Gestures in Motion: Exploring Referent-Free Elicitation Method for Hexapod Robot Control in Urban Environments

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

1 The manuscript states that gesture data from Phase 1 were manually clustered into 37 categories by three researchers and further reduced to 21 representative gestures for ranking in Phase 2. However, no details are provided regarding the coding protocol, criteria for determining similarity, or the process used to resolve disagreements among coders.

2 While the goal of implementing a real-time gesture recognition prototype on embedded hardware is commendable, the system’s actual performance metrics do not justify the claims made. A test accuracy of 30.75% for a 24-class classification task is effectively at or near chance level. Despite this, the discussion section presents the system as responsive and viable for embedded deployment, which risks misleading readers.

3 Table 1 presents average agreement rates across gesture elicitation conditions, suggesting that referent-free methods yielded higher agreement overall. However, no statistical analysis is provided to determine whether these differences are significant.

Author Response

Dear Reviewer,

We thank for the detailed assessment of the manuscript. We agree with all points raised, and corresponding changes have been implemented in the manuscript. Below, we provide detailed responses to each comment.

Comment 1:
“The manuscript states that gesture data from Phase 1 were manually clustered into 37 categories by three researchers and further reduced to 21 representative gestures for ranking in Phase 2. However, no details are provided regarding the coding protocol, criteria for determining similarity, or the process used to resolve disagreements among coders.”

Response 1:
We have expanded the Methods section to describe the clustering process in detail. Gestures were grouped by syntactic similarity (hand shape, direction, trajectory, amplitude, repetition, and spatial configuration). Three researchers independently coded the data, and disagreements were resolved through discussion until consensus was reached. This clarification is now included in Section 3.1.

Comment 2:
“While the goal of implementing a real-time gesture recognition prototype on embedded hardware is commendable, the system’s actual performance metrics do not justify the claims made. A test accuracy of 30.75% for a 24-class classification task is effectively at or near chance level. Despite this, the discussion section presents the system as responsive and viable for embedded deployment, which risks misleading readers.”

Response 2:
We revised the Discussion and Conclusion to soften claims and present the prototype as an early feasibility demonstration rather than a viable system. We clarify that only three visually distinct gestures were deployed live, while the overall 24-class accuracy remained low. The text now emphasizes that the implementation validates feasibility, not full performance.

Comment 3:
“Table 1 presents average agreement rates across gesture elicitation conditions, suggesting that referent-free methods yielded higher agreement overall. However, no statistical analysis is provided to determine whether these differences are significant.”

Response 3:
We have added a paired-sample t-test for overlapping tasks between referent-free and referent-based conditions. Results showed a significant difference in agreement rates, with referent-free gestures yielding higher consensus. We note the small sample size and advise cautious interpretation. This statistical analysis has been added to Section 4.3.

We believe these revisions address the reviewer’s concerns and improve the methodological transparency, accuracy of claims, and statistical rigour of the manuscript.

Best regards,  
Natalia Walczak  
on behalf of all authors

Reviewer 2 Report

Comments and Suggestions for Authors

I would like to congratulate the authors on a well-prepared manuscript. The paper is clearly structured, methodologically sound, and addresses a topic that will undoubtedly be of interest to readers. I do not find any suggestions for improvement, as the work is already suitable for publication in its current form.

Author Response

Dear Reviewer,

We thank for the detailed assessment of the manuscript. In response to the comment received:

Comment 1:
I would like to congratulate the authors on a well-prepared manuscript. The paper is clearly structured, methodologically sound, and addresses a topic that will undoubtedly be of interest to readers. I do not find any suggestions for improvement, as the work is already suitable for publication in its current form.

Response 1:
We sincerely thank the reviewer for the positive and encouraging feedback on our manuscript. We are glad that the structure, methodology, and contributions of the paper were found to be clear and relevant to the field. While no specific revisions were requested, we have nevertheless carefully reviewed the full text in light of the other reviewers’ comments and have made targeted improvements in methodology clarification, statistical reporting, and figure descriptions. We hope that these refinements further strengthen the paper and maintain the high quality that the Reviewer has recognized.

Best regards,  
Natalia Walczak  
on behalf of all authors

Reviewer 3 Report

Comments and Suggestions for Authors

The article is worthy of publication with a few minor additions.

The article presents a very interesting and important research into a way people will communicate with robots using gestures, and more specifically the way we could and/or should elicit those gestures and some related issues such as human attitudes and measure of agreement.

The research is clearly a very important one and the authors should be lauded for such a survey and their integrative approach with quite a few suggestions and facts found in the surveys and experiments.

My minor suggestions would be to add (and in some cases just clarify or detail) some aspects that would make the article even clearer to understand and to use its helpful findings and suggestions.

In general, if the authors had done more statistical research and processing, it would make the approach and findings even more convincing. If there is no such statistics, then the existing numbers and their analysis could be a little more expounded and explained.

In terms of structure of the article, the authors use "referent" quite frequently bu "referent' definitions and description in the context of the research should be given as soon as possible, and not only on line 102 after multiple uses and still is not tied to the context context, but a general description is given. A few more lines and example at much earlier a stage would improve the structure and readability of a wider audience of readers. 

The survey and experiment particulars could be detailed. For instance, Phase 1 (line 157) could be described in more details like "users were asked the question:"or  "users wrote  or performed ... while explaining ..." or other description.

More details could be given on the practical real life aspect. What was the robot role? Did it react and if so how? The mention on lines 350-360 could be less vague. Also, what exactly is shown in the pictures of users interacting with the robot and model accuracy graphs on those lines could be detailed.

The authors could give a little more of their view, their explanation and meaning of results. Proposed uses can be elaborate upon.

The dialogues between experimenters and participants and participants and robot, especially some detail,s maybe pics or graphics, of at least main gestures could be very helpful.

 

Author Response

Dear Reviewer,

We thank for the detailed assessment of the manuscript. We agree with all points raised, and corresponding changes have been implemented in the manuscript. Below, we provide detailed responses to each comment.

Comment 1:
“In general, if the authors had done more statistical research and processing, it would make the approach and findings even more convincing. If there is no such statistics, then the existing numbers and their analysis could be a little more expounded and explained.”

Response 1:
We have added a statistical test (paired-sample t-test) comparing agreement rates between referent-free and referent-based conditions (Section 4.3). In addition, we expanded the Results section to provide more detailed interpretation of agreement rates and the training/accuracy plots.

 

Comment 2:
“In terms of structure of the article, the authors use "referent" quite frequently but "referent' definitions and description in the context of the research should be given as soon as possible, and not only on line 102 after multiple uses and still is not tied to the context context, but a general description is given. A few more lines and example at much earlier a stage would improve the structure and readability of a wider audience of readers. ”

Response 2:
We revised the Introduction to include an explicit definition of “referent” (predefined robot functions such as “move forward” or “stop”) at the first mention, ensuring clarity for the reader.

Comment 3:
“The survey and experiment particulars could be detailed. For instance, Phase 1 (line 157) could be described in more details like "users were asked the question:"or  "users wrote  or performed ... while explaining ..." or other description.”

Response 3:
We expanded Section 3.1 to clarify Phase 1 procedures: participants were asked to create 20 gestures for a mobile robot in an urban context, with the robot present but inactive. The think-aloud method and the post-hoc use of floor rulers for proxemic analysis are now described in detail.

Comment 4:
“More details could be given on the practical real life aspect. What was the robot role? Did it react and if so how? The mention on lines 350-360 could be less vague. Also, what exactly is shown in the pictures of users interacting with the robot and model accuracy graphs on those lines could be detailed.”

Response 4:
We clarified in Section 3.1 that the robot was inactive during Phase 1 and controlled by the researcher in Phase 3 to visualize referents. This distinction is now also linked to Figure 1.

Comment 5:
“What exactly is shown in the pictures of users interacting with the robot and model accuracy graphs on those lines could be detailed.”

Response 5:
We expanded the text around Figures 1–2 to explain the panels: the top panel shows the lab setup, the bottom panel the recognition interface with overlaid labels, and Figure 2 the CNN training curves. Each figure is now explicitly tied to its role in the study.

Comment 6:
“The authors could give a little more of their view, their explanation and meaning of results. Proposed uses can be elaborate upon.”

Response 6:
We added to the Discussion and Conclusion our interpretation of the tension between user creativity and standardization, and elaborated potential applications (delivery, navigation, and assistance for children, older adults, and people with disabilities).

Comment 7:
“The dialogues between experimenters and participants and participants and robot, especially some details, maybe pics or graphics, of at least main gestures could be very helpful.”

Response 7:
The appendix table containing images of the elicited gestures is now referenced in the Results section. While transcripts of dialogues were not included to keep the paper concise, participant feedback was integrated into the analysis.

We believe these revisions address the reviewer’s suggestions and improve both clarity and practical relevance of the manuscript.

Best regards,  
Natalia Walczak  
on behalf of all authors

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

accept