NARX Neural Network for Safe Human–Robot Collaboration Using Only Joint Position Sensor

Round 1

Reviewer 1 Report

This manuscript developed a NARXNN training process for collision detection, which only use the joint position sensor. The NARXNN method simplified the training process and was verified to have qualified MSE and training error.

Some suggestions:

1.    The literature citation of NARXNN [33], [34] correspond to the wrong reference, which should be [35], [36], please check all of the references.

2.    In page 6, there is something wrong in the second paragraph which cause difficult to understanding the NARXNN methods, what is the input?

3.    The NARXNN design was suggested to give more details, especially about the NARXNN structure. The details of the hidden layer and the output layer are not clear.

4.    The way to obtained the training error was clearly based on equation(2), but how to obtain the smallest MSE is not clear.

Author Response

Kindly please see the attached file.

Author Response File: Author Response.docx

Reviewer 2 Report

Especially for the abstract and introduction English proofreading and a revision are necessary. 

In detail I dont find the article well suited for Logistics and it also lacks novelty. In fact it bears large similarities with previous publications by the authors while only providing marginal improvements. The amount of effectivity lacks a suitable n number and eventually being able to detect a collision after it has happened is also limited in relevance for safety reasons which was stated in the introduction. Increasing n numbers for testing,  a more detailed analysis of residuals and bias etc with a more clearer separation of test set and training set might be the minimum to increase the quality of the article for publication.

A few of the issues I noticed in the introduction and abstract:

Abstract, first sentence: "AbstractIn"

Abstract, first sentence: "nonlinear autoregressive ..." There seems to be something missing. "nonlinear autoregressive model" ?

Abstract, last sentence: "In final, " --> "eventually, " ??

Introduction, first sentence: Grammar. Did you mean "During collaboration between the human nd the manipulator, safety is the most relvant issue that needs to be considered in procedural design." ?

Introdcution, third sentence: "msut be" --> "must " ?

Introduction, first and second paragraph overall: Grammar and style need a revision.

Author Response

Kindly please see the attached document.

Author Response File: Author Response.pdf

Reviewer 3 Report

In the context of the human-manipulator collaboration background, this paper presents a NARXNN to detect the collisions between the human and the manipulator. Moreover, the proposed NARXNN is designed depending only on the signals of the position sensors, which enables the model to be adopted in extensive application scenarios.

This research topic is interesting and could offer some insights for both researchers and practitioners. Besides, the introduction of the research problem and the formulation of the model seems to be clearly and rigorously presented. However, there are several concerns with this paper, which have been detailed below.

1. The introduction seems to provide sufficient background and include some relevant references to form a solid foundation for the study. However, the presentation is somehow too wordy, with several instances of redundant phrasing. Some sentences are unclear or hard to follow. For example, “Collision detection methods are classified into two parts; the first part is the model-based methods which depend on the robot’s dynamic model, and the second part is the data-based methods which depend on data.”I suggest that the authors cut extra words, tighten up awkward phrasings and make this part more concise.

2. The presentation of some figures and tables is casual and definitely not qualified for publication.

For example, Table 1 has serious format issues which would confuse readership and make the conveyed information unclear. Furthermore, the figures should be high quality for publication. For more information, please refer to Instructions for Authors (https://www.mdpi.com/journal/logistics/instructions).

3. The authors state that the explicit dynamic robot model is not needed in the proposed method (page 3, paragraph 4). This setting is somehow confusing, and the underlying explanations should be detailed.

4.The major contribution of this paper is the proposed NARXNN method and the authors declare that the obtained results reveal that the trained NARXNN is an efficient method in estimating and detection the collisions. The authors should include more evidence leading to that conclusion. In addition, the comparison between the proposed method and other models should be more specific.

Author Response

Kindly please, see the attached file. 

Author Response File: Author Response.docx