Learning Human–Robot Proxemics Models from Experimental Data

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors In this paper, authors model human proxemics as robot navigation costs and distinguish between avoidance and interaction models depending on whether human-robot engagement is involved. The proxemic avoidance model enhances robot navigation by incorporating human-aware behaviors, treating humans not as mere obstacles but as social agents with personal space preferences. The model of interaction positions estimates suitable locations relative to the target person for the robot to approach when an engagement occurs. Here are my comments:
1)The contributions of this paper should be clarified in the introduction, and the major difference between the current work and existing ones should be delineated. All the models in Section 3.2 have been previously developed, and the innovation of this work is not clearly defined.
2)The method has only been tested using data from the public dataset. Nonetheless, it also requires validation through experiments on a platform featuring both robots and humans.
3)Since no comparison results are provided, it is unclear what the explicit advantages of the proposed modeling method are.
4)A diagram of the modeling process is suggested.
5)There are numerous errors. Even the funding section remains unchanged.  

Author Response

Thank you very much for taking the time to review this manuscript. Please find the detailed responses in the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

This paper presents a machine learning approach to modeling human–robot proxemics for socially aware navigation. Using the CongreG8 dataset, the authors develop an interaction model that estimates socially appropriate approach positions. The proposed models are compared with an existing hand-crafted asymmetric Gaussian model, showing improved interpretability of proxemic costs in close-range navigation scenarios. The work presents a learnable and extendable framework for integrating proxemic behavior into robot navigation. 

While the paper notes cultural differences in proxemic perception, in practice, many other factors require consideration, such as whether the robot has a vehicle-like appearance (e.g., a delivery robot) or a humanoid form, as well as its size, speed, and other physical attributes. The authors should clearly state the real-world scenarios they aim to address and review related work accordingly. Furthermore, if the proposed system can handle such variations, this should be explicitly argued. This issue is particularly relevant to the persuasiveness of the dataset choice: a thorough discussion is needed on whether the selected dataset is well-suited to the situations the authors seek to solve. 

Although the authors present this as an “avoidance model,” it is a static model of proxemic positioning rather than a dynamic avoidance model. Its primary function is closer to approaching conversational groups than to avoiding people in general. In the broader literature, “avoidance model” typically implies actively avoiding people. The current terminology may therefore be misleading, and a revision of the claim is recommended to prevent misunderstanding. 

If the model is to be used for dynamic navigation, it would need to incorporate factors such as the relative distance, speed, and predicted position of the other party. As it stands, directly applying the current model to navigation may not be enough. 

The discussion of results for the proposed model is rather limited. I suggest implementing a simulation using this model and conducting a user study to collect subjective likeability ratings from participants.  

Author Response

Thank you very much for taking the time to review this manuscript. Please find the detailed responses in the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

I have no further comments. 

Author Response

We sincerely appreciate your review and your recognition of our work. Thank you.

Reviewer 2 Report

Comments and Suggestions for Authors

I appreciate the authors’ efforts in revising the manuscript. The topic of modeling proxemics for socially aware robot navigation is important and timely, and the revised version shows progress. Nevertheless, there remain issues that need to be clarified and improved before the manuscript can be published.

-Abstract and Experimental Setting
The abstract states that the proposed models can guide the robot’s avoidance and approaching behaviors toward humans. However, the experimental setting is not described in the abstract, and the key results are also missing. Without this information, the claims appear overstated. Please add a concise description of the evaluation setting and main findings to support the statement.

-Necessity and Validity of Focusing on Three-Person Groups
The study relies on the CongreG8 dataset, which centers on triadic conversational groups. The manuscript does not clearly explain why a three-person group was chosen as the primary evaluation scenario. It would be important to clarify (a) why this particular group size is necessary for the study, and (b) to what extent the results obtained in this setting can be generalized to other situations, such as dyadic or larger group interactions. Without such discussion, the validity as evidence for the model’s effectiveness remains uncertain.

-

Introduction and Positioning of the Work
In the Introduction, the authors emphasize that their framework can be fine-tuned across robot platforms. However, the importance of the experimental setting itself is not discussed. Even if the framework is adaptable, the results depend on whether the chosen scenario is appropriate for the hypotheses. Please provide more explanation on why the chosen setting is essential to validate the proposed model, preferably with reference to prior studies on robot approaches to groups.

-Figures and Readability
In many figures, the text labels and vectors are very small, and the reader must rely heavily on captions to understand the content. This reduces readability. I strongly recommend revising the figures so that the main points can be understood more directly—for example, by enlarging the labels and simplifying the presentation.

 

Comments on the Quality of English Language

None in particular.

Author Response

Thank you very much for your valuable comments. In line with your suggestion, we added a more detailed description of the experimental setup, and our experiments cover scenarios with a single person, dyad, and triad. Please find the detailed responses below and the corresponding revisions in the re-submitted file.

Author Response File: Author Response.pdf