitions. It might be better to tone those claims down or explain more specifically how your system is going to serve toward those goals.
Comments on the Quality of English LanguageThe writing is mostly clear but frequent grammatical errors, redundancy, and complex phrasing affect readability. Some sentences are very long or technically dense without sufficient transition.
Author Response
Dear Reviewer, Thank you so much for your time and valuable feedback. We've made the requested changes and attached the revised manuscript for your review（Please see the attachment）. Your expertise and guidance are greatly appreciated. Best regards,
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for Authors- Clearly compare the differentiated advantages of Gazebo and Isaac Sim in the abstract, and quantify the specific technical means to improve offline teaching efficiency?
- To supplement the specific problems existing in current robot offline teaching and clarify the necessity of this research?
- How to combine Gazebo lightweight modeling with Isaac Sim high-precision rendering to avoid using two separate systems for isolated description?
- Add a curve graph of the influence of different parameters on path planning time in “2 Experimental Validation”to support the conclusion that “5cm resolution is optimal”?
- Add calibration tools and adjust thresholds in the “manual calibration strategy”to enhance operability?
- Refine “high-precision scene modeling”into “dynamic reconstruction technology based on neural radiation field” and clarify the key modules of the digital twin system?
- Add quantitative comparison data between Gazebo and Isaac Sim in terms of simulation efficiency and modeling accuracy in the conclusion?
- Comparing the costs of offline and on-site teaching to support the economic argument of “increasing efficiency by 15 times”?
Author Response
Dear Reviewer, Thank you so much for your time and valuable feedback. We've made the requested changes and attached the revised manuscript for your review（Please see the attachment）. Your expertise and guidance are greatly appreciated. Best regards,
Author Response File: Author Response.pdf
Reviewer 3 Report
Comments and Suggestions for AuthorsReview Report
"Robotic Simulation Systems and Intelligent Offline Teaching for Urban Rail Transit Maintenance"
The study addresses an important issue in urban rail transit maintenance: the inefficiency and safety risks of manual inspections. The suggested robotic simulation and offline teaching framework provides a unique solution with obvious practical applications.
The combination of Gazebo (for lightweight simulation) and Isaac Sim (for high-fidelity modeling/AI training) offers a balanced approach that takes advantage of the strengths of both platforms.
The methodology is well-organized and includes data collecting (multi-source point cloud fusion), modeling (reverse engineering), simulation (Gazebo/Isaac Sim), and validation (field tests).
The use of Trimble X7 scanners to generate high-precision point clouds, as well as the creation of a unique annotation platform, demonstrate technological depth.
The study coincides with current trends in digital twins and AI-driven automation, and it lays the groundwork for future rail transit digitization.
The study focuses on a particular robot model and metro undercarriage geometry. Variability in rail car designs (e.g., different manufacturers, aged infrastructure) may necessitate adaptive modeling, which has not been widely explored.
The 84% initial success rate indicates a reliance on manual corrections, implying unsolved issues in automatic error compensation (e.g., sensor differences, modeling flaws).
Isaac Sim has high hardware requirements (for example, an NVIDIA RTX 4090 GPU), which may limit accessibility for smaller train operators.
The paper glosses over real-time synchronization concerns between simulation and physical robots (such as lag in ROS-Isaac Sim communication).
A comparative analysis with existing offline training tools (such as ROS-Industrial and ABB RobotStudio) is missing. A benchmark would improve the argument for the suggested framework.
The qualifying percentage of 84% is based on a small-scale test (44 targets). Larger datasets (say, 10,000+ points) may reveal scalability limitations.
While safety risks (such as tight places) are acknowledged, there is no consideration of fail-safes for real-world deployment (such as dealing with robot accidents or sensor failures).
To assess robustness, apply the framework to a variety of metro types and situations (for example, outdoor tracks and tunnels).
Determine the computational load of the simulation systems (e.g., Gazebo vs. Isaac Sim) to help guide hardware selection.
Propose AI-based error correction (for example, reinforcement learning for pose corrections) to eliminate the need for manual changes.
Investigate adaptive modeling strategies (e.g., real-time point cloud updates) for handling structural variations in rail vehicles.
Release the annotation platform code and simulation datasets as open-source resources to encourage community adoption.
Create step-by-step procedures for integrating ROS, Gazebo, and Isaac Sim to reduce access barriers.
The "high-precision scene modeling" and "digital twin" objectives are broad. Prioritize certain milestones (such as real-time LiDAR-to-model alignment).
This research describes a substantial progress in robotic maintenance for rail transit by combining rigorous simulation methods with practical validation. While the results are encouraging, there are still areas for improvement in terms of application and automation. With few tweaks, this work could become a benchmark for intelligent infrastructure maintenance.
Strong technical contribution with potential for scalability and comparison analysis.
Sincerely,
Reviewer
Author Response
Dear Reviewer, Thank you so much for your time and valuable feedback. We've made the requested changes and attached the revised manuscript for your review（Please see the attachment）. Your expertise and guidance are greatly appreciated. Best regards,
Author Response File: Author Response.pdf
Reviewer 4 Report
Comments and Suggestions for AuthorsYou should mention the deep learning algorithm.
4.1. Simulation Scenario Construction 523
In the previous chapter (needs to be removed "chapter word")
Illustrations of the robot arm's trajectory are required.
The end-effector's fixed camera requires additional details.
One of the most important points is the collision-free operation for the robot arm and UGV. You should mention the sensors and the techniques for avoiding collision for the robot arm and UGV.
How was the registration accuracy of the point cloud data quantitatively evaluated? Were any benchmarks or ground truth models used?
What preprocessing or mesh-cleaning techniques (if any) were applied to the point cloud-derived models to address deformation and missing color information?
What metrics or methods were used to assess the color fidelity of the point cloud data? Is this subjective or quantitatively validated?
What criteria led to selecting the Trimble X7 over other scanners? Were cost, accuracy, and scanning speed benchmarked against alternatives?
For your robot, using a 3D scanner lidar is critical for its application; since you mentioned that it is not used, it would be beneficial to add justification or recommendations for this decision.
Author Response
Dear Reviewer, Thank you so much for your time and valuable feedback. We've made the requested changes and attached the revised manuscript for your review（Please see the attachment）. Your expertise and guidance are greatly appreciated. Best regards,
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
Comments and Suggestions for Authors- The paper lacks a well-defined research design or methodology. There is no clear articulation of the research objectives, hypotheses, or experimental setup. Without this foundation, it becomes challenging to evaluate the scientific rigor and replicability of the proposed system. The authors should clearly define these details.
- Although the paper discusses simulation and offline teaching capabilities, it fails to provide empirical results or performance benchmarks. For a system designed to improve urban rail maintenance, key performance indicators such as simulation accuracy, robot response time, training success rates, or usability metrics should be reported. The absence of such data weakens the claims of effectiveness and limits the contribution to the field.
- Much of the technical content in the manuscript remains at a high level without delving into implementation details. For instance, the paper lacks detailed information about how the digital twin interfaces with simulation engines, what robotic control layers are involved, or whether the platform integrates with real hardware or Programmable Logic Controllers (PLCs).
- The cited references are relatively sparse and do not reflect recent research developments in robotics simulation, intelligent tutoring systems, or digital twins in transportation maintenance. Moreover, many citations are either broad or tangential, missing opportunities to ground the proposed system in the current body of knowledge.
- The conclusions claim that the system is effective, scalable, and applicable to diverse scenarios, yet these claims are not substantiated with evidence in the paper. Practical conclusions should directly reflect the results presented and discuss implications, limitations, and future directions based on the actual evaluation.
- Several figures in the paper are underdeveloped and need to be enhanced.
The English could be improved to convey the research more clearly.
Author Response
Dear Reviewer,
Thank you for your review and the valuable comments. Your professionalism and attention to detail during the review process are highly appreciated and have been instrumental in helping us enhance the quality of our manuscript.
We have carefully examined each of your comments and made corresponding revisions to the manuscript. To facilitate your review, all modifications have been highlighted in the PDF version of the manuscript. Please see the attachment.
Your dedication and expertise have been crucial in elevating the quality of our research. We are truly grateful for your time and input!
Best regards,
Author Response File: Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for AuthorsIt can be accepted.
Author Response
Dear Reviewer,
thank you for your professional and meticulous reviewing.
Best regards,
Reviewer 3 Report
Comments and Suggestions for AuthorsDear authors,
Thank you for incorporating the suggestions/remarks in the revised manuscript. I can recommend it for publication in Electronics.
SSincerely,
Reviewer
Author Response
Dear Reviewer,
thank you for your professional and meticulous reviewing.
Best regards,
Round 3
Reviewer 1 Report
Comments and Suggestions for Authors- While the introduction situates the work within Industry 4.0/5.0 and mentions existing simulation tools (Gazebo, Isaac Sim), it should more precisely highlight what is truly contribute compared to the latest studies on digital twins and offline teaching for rail maintenance. I recommend explicitly contrasting your pipeline (the combined Gazebo → Isaac Sim workflow for full undercarriage modeling) with the most recent MDPI and IEEE articles on robotic metro inspection (e.g., Sensors 2024, IEEE T-ITS papers). A focused “gap statement” paragraph that names those works, explains their limitations (e.g., no high-fidelity model for an entire six-carriage train), and then states exactly how your framework fills that gap will help readers understand the unique contribution from the outset.
- Several core modules (URDF and XACRO conversion, Isaac Sim robot import, MoveIt configuration details) lack critical numeric parameters. For instance, in the Robot Model Design (Section 3.5), please supply a sample URDF/XACRO snippet showing joint axes, limits, drive strengths, and inertia values used in Gazebo. Similarly, in Isaac Sim (Section 4), specify the exact joint controller gains, material properties (e.g., PBR vs. Phong settings), and lighting/RTX parameters so others can reproduce your high-fidelity rendering. For the Safety Motion module (Section 3.6), include which MoveIt planner (e.g., RRTConnect or PRM), planning time limit, and collision avoidance thresholds were employed.
- In Section 5.2, the manuscript reports that programming each manipulator action in simulation takes “about 30 seconds” compared to “5–6 minutes” on site, and later notes success rates for different target types (e.g., 84 % for RGB targets, 95 % for infrared) . However, it does not provide any numerical measures of how closely the offline‐taught poses match real‐world results, nor does it quantify registration errors from the point‐cloud modules. To strengthen this section, the authors should include any actual metrics they have—such as positional or orientation deviations observed when executing offline‐taught commands on the real robot (e.g., average millimeter or degree offsets), and the RMSE from point‐cloud alignment—rather than hypothetical values. If exact pose‐error numbers were not collected, the paper should explicitly state that only success rates and timing were measured; otherwise, a concise table comparing simulation success rates, on‐site success rates, and any measured pose offsets (mean ± std) across representative inspection points would clarify how well the simulated workflow translates to real‐world accuracy.
- Several figures lack essential labels and scale references. For example, Figure 19 (OctoMap visualizations at different resolutions) shows two panels—(a) 0.01 m and (b) 0.05 m—but neither includes axis labels (e.g., “X (m)”, “Y (m)”, “Z (m)”) or a scale bar indicating real-world dimensions (p. 21). Similarly, Figure 20 (“The locally optimized 3D OctoMap data is loaded into the MoveIt controller”) depicts the OctoMap within the planning scene but does not indicate voxel size visually or annotate the robot’s position in meters (p. 22). Also, check other figures too.
The language issues are all minor (typos, missing spaces, and a few awkward phrases) and can be resolved with a focused copy-edit rather than major rewriting.
Author Response
Dear Reviewer,
I hope this message finds you well. I am writing to express my heartfelt gratitude for your third round of thoughtful and detailed feedback on our manuscript. Your professionalism and dedication to thorough review are truly admirable, and the care you have put into your comments is evident.
Your expertise has been instrumental in shaping our work, and your attentiveness to detail has been both humbling and inspiring. Each of your comments has provided us with significant guidance, helping us to refine our research and strengthen our paper. The depth of your insights has not only highlighted areas for improvement but has also challenged us to think critically about our approach and findings.
I want to assure you that we have taken your feedback to heart and have made every effort to address your concerns comprehensively. Your suggestions have led us to re-evaluate our methodology, expand our analysis, and clarify our arguments, all of which have contributed to a substantial enhancement of our manuscript.
In response to your latest comments, we have meticulously revised our paper, focusing on the specific areas you have highlighted. We have also prepared a detailed point-by-point response that outlines the changes we have made and the reasons behind them. This response will be included in the DPF (Document Production File) to ensure transparency and to demonstrate how we have incorporated your feedback into our final submission.
Once again, I extend my sincere thanks for your support and guidance throughout this process. Your commitment to scholarly excellence has been a valuable learning experience for me, and I am genuinely grateful for the opportunity to improve our work based on your insightful suggestions.
Warm regards,
Changhao Sun
Author Response File: Author Response.pdf
