Improving VR Welding Simulator Tracking Accuracy Through IMU-SLAM Fusion

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The manuscript presents a well-structured and technically rigorous study, combining SLAM, IMU modeling, EKF fusion, and domain-specific motion constraints. The experimental evaluation is thorough and includes statistical tests, error analysis, and performance profiling. However, there are few improvement are needed as below:

Minor Correction:

1. The abstract provides a clear overview of the problem but the overall abstract is too lengthy, you should control your abstract in 200 – 250 words.
2. The abstract includes too many specialized terms like “Extended Kalman Filter,” “ORB-SLAM3,” etc. These can be briefly mentioned but not elaborated in depth.
3. Section 3 contains extensive derivations may overwhelm readers and distract from the main contributions. Suggest move detailed derivations and matrices to an appendix
4. The introduction repeats information about VR tracking errors (3–10 mm), cost limitations of external systems and welding environment challenges. Try Condense these discussions to improve readability.
5. Author claims that is it cost-effective, but the experiment it relies on high-end GPUs (RTX 3080), multithreading, and precise VR hardware calibration. Please clarify whether the method can run on affordable hardware (Quest, Pico, integrated GPUs). What is the minimum requirement for the hardware for this method.

Major Correction

1. The authors evaluate the system exclusively in VR simulator environments supported by OptiTrack ground truth. No experiments include real welding torches, real metal workpieces, and etc. This omission undermines the manuscript’s claim that the method is applicable to VR welding training systems because real welding environments pose unique challenges. Suggest to include at least a small-scale pilot test in a real welding environment or clearly state this limitation and moderate claims regarding real-world applicability.
2. Wrong references. Reference [18] is used multiple times for unrelated statements. While some references numbering is missing eg. Reference [19] – [38].
3. Novelty statement is not discussed clear. Please clearly state why existing VR tracking systems cannot address welding-specific dynamics.

 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

- starting from Section 2.4 up to 2.6, all the references are cited to [18]. I think this is a mistake. Authors need to double check such small details before submission.
- Currently, there are too many equations (total 42?) making it hard for me as a reader to follow the logic. The authors do not need to show the equations related to previous SLAM and sensor-fusion literature, just doing proper citation is enough and the readers can direct themselves to that corresponding paper. The authors should only show in the main paper the equation that is relevant to their core contribution. Additionally, derivation of these equations can be moved to the appendix as it is not core to the paper's value but useful for readers who want to know how they were formulated.
- The use of space is not good. There are too many to list down, but I list two as examples: 1) there is too much whitespace before Algorithm 1. 2) There is one page that only contains figures (Fig. 2 and 3).
- Starting from page 8 up until before Section 7 Conclusion, there is a lack of explanation that should be present for academic papers. There are too many equations that it is difficult to understand what is its connection to the overall paper. Algorithm details are shown in Section 4.2 but no explanation to its logic. "Table 1 summarizes the theoretical computational complexity of each module", but no explanation to its implications. Same with Tables 2-4. The totality of Section 6 just shows the raw data and figures of the results without interpreting what these results mean. There is no flow or synthesis or even the minimal commentary that can be  found in these Sections, making the Conclusion feel like a sudden jump.
- As the 3rd core contribution, the authors claimed "comprehensive validation", but I am a bit doubtful about this. The authors described they did a total of 15 datasets across different types of operators (novice, intermediate, expert) and different weld types. However, there is no description about how this dataset was collected, what kind of welding session and how difficult it is, and other important details that are needed to be described for a user study experiment. There are also no description about the environment factors such as the lighting, possible occlusion variations, etc. Comprehensive seems a bit overstated, and it is very difficult to judge whether their results can really be generalized or not.

Overall, I would recommend a major revision of this manuscript. The research seems to  technically sound with moderate novelty, however how the paper is presented is clearly lacking. The biggest weakness of the current manuscript is adding more details to the authors insights in between Sections 4 to 7, fix how tables/figures/equations/whitespaces are presented, and a better narrative flow to explain the story of their research. If the authors address each points above, and more importantly improve the readability of the manuscript, it could have an argument for acceptance.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

The paper presents an IMU-SLAM fusion approach to improve the tracking accuracy of VR-based welding simulators using only standard VR hardware. The authors integrate IMU data with ORB-SLAM3 through an Extended Kalman Filter (EKF) and propose a novel drift correction technique exploiting the periodic weaving motion (3–7 Hz) characteristic of welding torches. Experimental validation using 15 datasets and OptiTrack ground truth demonstrates an average RMSE of 3.8 mm, showing a significant improvement over commercial VR trackers such as SteamVR and Oculus Insight. The proposed system also maintains latency below 100 ms, meeting real-time feedback requirements. The paper is well organized and addresses a relevant problem for VR-based industrial training systems. The proposed method appears technically sound and validated by quantitative experiments. I just provide some remarks:

- The paper compares primarily against SteamVR, Oculus Insight, and standard VI-SLAM baselines. Including more recent learning-based fusion approaches (e.g., LSTM/Transformer-based IMU drift correction) could provide a broader perspective.

- Although the study references the challenges of arc brightness and feature-poor surfaces, it is not clear if real welding light or reflective surfaces were simulated in the experiments. Clarifying this is important for assessing robustness in true operational settings.

- Broaden the discussion on potential adaptation to multi-tool or multi-user scenarios.

- I suggest to improve Figure 1 readibility.

- Authors contribution is not reported correctly.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

- as one of the authors' contribution, they stated that they provide the implementation as open-source, but there is no reference of it anywhere on the paper, so I am confused. This is not a double-blind review so I cannot understand the reason why this is hidden.

- all the formatting problems (such as the white space, wrong citations, messy equations, etc.) are now fixed.

- the authors improved their wording to make their novelty stronger, connect related work to theirs, improved limitations and future work, and stronger conclusion.

- the authors also improved and strengthened their experimental setup and results section.

Overall, the paper has been majorly improved, so I can recommend this for minor revision. However there are two points that I want to point out. First, this is not a double-blind review, making me skeptical of contribution number 5. Either remove that contribution, or share the information where the readers can access that open-source claim. Second, when writing a major revision paper it is very difficult to keep track of the changes between the previous and the current version of the paper. A common courtesy to reviewers is highlight the added text as blue text, and strikethrough the deleted ones. This is very easy to do with some Latex package. I  hope this will be a learning point for the authors moving forward. FYI minor revision is the first point about the open-source.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf