From Knowledge to Leverage: How to Use Musculoskeletal Simulation to Design Exoskeleton Concepts
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsThank you for this very timely and important article. It would be great if digital human modelling could be used to predict how exoskeletons work and how people react to the mechanical effects of exoskeletons. Unfortunately, the issue is more complex than it first appears. Basically, you address all aspects. However, I think you should be a little more specific about the limitations of this approach. It takes a lot of expertise to interpret the results of such human modelling approaches (I love such approaches) and to draw the right conclusions. Unfortunately, it is not plug and play in the end.
Please forgive me for leaving a few (somewhat critical) comments here and there. I would appreciate your critical feedback so that we can improve the paper for the reader.
Line 113: The criterion is based solely on muscle forces - not joint forces or other passive structures. Please provide more information to clarify how it is used in the optimisation problem (1)-(3).
Figure 2:
- It seems that the box is placed further in front of the body, which increases the L5-S1 force at the end. An additional kinegram would be desirable here to make the whole thing clearer.
- Why don't you show the tibiofemoral joint forces? Especially when the hamstrings are also active?
Line 155: Why don't you look at the hip joints? When lifting boxes, the back is probably under much more strain than the knee extensors.
Line 164: Do you mean biarticular instead of bilateral?
Figure 3 (and Figure 5): Directional arrows would certainly be helpful here. Ultimately, it is somewhat strange that even the large deviations from the trend line in the range below 20° have little influence on the results in Fig. 5. The same applies to the 50° range of the left patella (Fig. 5)!
Figure 4: Please comment on my statement: This only applies precisely to the one movement examined. With every change in movement, the ideal characteristic curve would no longer fit. A control system would therefore be required that could achieve this in real time (very theoretical – probably far from feasible). So, in the end, it does not seem to be of any help to designers.
Figure 6: Please comment on my statement: That would be something to try to implement technically at the end. I have my doubts.
Line 220: Please comment on my statement: What happens if the work/movement situation changes slightly? Could it be that in the deep initial position, the hamstrings need to be active in order to maintain the knee bend? The whole idea seems impractical to me. Multiple joints and only one rod spring.
Line 236/237: Please comment on my statement: It seems to me that the actual disadvantage is being turned into an advantage.
239-240: Please comment on my statement: Yes, the muscle pathways imply this.
267: Please comment on my statement: Well, that basically disproves the concept. Multisegment exos are therefore only possible with regulation and probably only as active systems. So the concept isn't really that convincing.
270: Please comment on my statement: Well, I don't think that the pole system you suggest would ultimately be practical.
Author Response
Dear Reviewer,
Please let me first thank you very much for your constructive approach to this process. I absolutely share your goal of improving the paper for the benefit of the reader. Thank your for your contributions to this end.
Line 113: The criterion is based solely on muscle forces - not joint forces or other passive structures. Please provide more information to clarify how it is used in the optimisation problem (1)-(3).
This is a very insightful comment, and the informed reader might certainly be puzzled about this. It is actually already mentioned in the discussion, but I have added further explanation about this issue to the section explaining the three types of forces in the mathematical program (1)-(3).
Figure 2:
It seems that the box is placed further in front of the body, which increases the L5-S1 force at the end. An additional kinegram would be desirable here to make the whole thing clearer.
Yes, this is definitely an effect of the test subject moving the box further from the spine at the end of the movement. I have added comments about this and also an explanation for the choice of patella-femoral force to the first paragraph in section 3.
Why don't you show the tibiofemoral joint forces? Especially when the hamstrings are also active?
It was also my initial suspicion that tibiofemoral forces might be large, but it turns out that they are not. There are literally thousands of other forces that are also not included in this rather simple example, and I think it would be going to far to explain why each of them is not included in the study.
Line 155: Why don't you look at the hip joints? When lifting boxes, the back is probably under much more strain than the knee extensors.
Yes, exactly, the load from the box acts in front of the knees, so it stresses the hamstrings rather than the quadriceps, and this is one of the reasons why the tibiofemoral force is not overly large. In the interest of focus on the method, I wanted to limit the selected variables to a few, and I think the L5S1 force is more ergonomically relevant than the hip joint force in terms of injury risk. However, the necessity to consider tissue loads more comprehensibly than what is presented here is an important point, and I have therefore added further remarks about this to the discussion.
Line 164: Do you mean biarticular instead of bilateral?
Yes, of course. This has been corrected. Thanks for spotting it.
Figure 3 (and Figure 5): Directional arrows would certainly be helpful here.
I think, for the directional arrows, you are hinting at figures 3 and 8? I have added arrows to both of them indicating the direction of movement.
Ultimately, it is somewhat strange that even the large deviations from the trend line in the range below 20° have little influence on the results in Fig. 5. The same applies to the 50° range of the left patella (Fig. 5)!
Yes, I can see how this could make the reader puzzled. I have added a paragraph to the discussion about this.
Figure 4: Please comment on my statement: This only applies precisely to the one movement examined. With every change in movement, the ideal characteristic curve would no longer fit. A control system would therefore be required that could achieve this in real time (very theoretical – probably far from feasible). So, in the end, it does not seem to be of any help to designers.
I agree in principle. However, in lifting situations, the user is fighting gravity most of the time, and this gives passive exoskeletons that counterbalance gravity a chance to be reasonably effective in a variety of different but related movements. This paper just illustrates the method. In practical use, the design should be simulated in variety of situations including walking. I have added a remark about this to the discussion.
Figure 6: Please comment on my statement: That would be something to try to implement technically at the end. I have my doubts.
I agree, but please remember that the aim of the paper is to present a design methodology. It is not the aim to design a particular exoskeleton. The design in Fig. 6 can be impractical for a number of reasons, some of which are mentioned in the discussion. If I were to further develop the rod exoskeleton, I would probably exploit the opportunity that the rods do not have to be straight. They can curve to follow the body's articulations in the sagittal plane a bit better and thereby take up less space outside the body.
Line 220: Please comment on my statement: What happens if the work/movement situation changes slightly? Could it be that in the deep initial position, the hamstrings need to be active in order to maintain the knee bend? The whole idea seems impractical to me. Multiple joints and only one rod spring.
Yes, that is a very probable problem. I hope the situation is covered by the additions I have made to the discussion.
Line 236/237: Please comment on my statement: It seems to me that the actual disadvantage is being turned into an advantage.
Yes, this goes to the first of your remarks as well, and I think is is now discussed in the additions under the mathematical problem formulation (1)-(3).
239-240: Please comment on my statement: Yes, the muscle pathways imply this.
I am not exactly sure what you mean, but I think you mean that the muscle paths pass closer to the joints than the action line of the external force. If so, I think this is covered by the aforementioned addition.
267: Please comment on my statement: Well, that basically disproves the concept. Multisegment exos are therefore only possible with regulation and probably only as active systems. So the concept isn't really that convincing.
Well, whether the exo will wobble or not during walking, and whether this can be mitigated in the detailed design phase or not is only speculation from my side. I think there may be a problem, but the exo design is not the point of the paper. The point is the simulation-based design approach.
270: Please comment on my statement: Well, I don't think that the pole system you suggest would ultimately be practical.
Possibly not in its current form. Again, the exo not the point of the paper. It is only an example of the kind of ideas that can be generated with the design method.
Thanks again for your insightful comments. I think the resulting changes in the paper has improved its quality a lot.
Reviewer 2 Report
Comments and Suggestions for AuthorsThis article presents a compelling and methodologically rigorous approach to integrating musculoskeletal simulation into the early conceptual design of exoskeletons, offering significant advancements in reducing design iterations and optimizing passive actuator parameters. By leveraging inverse dynamics and mathematical programming within the AnyBody Modeling System, the authors effectively demonstrate how simulation-derived insights—such as ideal spring stiffness and multiarticular rod configurations—can guide practical exoskeleton development. The case study on box lifting provides clear, data-driven validation of the methodology’s ability to offload critical joints (e.g., knees and spine) while maintaining computational transparency through open-access datasets. The work bridges a critical gap between biomechanical theory and industrial application, highlighting the potential for cost-efficient, simulation-aided design processes in wearable robotics. Overall, this research sets a valuable precedent for future studies aiming to enhance exoskeleton functionality while addressing real-world ergonomic and economic constraints. There are several concerns:
1.How is the assumption that human motion remains unchanged with exoskeleton intervention (based on inverse dynamics simulation) validated? Is the validity of the AnyBody Modeling System for spine and knee joint simulations sufficiently supported by peer-reviewed literature?
2.Given the use of a single healthy male subject, how can the results be generalized to diverse populations (e.g., age, gender, body type)?
3.Does the HD5 data on Zenodo include full simulation parameters and raw motion capture data? Can other researchers independently replicate the experimental results?
4.Is the inverse dynamics simulation method limited to passive exoskeletons? For active actuators, what specific control strategies or algorithms are proposed to manage real-time adjustments?
5.Are parameters like the spring stiffness (e.g., 1.38 Nm/degree) and linear trend lines derived from systematic optimization? How is subjective bias avoided in parameter selection?
6.How does the bilateral rod exoskeleton balance task-specific support (e.g., box lifting) with interference in other movements (e.g., walking)? Are ergonomic factors like comfort or added physical burden evaluated?
7. Are the simulation results validated through physical prototypes? If not, how is the predictive accuracy of the model ensured?
8.Can the methodology be extended to dynamic or non-repetitive tasks (e.g., handling irregular objects)? What modifications to the model assumptions would be required?
9.The study claims to "reduce design iteration costs," but no quantitative analysis (e.g., time, budget savings) is provided. Are there real-world case studies or data to support this claim?
Author Response
Thank you very much for your kind words about the paper. Below, please find pointwise responses to your detailed comments.
1.How is the assumption that human motion remains unchanged with exoskeleton intervention (based on inverse dynamics simulation) validated?
This is not validated at all, and I would be very surprised if it were completely true. This limitation is discussed openly in section 4, however, and it is mentioned that it is difficult to mitigate in the initial design phase, where a prototype is not available for testing.
Is the validity of the AnyBody Modeling System for spine and knee joint simulations sufficiently supported by peer-reviewed literature?
You are right: I have neglected to insert references for validation studies for these body parts. These references have now been added to section 1.1.
2.Given the use of a single healthy male subject, how can the results be generalized to diverse populations (e.g., age, gender, body type)?
Please notice that this is not a paper about a new type of exoskeleton. The purpose of the paper is to demonstrate a design methodology. The methodology would have been the same but the resulting exoskeleton possibly different, if we had used a different subject with a different movement pattern. The exoskeleton only serves as an example.
3.Does the HD5 data on Zenodo include full simulation parameters and raw motion capture data? Can other researchers independently replicate the experimental results?
Yes, the hd5 file contains all simulation results including motions of all joints in the model from which the movement can be reconstructed. The hd5 file also contains all analysis results including muscle forces and joint reactions. The hd5 file format is a standard, which can be read by popular programming frameworks such as Python and Matlab.
4.Is the inverse dynamics simulation method limited to passive exoskeletons? For active actuators, what specific control strategies or algorithms are proposed to manage real-time adjustments?
The use of idealized actuator moments/forces as presented in Figures 3 and 8 would require an active exoskeleton. The assumption of a passive exo only comes into the picture when the actuator forces of those figures are approximated by linear regressions. You could say that the Figs. 3 and 8 provide a baseline actuation strategies for active actuators. As far as I know, the simulation does not contribute to understanding dynamic real-time adjustments. I have added a paragraph about these considerations to the discussion.
5.Are parameters like the spring stiffness (e.g., 1.38 Nm/degree) and linear trend lines derived from systematic optimization? How is subjective bias avoided in parameter selection?
These values are the slopes of the linear regressions, so they are completely objective.
6.How does the bilateral rod exoskeleton balance task-specific support (e.g., box lifting) with interference in other movements (e.g., walking)? Are ergonomic factors like comfort or added physical burden evaluated?
They are not, but they certainly should be in any practical use of the method. A paragraph about this has been added to the discussion.
7. Are the simulation results validated through physical prototypes? If not, how is the predictive accuracy of the model ensured?
The results, i.e., the designed exoskeleton, is not validated. The purpose of the paper is to present a design method. The resulting exoskeleton is only for demonstration purposes. Its practical design and production would require several additional steps. This has been added to the discussion c.f. the answer to the preceding comment.
8.Can the methodology be extended to dynamic or non-repetitive tasks (e.g., handling irregular objects)? What modifications to the model assumptions would be required?
This is a good question. The methodology should be exactly the same for a longer sequence of diverse working tasks, but the analysis may reveal that a simple exoskeleton is less useful when the work task is less repetitive. I have added a paragraph to the Discussion to address this issue.
9.The study claims to "reduce design iteration costs," but no quantitative analysis (e.g., time, budget savings) is provided. Are there real-world case studies or data to support this claim?
There are multiple case studies in design science supporting the idea that the majority of the cost is committed in the early stages of the design process. However, design studies tend to be qualitative in nature. The reader is referred to reference 21.
I thank you sincerely for your constructive comments, which have improved the manuscript significantly.
Reviewer 3 Report
Comments and Suggestions for AuthorsThe paper proposes the use of musculoskeletal simulation since the early stages of the design of exoskeletons, instead of using them when the prototype is in a more advanced design stage, where their benefits can be late. The author states that this process can be automated and applied to full body working tasks, if a particular structure in its underlying mathematical formulation is exploited.
My main concern with the paper is that, since the proposal is just presented with an example, the automation of the design process that is pretended is not clear. The objective of the paper would be achieved better if the author presented first his ‘philosophy’ of design with a block diagram or a sequence of tasks to be followed by any designer to achieve the desired goal. And then, apply that process to the example that is given.
I think that the author can vastly improve the significance of the contribution of the paper in this sense, so any reader can apply his methodology to a new design.
Apart from that, I have some other questions and comments:
Line 166: Author says that the motor moments can be ‘reasonably’ approximated by a single linear trend line. What I see is two non-linear functions not easy to approximate with a linear trend. At a knee extension angle of 40º for example, the idealized motor moment can zero or almost 60 Nm, which is quite a difference.
Figure 8: Is there an explanation for the rod force trend going vertical near 1.2 m? If the trend doesn’t cover that part of the motion, what do the author propose to compensate that?
Figure 7 and 9: It would be nice to have the same scale in the Y axis for a better comparison.
Conclusions are short. How are currently done the early design stages of exoskeletons? It seems difficult to quantify the benefit of the proposed methodology, but it would be nice.
Author Response
Dear Reviewer,
Thank you very much for your efforts and contributions to improving the paper. Your input is very much appreciated.
My main concern with the paper is that, since the proposal is just presented with an example, the automation of the design process that is pretended is not clear. The objective of the paper would be achieved better if the author presented first his ‘philosophy’ of design with a block diagram or a sequence of tasks to be followed by any designer to achieve the desired goal. And then, apply that process to the example that is given.
I think that is a good idea to give the reader an initial roadmap for the process, and this was also suggested by another reviewer. It can be realized with a list of steps to take in the process. I have added such a list to the Introduction just before section 1.1.
Line 166: Author says that the motor moments can be ‘reasonably’ approximated by a single linear trend line. What I see is two non-linear functions not easy to approximate with a linear trend. At a knee extension angle of 40º for example, the idealized motor moment can zero or almost 60 Nm, which is quite a difference.
Point taken. Perhaps "reasonably" is not a good word. What happens is a linear regression, which can be done with any data set regardless of how the data are correlated. I have changed the wording to avoid misunderstanding, and a paragraph about the relationship between the quality of the fit and the effect of the exoskeleton has been added to the discussion.
Figure 8: Is there an explanation for the rod force trend going vertical near 1.2 m? If the trend doesn’t cover that part of the motion, what do the author propose to compensate that?
Yes. This is because addition of the idealized forces is costless for the algorithm. This issue is now addressed in the same added paragraph in the Discussion.
Figure 7 and 9: It would be nice to have the same scale in the Y axis for a better comparison.
Good point. This has been done.
Conclusions are short. How are currently done the early design stages of exoskeletons? It seems difficult to quantify the benefit of the proposed methodology, but it would be nice.
I deliberately placed the wordy discussion in the Discussion section in the interest of brevity in the Conclusion. Good comments from you and the other reviewers have added several paragraphs to the Discussion, which is now significantly longer than before. I do not have access to all exoskeleton design labs, so it would not be right for me to state their design processes. I do, however, refer the reader to reference 21 (in the updated version) which is a pointer to the world of design science, and which discusses cost estimations of different approaches.
Thanks again for your valuable contributions.
Reviewer 4 Report
Comments and Suggestions for AuthorsPlease see attached file.
Comments for author File: Comments.pdf
Author Response
Dear Reviewer,
I thank you sincerely for your efforts in reviewing the paper and for making valuable points that have contributed to the quality. I also thank you for your kind words about the manuscript. Please find below responses to the detailed comments.
I recommend removing the subtitle "Featured Application: Using musculoskeletal simulation to obtain conceptual designs for exoskeletons" from the manuscript. Although it may have been intended to highlight the practical relevance of the work, its placement directly below the article title may confuse readers and create ambiguity about the actual title of the paper.
I completely agree. I believe that the "Featured application" subheading was a part of the template, but I have now removed it, and we shall see whether the editor requires it.
I suggest revising the structure of the abstract by removing the numbered labels (1) Background; (2) Methods; (3) Results; (4) Conclusions). While this structure can be helpful in certain contexts, the content of your abstract is already clearly organized and understandable without these headings. Removing the numbered format would improve the flow and readability, making the abstract more aligned with standard practices in scientific publishing.
Again, I agree, but with the same comment as above: I think this is the standard journal template.
To better align with the journal’s formatting standards and general scientific writing conventions, I suggest removing the subsection title “Simulation-Based Design” from the introduction. Instead, consider integrating a brief sentence at the beginning of the paragraph to indicate that it discusses simulation-based design. This will help maintain a smoother flow within the introduction while preserving the clarity of the content.
I agree. It seems silly to have section 1.1 when there is no section 1.2. The subheading has been removed.
To improve the readability and structure of the article, consider adding a brief paragraph at the end of the introduction that outlines the organization of the remaining sections. This overview will help readers follow the logical flow of the paper and anticipate its key content. For example, you could mention that Section 2 presents the Mathematical method. The subsequent sections can be introduced analogously, summarizing their primary focus.
One of the other reviewers suggested almost the same, namely addition of an outline of the approach, which would make it easier to follow the detailed subsequent derivations. This has been done.
Adjust the figures, including their respective captions, to align with the journal’s standards (e.g., by centering them).
The journal template provides predefined styles to use for each text element, and those have been used also for the figures. I trust that the editor will re-arrange to the journal standard in case there is some error.
On line Equation (3): Replace "i = 1..nm" with "i = 1 . . . nm".
Done. Thanks for spotting this.
Thank you very much once again for your help and kind words.
Round 2
Reviewer 1 Report
Comments and Suggestions for AuthorsThank you very much for responding to my comments. I think the article has turned out really well. Thanks again!
Reviewer 3 Report
Comments and Suggestions for AuthorsI think that the author have addressed the proposed changes and the paper has been considerably improved. I recommend it for publication.